This one is from Season 2, Episode 5, with Alex Millar, co-founder and CTO of GovAI, and previously the co-founder of Bonfire, a procurement platform built for government. First released almost exactly a year ago.

Let’s look into it! 👓

When first spoke with Alex, people were already arguing about whether AI products were “real” products or just wrappers around foundation models. That question still floats around a lot today, surprisingly enough, and I appreciated that Alex didn't just stop there.

He challenged: is that really the interesting question to ask?

🎙 From the Archives: Alex Millar on How Everything Is A Wrapper

“Well, technically everything’s a wrapper.”

The question was simple: how does a product stand on its own if it's just a ChatGPT wrapper? An app that changes the way you interact with ChatGPT… but still, fundamentally, is just ChatGPT under the hood.

Alex's answer made me laugh out loud (and not figuratively). 😂

Yes, a product may wrap a model. But, if you think about it, every SaaS app out there is also just a web interface that’s a wrapper to a database somewhere. The real interesting question is whether the app enables more value than the raw underlying capability would on its own.

For a product to do that, it means doing the work to understand the niche well enough to know what the buyer needs around the AI itself.

In Alex’s case, that included things like privacy controls, acceptable use policies, PII detection and redaction, SSO, staff training, peer groups, and pricing that works for a city budget instead of a generic per-seat SaaS motion.

Today, a lot of AI discussion still stays at the level of capability. You know, can the model do this? Can it retrieve that? Can it call tools? Can it act agentically?

But, in reality, buyers care more about purchasing something they can roll out, govern, budget for, explain internally, and trust enough to keep using… rather than the raw list of features and capabilities.

💡 One Core Insight: The Product Is the Fit

One of the parts that stood out the most to me was Alex’s point that positioning and offering are part of the product.

Those are often seen as marketing terms, but really, they're more fundamental.

He gave a very practical example with pricing. Instead of charging per user, GovAI was working from a flat-fee model based roughly on city size and employee count. He learned that a city needs fixed costs. A city also doesn't want to decide which staff get access and which staff don't. That's just an overhead nightmare.

The pricing was part of product judgment that became part of their unique value proposition.

Same with the way he described training and peer groups. And, same with the way he talked about serving a niche tightly enough that you can identify common workflows and put useful rails around them.

Product strength often shows up in the surrounding decisions, not just in the model choice or the interface.

If you are building in AI right now, that still feels like a very useful filter:

Are you just shipping a capability? Or, are you shaping something people in a specific context can adopt?

🎧 Listen to the full episode now!

🔑 One Key Clip: A Great Product Still Has to Make a Noise

“But at the end of the day, if you have a great product and no one knows about it, does the product even exist?”

In the bonus episode, Alex said something I think a lot of technical founders need to hear.

Yes, you have to build something that genuinely resonates with people and delivers value. But, that's only half the battle.

You also have to figure out how to tell people about it and get it into their hands.

A lot of technical people are comfortable with that first half and uncomfortable with the second. They want the product to speak for itself. They hope a good build will naturally get discovered. Sadly, that's not how it works.

That felt especially relevant in the context of AI, where new products appear constantly and where the market is crowded with demos, launches, experiments, and noise.

A useful product, no matter how innovative, still needs distribution, timing, and sales and marketing that fit the stage of the company… just like every other business.

👂 Listen to the bonus episode now!

🕔 One Year Later

The part that still stands out to me now is how grounded Alex’s framework was.

He wasn’t talking as though the hard part was having access to the model or the tech. He made it clear the hard part was understanding the buyer, shaping the offer, handling the edge cases, and building something people could actually adopt.

And after all this time, and all the other conversations we've had with other founders and builders… that still feels… right.

It seems that a lot of strong AI product thinking looks pretty ordinary from the outside. It looks like pricing choices, rollout choices, training choices, workflow choices, and careful decisions about what to support now versus later.

That may not be the flashiest part of AI… but it may just be the most valuable.

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

This is TL;DL where I write about what stood out to me in each episode, share some food for thought, and share what’s next… for those of us who prefer to read.

🤗 Today, we’re spending time with one of our favorite people!

I really like getting to hear from people who’re willing to say what’s actually hard. Pat Belliveau is one of those people. He’s candid, practical, and really generous about sharing what he’s learned.

We really appreciate that about him!

We like him so much, this is the second time he’s been on the show! The first time we interviewed him was almost exactly one year ago. And it’s been fun to watch him and his company grow this whole time.

🎙 Just Interviewed: Pat Belliveau on Why Getting AI to Work Once Is the Easy Part

“Anybody can get an AI to do something once. That’s not hard. But to get it to do it twice, three times, four times, and repeatable and reliable...”

Pat Belliveau is the managing partner at Gambit Co, where he and his team build applied AI systems for real business workflows. Throughout this interview, Pat repeatedly talked about reliability, trust, and the discipline required to make AI useful beyond the prototype stage.

He’s also one of those guests who’s willing to say what many people avoid saying out loud:

The hard part isn’t successfully getting a flashy result with AI. That’s actually super easy, especially nowadays. The real hard part is making that flashy thing happen more than once.

You know, beyond the demo.

Listen to Daniel and Pat get into what changed for Gambit between year one and year two, why retainer relationships unlocked better work, and how Pat thinks about solving one real problem at a time.

🎧 Listen to the full episode now!

💡 One Core Insight: AI Gets Hard in Production

AI gets much harder the moment it has to fit inside an actual workflow.

It’s one thing to show that a model can produce a result… Turns out, it’s a whole other thing entirely to make that result reliable, explainable, and useful at scale.

The real work starts when AI has to survive messy documents, repeated use, human review, edge cases, and the expectations of a business that can’t afford surprises.

A prototype can create excitement quickly. That’s why they’re helpful! It can help people imagine what might be possible, and people need all the help they can get with their imaginations. But, production is where all the hidden complexity shows up. Suddenly, the question becomes, can this stochastically-determined AI system do that really amazing thing again, with the same result, tomorrow and the day after tomorrow? Does the team understand what the AI is doing enough to be able to verify it? Can they trust it enough to build a real, business-critical process around it?

Pat never talked about all this in abstract terms. He’s currently building in the middle of it and he distinguishes between a promising prototype and something that can actually hold up in production. That gap between impressive and dependable is where the hard work lives.

🔑 One Key Clip: Start Small Enough to Earn Trust

“Find something small, simple that you can crush. Use it as a use case.”

In the bonus episode, Daniel asks Pat what he would do if he were starting an AI consultancy from scratch today. Pat’s answer? Do something small, useful, and real for someone already in your network.

Then, turn that win into a case study, document what changed, and ask who else could benefit from the same kind of solution. Maybe it’s a logistics workflow, or its document extraction (who doesn’t use documents?). Maybe it’s something that’s only sort of adjacent. The point is to find the layer of the problem that applies beyond one organization.

AI is a low-trust market (anyone else feel like there are a few too many “experts” with 40 years of experience with ChatGPT?), and in a low-trust market, you can’t begin with big promises. You have to begin by solving one visible problem well enough so that people can see the value for themselves. Then, you let that story open the next door.

The short clip is worth it for anyone thinking about AI and business.

👂 Listen to the bonus clip now!

🥡 Food For Thought: Where Is the Real Friction in Your Workflow?

AI can probably do whatever you’re wondering about right now. It might take some work to set up, but it’s usually possible to figure it out.

The more interesting question is, “Where is the friction, and what would change if this step became easier, faster, or more consistent?”

It’s so easy to get distracted by tools, features, and demos. But, what’s the actual work people are trying to get done? What’s the real bottleneck? What’s repetitive? What’s slowing the team down? What’s creating drag in the process?

Those questions change the starting point. The answer might not always be flashy, but if you start looking for the part of the workflow that’s most ready to improve, you can get an ROI much more reliably and much faster.

Sometimes that means cost savings, sometimes it means speed... sometimes it means just giving a team more room to focus on the work that actually requires a human (i.e. not data entry).

I’d love to know how you think about this in your own work: when you look at your team’s workflows, where do you see the most friction right now? What would change if that part suddenly became easier?

👇 Let’s chat in the comments!

Leave a comment

🥁 Up Next: March Break Student Special

Next, we’re doing a March Break special!

(Even if it is arriving just after March Break 😅.)

First, we’ll be revisiting Pat Belliveau’s bonus episode from Season 1 on AI in schools, and on the ways school systems may be falling short of what students actually need right now. Even though it was recorded a whole year ago, it’s still poignant now… and that’s kind of sad.

Then, next week, we’ll share a set of short interviews with high school students about their day-to-day experiences with AI, what they think about it, and how they see it shaping their future.

I’m especially excited about those episodes! We hear a lot about students when people talk about AI, but not nearly enough from them. Getting to hear their perspective directly feels important, timely, and honestly just really needed.

Follow the podcast so you don’t miss the special episodes, and subscribe to the newsletter if you want the summary in your inbox next week!

▶ Listen on your favorite player!

✨ The Future of AI Will Belong to What Holds Up

This podcast exists to stand at the gate of businesses using AI, separating hype from lasting impact. One of the clearest ways to tell the difference is reliability.

A flashy result can get attention, but a tool that keeps working inside a real workflow is something else entirely. That’s where trust is built and where teams start to believe AI might actually deserve a place in how they work.

If you know someone who is building AI carefully, thoughtfully, and with real operational stakes, we would love to hear from them!

As always, thanks for listening! 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

This one is from Season 2, Episode 4, with Bijan Vaez, CEO of Merchkit, three-time founder, and former CTO who has spent years building SaaS products through more than one technology shift.

🛍 Let’s go shopping!

This conversation aged really well and it was fun to revisit. You know, like good wine. 🍷

When first spoke with Bijan, the AI world was still in one of those especially dizzying phases where everything felt possible and very little felt settled. New capabilities were showing up constantly and everyone was trying to figure out where the real opportunities were.

No one could really tell.

That was the reason this podcast started at all: to cut through the noise and distinguish between hype and lasting impact.

So, let’s dig into it: what does actually last?

🎙 From the Archives: Bijan Vaez on Why We Shouldn’t Get Stuck On What AI “Can” Do

“When you’re just focused on the feature that AI enables, that could be wiped out very quickly.”

Bijan is the founder and CEO of Merchkit, where he works on AI-powered product catalog management and enrichment for retailers, marketplaces, and enterprise brands.

I loved how honest he was about the path that got him there. He didn’t start with a neat vertical AI thesis already in hand, he started where a lot of curious builders started: with the frontier itself.

AI inspired him to get back into being a technical person again, not just a builder of companies. That was cool.

He talked about getting pulled into GPT-3 early, experimenting with Stable Diffusion, and even trying to dig into the deeper research side of the field. Then he went to CVPR, asked researchers all his hardest questions, and came away with a very clarifying realization: this space was much bigger, and much less settled, than he’d assumed.

“And I was like, oh no — if they don’t know, I have no chance of figuring this out. I just realized how big and how nascent this whole space is.

That realization changed the direction of his work.

Instead of trying to move the frontier forward himself, he stepped back and asked how the new capabilities could be applied to a business problem he could understand deeply enough to solve.

💡 One Core Insight: The Hard Part Is Usually Upstream

Turns out, the problem was pretty unglamorous.

Most people start with, “What is the coolest thing AI can do?”

Bijan instead asked why customers kept getting disappointing results from AI tools they were already excited to use.

And the answer, again and again, was… dun dun dun…

Data.

Teams were dumping huge amounts of messy product information into ChatGPT and hoping it would somehow transform that into useful output. Instead, they got results that created more cleanup work. The model was, essentially, being asked to perform magic on bad inputs.

I suppose, that was the promise, right?

Bijan and his team realized the real work that needed to be done: helping retailers ingest, clean, contextualize, and enrich product data, so the downstream workflows became tractable. Once the data quality improved, a lot of the “AI magic” people wanted actually started to become possible.

This episode was a reminder that lot of AI conversations, even today, stay at the level of capability. Can the model do structured output? Can it reason? Can it summarize? Can it generate? The answer is usually “Yes, but…”

If the workflow is weak and the source data is unreliable, the capability itself, no matter how magical, won’t help you.

Anwar talked about how AI companies run into that trouble all the time in the very first episode of the podcast: real-world data is messy… and it turns out AI can be quite picky about its data.

🎧 Listen to the full episode here!

🔑 One Key Clip: Big Markets Give You Room to Learn

 “… if there’s a big enough market, you can keep tinkering, and there’re enough customers to throw a new idea at.

In the bonus episode, Bijan talked about something that sounds obvious once its been said: when a technology shift is this large, it’s actually hard to know where to start. There are too many opportunities, too many possible wedges, too many things that might work…

So, best to start in a very large market.

That way, if the first idea misses, there’s still enough room to keep testing, adjusting, and learning from the same kind of customer without immediately running out of runway… or actual customers.

He started in AI product photography. That was the first swing.

It led him into conversations with customers. Those conversations exposed mismatched expectations, thin budgets, and a much more painful operational problem hiding underneath the original use case. And that’s how he ended up much deeper in retail data infrastructure and enrichment.

That still feels like very good advice for founders building in AI now:

Bijan didn’t get overly attached to the first story he told himself about where the value was—he used his first product idea as a listening device. And then, through relentless focus on trying to discover what the jobs that needed to be done were, he found a bigger, juicier problem to solve.

👂 Listen to the bonus clip now!

🕔 One Year Later

The surface-level feature race feels like it’s only intensified—more and more capabilities are getting bundled into the major platforms. Tools that once looked differentiated get crowded out very quickly.

In that environment, Bijan’s warning is more relevant now than ever: if your whole business is a thin wrapper around a feature, you’re in a fragile position.

I’d argue, you technically always have been… but it’s just a lot more obvious now. The market is less forgiving.

But, if your business is built around a real workflow, real context, and real operational pain, you have something sturdier to work with.

That might not be as glamorous, and a lot of times, it makes the work much harder. You have to understand how people actually do their jobs and put in the work to understand the data they rely on, the edge cases they run into, the review steps they can’t skip, the trust thresholds they care about, and the parts of the process where human judgment is still doing indispensable work.

All that is way slower (and way less fun) than shipping a vibe-coded demo.

But, it’s also much closer to building something useful.

This conversation reminded me that “Applied AI” is often a much “humbler” discipline than the phrase makes it sound. Instead of flashy, frontier tech, it means going deep into a business context, solving old problems in old systems with newly available tools.

It means caring whether a team can trust the output enough to actually use it.

That’s still so relevant now, Pat Belliveau and Daniel talked about it in the most recent episode of the podcast!

(Totally organic connection, I promise. It just stood out to me.)

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

This is TL;DL where I write about what stood out to me in each episode, share some food for thought, and share what’s next… for those of us who prefer to read.

🤑 Let’s talk AI and money.

There’s something oddly reassuring about hearing a fintech founder say that lenders actually do want to lend money.

Most of the time, the process feels overwhelming and stressful. Sometimes, it feels a little unfair—like the system is built with overly restrictive rules and onerous requirements. But, what if all that is just because the information borrowers need is scattered and disorganized?

This episode with Sharmeen Aqeel sat at the intersection of two things I particularly care about: trust and design. And how the two are very intertwined. As a designer, I often think about (and evaluate) how many systems feel outright hostile when, really, they’re just poorly designed.

Yes, I’m a great person to have coffee with. ☕

🎙️ Just Interviewed: Sharmeen Aqeel on Using AI To Help Businesses Borrow Money

Sharmeen is the founder and CEO of Lyyvora, a Lending-as-a-Service platform helping healthcare and medical aesthetics clinics access capital more clearly and efficiently.

Her background is in product design, shaped in France and Canada. And she speaks like a designer, too. You can hear it in the way she talks about the lending gap in a user experienced-centered way: lenders are there, the information is there, but borrowers feel confused and stop trying after one rejection—even if, maybe, all they needed was a different document.

Her UX-focus allowed her to see lending as a workflow problem. Rather than think of it like a huge, insurmountable finance problem (which would have been my first instinct), she figured making the user experience easier would solve the problem she saw.

After all, lenders want qualified borrowers. So, what’s actually broken?

🎧 Listen to the full episode now!

💡 One Core Insight: Accessibility, Not Obstruction

What Sharmeen describes isn’t a system built to exclude borrowers, per se, but a system that’s just become difficult to navigate for the people who most need to use it. The criteria for lending are generally well defined. Lenders care about revenue levels, time in business, documentation, and other indicators that help them understand risk. None of that is unusual or surprising within financial services.

Where the difficulty arises is in how those requirements are communicated and experienced. Information about lending criteria is often scattered across websites, buried inside lengthy explanations, or delivered through formats that require time and interpretation to understand. For someone running a healthcare clinic, that level of research quickly becomes untenable. They’re managing staff, patients, and operations—spending hours comparing lenders and trying to decode eligibility requirements is a difficult thing to find time for.

The result is a pattern Sharmeen encountered repeatedly while chatting with people in the space. A clinic receives a rejection from its primary bank, assumes that decision reflects the entire lending market, and stops applying. In reality, other lenders with slightly different risk models might’ve been willing to consider the same borrower.

Lyyvora doesn’t try to rewrite the financial rules that lenders use to evaluate borrowers. Instead, it tries to improve the experience of borrowing through human-centered design. The platform consolidates the application process into a single intake flow, then uses AI to assess borrower readiness and compare that profile against the criteria of multiple lenders.

From the borrower’s perspective, this changes the experience from a series of opaque applications into something closer to guided navigation. Instead of guessing which lenders might say yes, they receive a clearer picture of where they stand and which options are realistically available.

Daniel’s conversation with Sharmeen highlighted how AI fits into this structure. The technology doesn’t replace lender judgment or remove risk evaluation. But, because it helps interpret criteria at scale and presents the results in a form that borrowers can actually use, it opens the doors for borrowers who previously wouldn’t have knocked.

For founders building in regulated or high-stakes industries, this is a useful reminder that many problems labeled as “difficult” are actually problems of accessibility. The underlying system may already work reasonably well. The missing piece is often the design layer that translates complexity into something understandable.

What AI makes possible is a different interface to that complexity.

Instead of asking borrowers to climb the entire mountain of information themselves, systems can now interpret those rules and present them as smaller, usable decisions along the way. What documents are missing? Which lenders might realistically consider this application? What step should come next?

When that translation layer works well, the experience changes dramatically. Borrowers no longer need insider knowledge of the financial system to navigate it (what small business owner signs up for that?)—just the ability to respond to the next clear step.

🔑 One Key Clip: What If ChatGPT Makes You Obsolete?

“One thing that I don’t see AI doing is the network that I’m creating.”

In the bonus clip, Daniel asks a forward-looking question. What happens if matching borrowers to lenders becomes trivial? What if, in five years, a single GPT can gather criteria and return offers instantly?

Sharmeen didn’t flinch. Yes, the core workflow may get easier. And yes, ChatGPT may one day start doing what Lyyvora does out of the box. But, the network she’s building right now? That will still matter.

She’s cultivating relationships with lenders, building a close community of borrowers, staying hands-on in early deals to protect credibility…

AI may automate the intake, but it can’t automate trust.

If you’re thinking about defensibility in an AI-heavy world, this short clip is precise and honest. Give it a listen.

👂 Listen to the bonus clip now!

🥡 Food For Thought: What Becomes Your Moat?

Consider this:

If AI makes the core work easier, what becomes the moat that sets you apart?

It’s easy to think about this question in the context of companies—if the technical capability that once defined your product suddenly becomes easier to build, easier to automate, or even embedded inside general AI tools, then the source of your defensibility has to shift somewhere else.

Maybe that “somewhere else” is distribution, or brand, or even a network of relationships that took years to build. Or, maybe, it’s the trust people place in the humans behind the system.

But, the same question applies to individuals, too.

Many of us built careers around skills that required time, training, and specialized knowledge. If AI begins to reduce the effort required to perform those tasks, the question naturally becomes: what makes someone stand out when many people can now perform similar tasks?

Is it experience interpreting messy situations? The relationships you have built? Your ability to see patterns and ask better questions? The way people trust you to guide decisions when the stakes are high?

I don’t think this question has a single answer, but I do think it’s becoming one of the most important questions for anyone in the workforce right now—business owner or not.

I would genuinely love to hear how you are thinking about this. What becomes the moat when the work itself becomes easier?

Share your thoughts in the comments or reply and tell us how you are approaching this in your own work!

Leave a comment

🥁 Up Next: Patrick Belliveau Returns

“What happens when an AI prototype works once, and then has to work every day on a million real inputs?”

Patrick Belliveau is coming back to the show just over a year after his first appearance, and the evolution is real.

When Daniel first spoke with him, Gambit was building chat-based AI personalities. Now the focus is agentic workflows with audit trails, fallbacks, and human-in-the-loop checks.

He also shares concrete patterns they use to keep systems honest, including multi-model validation when extracting from documents. And, you’ll hear about “Bogan”, an AI personality paired with a research layer that wrote a cold email and got a reply in 35 minutes.

If you’re curious about what it looks like when an AI company moves from prototype to operational discipline, you’ll want to hear this one. The episode drops tomorrow!

▶ Listen on your favorite player!

🔍 The Part AI Still Can’t Automate

AI can reduce the cost of analysis, accelerate communication, and help systems surface patterns more quickly than a human team could on its own.

But, at the same time, relationships, credibility, and community still develop through slower and more human channels.

Listening to Sharmeen describe the network she is building around Lyyvora made me think about how many successful systems ultimately rely on both at once: automation where speed helps and human presence where trust must accumulate over time.

At least for now, it seems trust isn’t something that can be automated.

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

This one is from Season 2, Episode 3, with Jennifer Moss, an international speaker and the author of Why Are We Here?.

🤩 Let’s get to it!

Some conversations were recorded in a different season of life, but we can still feel the weather in them today. When first spoke with Jennifer, AI still felt brand new to many people, and “AI Anxiety” was a new term going around.

Today, it feels simultaneously like things have and haven’t changed at all.

🎙 From the Archives: Jennifer Moss on The Post-Pandemic Multiverse of Work

“We’re in the multiverse of work. We’re not in the future of work. We’re in a whole different timeline.”

Jennifer Moss is an award-winning journalist, a Harvard Business Review columnist, and a globally recognized voice on workplace wellbeing and leadership. She’s written The Burnout Epidemic and Why Are We Here?, and she’s spent years doing the kind of work most of us don’t have the time or access to do: interviewing widely, reading deeply, and tracing patterns across what employees, executives, and researchers are all seeing at the same time.

I particularly liked this episode because she came at the conversation as a workplace culture person, not an “AI person”, who kept getting pulled into AI conversations because the human effects were showing up everywhere.

Like she just couldn’t escape it.

I suppose, no one could.

In the episode, she talked about how the pandemic changed the baseline for how people relate to work, why “why am I here?” became more than a passing thought for so many, and why leaders kept reaching for the wrong levers when they felt the pressure to move fast.

It was fun!

💡 One Core Insight: We’re All Still Living in The Aftershocks of the Pandemic

This interview was released almost exactly a year ago… and yet, I think everything that was said then still holds true. Especially the bit about how we’re in a “multiverse of work”.

(Still one of my favorite clips. 😊)

We didn’t return to the old timeline after 2020. A lot of people changed how they personally related to work, meaning, and time, and they did it in the middle of shared loss, shared grief, and the unsettling awareness that normal life can disappear suddenly and unceremoniously.

We all had to rethink life in the unsettling company of mortality salience.

As a society, we were forced to face the finitude of life, and when that happened, it created stress, reactivity, and polarization for some, and a deeper pull toward meaning and impact for others. In other words, the pandemic shaped what people tolerated, what they prioritized, and what started to feel pointless.

In the main episode, Jennifer connected this to why so many people made big career moves, why the “why am I here?” question became persistent, and why it’s naïve to treat the pandemic like a thing we can simply put in the rearview mirror.

Jennifer made the observation that many workplaces are still trying to solve today’s morale and engagement problems with pre-2020 instincts. They tighten the rules, increase visibility, reassert control, push speed and urgency as a substitute for clarity.

However, mortality salience changes the context people are bringing into work. When someone has been shaken awake to “life is short”, they don’t automatically snap back into old patterns just because a mandate says it’s time. Unconsciously or not, they start looking for purpose, they start measuring leaders differently, and they start asking harder questions about whether work is fuel or just grind.

AI landed in the middle of all that… as an accelerant. If leaders treat AI adoption as a race, while workers are still navigating a deepened sensitivity to meaning, safety, and direction, it creates the conditions for disengagement that looks like laziness from far away and feels like self-protection from the inside.

If you’ve been wondering about “AI Anxiety”, “FOBO”, and why engagement at work is at an all-time low and hopelessness is at an all-time high, the full episode explores all of that and includes some practical ideas on what to do.

🎧 Listen to the full episode here!

🔑 One Key Clip: How AI Turned Into a Race Before It Became a Plan

The bonus clip is a tight explanation of why so many AI rollouts feel frantic.

Jennifer shared a surprising data point: in 2019, technology wasn’t even in the top six disruptors CEOs were most worried about. Then, in 2020, it finally showed up on the list. Within two years, it shot up to number one.

That shift illuminated a specific kind of executive pressure:

“We can’t get behind.”

Jennifer described the push and pull she saw inside leadership teams. On one side, urgency and competitiveness. On the other side, a quiet admission that the fundamentals were missing:

“I don’t really have a strategy.”

“I don’t really have a plan.”

“I don’t really have a communication strategy.”

And employees felt that absence immediately—they felt the adoption pressure, but they didn’t hear a coherent story about what changed, what stayed, and how people would be supported through the transition.

Jennifer pointed out that the uncertainty became a subtle, persistent force. If people stopped seeing themselves in the future of an organization, hopefulness dropped, and employees didn’t see a point to being loyal or engaged.

She also named the missed opportunity: AI was supposed to remove the mundane tasks and open room for more meaningful work. But you can’t just automate someone’s tasks and call it progress if you don’t redesign the role and invest in upskilling.

Where was all the meaningful work that we were promised?

👂 Listen to the bonus short here!

🕔 One Year Later

It’s been a whole year, and yet, if you look at the world… things haven’t really changed.

Many teams are still living with the aftershocks Jennifer described. Mortality salience didn’t actually evaporate when restrictions ended—people still feel that low-grade awareness that life is finite, and there is still a tangible sense of existential angst.

I think that might be why Matt Schumer’s viral essay “Something Big Is Happening” hit everyone the way it did. It framed this current “AI moment” as “February 2020 again”, with AI as the incoming wave, and it tells people to prepare for job disruption fast or be hit unawares like most of the world in 2020.

It’s almost a little triggering.

But panic isn’t a strategy. If leadership can’t explain what AI is for, what changes in real jobs, and how people will be supported, workers interpret “move fast” as “you’re on your own”.

And more than anything else, that’s just a surefire to get AI Anxiety, Fear of Becoming Obsolete, and disengagement, even in high-performing teams.

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

This is TL;DL where I write about what stood out to me in each episode, share some food for thought, and do a roundup of what happened and what’s next for those of us who prefer to read.

Let’s read! 🤓

We’re having a bit of a false spring here in Canada. An unexpectedly warm day, the snow starts retreating, and the kids start asking for their bikes.

And you try explaining to a 3-year-old, “It’s warm right now, but it’s winter, so no, we’re not digging your bikes out of the garage.”

And it just isn’t fair.

That’s kind of how “fair” shows up in AI conversations right now. Everyone uses the word, almost nobody defines it, and it can feel like we’ve already settled the question.

Underneath the noise is a practical issue that keeps getting treated like a side note: who should get paid when AI uses human work… and how?

In this interview, helps put shape around it. She names the different kinds of AI use that matter, and she talks plainly about what paying creators could look like if it happens at scale.

🎙️ Just Interviewed: Julie Trelstad on How AI Is (and Isn’t) in Publishing

Julie Trelstad has spent 30+ years in book publishing, riding wave after wave of new technology, from desktop publishing and eBooks to print-on-demand and self-publishing.

Today, she’s Head of US Publishing at Amlet.ai, working on rights and licensing infrastructure for the AI era, and she runs Paperbacks & Pixels, where she helps authors build sustainable publishing and marketing systems.

Julie frames the current debate in a really practical way: questions about AI and human work are turning into an infrastructure problem. The conversation digs into what “AI rights” can mean in practice, from training to research to generative use, and why creators need a way to prove ownership, declare permissions, and get paid when their work is used.

If you’ve been trying to figure out what “fair” could actually look like beyond slogans, this episode is worth a full listen.

🎧 Listen to the full episode here!

💡 One Core Insight: “Fair” needs infrastructure

Julie offers a simple picture of what “fair” could look like when AI uses human work:

“Content providers should be paid on a token basis, like Spotify artists.”

That one sentence forces the conversation out of vague ethics and into operational reality.

and I have talked a lot about the impossibility of this. (You know, over coffee at breakfast, while the kids are running loose. What else do you do with your spouse?)

Once a model is trained, it’s not pulling a neat list of sources into the thing it generates. It’s just absorbed patterns across a huge corpus. Expecting a citation inside a generated artifact isn’t supported by the current LLM architectures. It’s more like… it listened to all of Disco, and then wrote a Disco song.

Which specific track do you footnote? Can you even do that?

Even in the best case, determining if a piece of content was necessary for generating something else would be extremely expensive. Like, training a bunch of LLMs for each citation expensive.

And, even if you did have infinite budget, you would only get an answer like "this source influenced this output," not "this source caused this output" or "this output quotes this source".

So, if we actually care about fairness, we have to talk about the moments where payment and permission can be made real.

Julie’s framing helps! She separates training from other kinds of use, like research and generation, and she argues that creators need a way to declare what’s allowed and get compensated based on how their work is actually used.

At the same time, the “good stuff” is moving behind paywalls, which makes high-quality data harder to access through the open web alone.

And we want our AI trained on high-quality data. As Julie says, if we’re thinking about a healthcare model, for example:

 “You’re gonna wanna build that on textbooks and peer reviewed journal content and not on the comments from WebMD. And if you don’t know where your content is coming from, it’s gonna be harder to say this is high quality and it’s gonna hallucinate less…”

The practical question becomes: if we want AI built on high-quality human work, how do we build systems where permission and payment happen at the right layer, without pretending the model is a search engine with footnotes?

🔑 One Key Clip: The “Good Stuff” Is Disappearing

“High value content is being put behind paywalls… the good stuff is disappearing from the web.”

This bonus clip is short, but it surfaces a real constraint that a lot of AI conversations skip over. We talk about “better models” as if the inputs are a given. But, Julie reminds us that if the best material gets locked down in response to scraping, then builders will have a pretty serious data access problem.

That’s where her argument for rights infrastructure comes in. If you want small, high quality models trained on textbooks, journals, and serious research, you need a way to know where content came from, whether you are allowed to use it, and how to license it… without months of back and forth emails and lawyers.

The clip also reframes the fairness debate. We know creators are owed compensation for their work… but it’s about more than that now. It’s also about whether high quality knowledge stays discoverable and usable at all, or whether it retreats into private silos.

Like in the olden days where knowledge and learning was gated by universities and only accessible to those who were able to go there.

If you build AI products, or you publish anything you care about protecting, this one is worth the few minutes.

👂 Listen to the short now!

🥡 One Takeaway: Separate “Training” from “Using”

I think a lot of the heat in AI content rights debates comes from treating every kind of use as the same thing.

Training is one thing. A model takes in a mountain of text then compresses it into weights. You don’t get a clean audit trail back out. That’s why “just cite the sources” sounds reasonable… but misses the mechanism.

Using is another thing. When an AI product searches, retrieves, summarizes, or quotes, it’s operating much closer to a traditional content workflow—you can cite. That is the layer where attribution, permission, and payment can actually be enforced. It’s also the layer where creators can plausibly be compensated based on real usage.

If we want creators to be treated fairly, the question is more than just, “should they be paid?”, it’s also where can fairness be enforced in a way that actually holds up in the real world?

💬 When you say “AI should be fair to creators,” what would you want that to mean in practice?

Leave a comment

🥁 Up Next: Sharmeen Aqeel on Trust, AI, and Founder Velocity

Sharmeen Aqeel is a product design leader turned founder, building Lyyvora, a lending marketplace for healthcare clinics. She is using AI everywhere it helps her move faster, from prototyping to outreach to borrower lender matching, and she’s equally clear about where she refuses to use it.

Up next, Daniel and Sharmeen talk about how AI can accelerate almost everything… but it can’t borrow trust on your behalf. They get into what she automates, what she keeps human, and how she thinks about building a network and a community that will outlast whatever the next wave of tools makes “easy”.

The podcast episode drops this Friday! Follow along in your favorite player so you get notified. 😍 Or check out the YouTube Channel and hit the 🔔.

✨ When “Fair” Moves from Principle to Practice

For years, the default assumption was that if the content is online, someone will scrape it. If creators wanted protection or compensation, they lock it down with paywalls.

AI companies took advantage of that and mostly got away with it. Yes, Anthropic got a $1.5B bill… but, if you look at it closely, it was for using pirated content, not for scraping content… so…

Now, high-value writing is moving behind paywalls, creators are looking for ways to prove ownership and set permissions, and builders are realizing that “just use the open web” isn’t actually a stable data strategy if you care about quality.

Julie’s point is that this whole game only works long-term if there’s a middle layer that can do the boring parts well: the ability to fingerprint content, declare which kinds of AI use are allowed, and make compensation frictionless enough that it happens in real workflows.

This makes me think of the conversation Daniel had with Dr. K last year on how the tyranny of convenience will always be something we’ll have to deal with as a species—tech companies are going to inevitably take the path of least resistance.

I wonder what it will look like.

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Today we’re rewinding to Season 2, Episode 2 with Atif Khan, who at the time was Chief AI Officer at Messagepoint, and who is now AI Architect at MindBridge, as well as a Data Growth Coach at Communitech.

Let’s get into it!

A year ago, “AI-first” meant something specific.

I mean, it still does.

I think most people just forgot.

Upon revisiting this, I realized how often we now use “AI-first” without defining it. Atif did a year ago, and in a way that still holds up.

🎙️ From the Archives: Atif Khan

“I organically grew in the industry as a software developer, computer scientist… and then went back to university… to understand not just the engineering part of things… but also what’s the science behind it.”

Atif lives at the seam between research and production. He cares about the theory, and he cares just as much about whether said theory survives contact with the real world.

One story he shared really stood out to me. A talented student joined his team and, within a week, was in tears.

“My classifier is not working.”

In class, the data had been clean. In the real world, it wasn’t. The theory was applied correctly, but the context she found herself in wasn’t what she expected… and she wasn’t trained to know how to compensate.

That gap between elegant theory and messy implementation is where most AI efforts stall. Atif’s message was poignant: understanding how something is built is only half the work. The other half is adapting it to reality.

💡 Core Insight: AI-First Is a Change in Expectations

In this episode, Atif defined “AI-first.”

If someone hands you a thousand numbers to add, you wouldn’t add them manually. You could, but that would be a very bad idea.

Instead, you’d use Excel because it compresses effort, removes friction, and would do it way more accurately and perfectly. “AI-first” is the next step. You show the machine the problem, and it returns not just the answer, but the reasoning behind it.

“Potentially it can look at those numbers. It can write a Python script in the background… give you the Python script so you can validate that this answer is correct, as well as give you an answer in real time.”

If your team had access to tools that compress cognitive effort, but your expectations haven’t changed, you are not AI-first. You are AI-enabled at best.

At Messagepoint, this became tangible. Copilot was rolled out to developers. They were not micromanaged in how to use it. Instead, the expected outcomes were adjusted.

“Less code reviews, more robust code… better coverage from unit testing… you can definitely measure each and every one of those on a daily basis with every commit.”

That is AI-first: a recalibration of what competent work looks like.

🎧 The full episode unpacks this mindset in more detail, and it’s worth revisiting.

🎧 Check out the full episode here!

🔑 Key Clip: Go Touch the Elephant

“There’s a story… people were taken to a dark room… and they had never seen an elephant before… everybody described the elephant differently.”

In the bonus episode, Atif uses this parable to describe AI. Depending on what you’ve touched, the beast looks different.

And then, he ups the stakes.

“This specific beast… it’s not just an elephant, it’s a dinosaur.”

Confusion, he argues, doesn’t go away by staying outside the room.

“We all have to go into the dark room and touch the elephant.”

And the barrier to entry is lower than ever.

“You don’t need to learn a programming language… you can literally just get on a browser, talk to it and get a feel for what it can do.”

What surprised Atif most over the last two years wasn’t that it could generate text, but that it could reflect.

“It’s going to be… at a point where it’s a cognitive companion.”

That phrase still stands out to me. AI not just as a text generator, but a cognitive companion. Something you can test ideas against, refine thinking with, and use to accelerate learning.

“I can potentially tap into the voice of many experts.”

And there’s no perfect time to start… or rather, any time is the perfect time to start.

So, the question is simple: have you stepped into the room yet?

👉 Check out the short

🧭 One Year Later

Revisiting this conversation, what stood out to me, again, was the mindset shift that seemed required to wrap our heads around AI properly and correctly.

AI-first was never about hiring researchers or announcing a roadmap… it was about recognizing that cognition had become cheaper and adjusting your workflows accordingly.

If you say your company is AI-first, what changed in how you measure work?

That question still feels uncomfortable in the best way.

A year later, the tooling has improved and the headlines have multiplied, but the core insight remains the same.

“AI-first” isn’t just a feature roadmap, it’s a necessary change in expectations.

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Let’s get into it!

On the surface, this was a conversation about AI, growth, and go-to-market. Underneath it, though, was something more familiar. has lived through multiple shifts that once felt just as destabilizing as this one: the early web, the rise of SaaS, the move from sales-led to product-led growth.

Each time, the tools changed.

Each time, the business models followed.

Each major shift in software eventually changed how companies acquired customers, where humans added value, and what kinds of roles actually scaled.

Now, AI is accelerating that process again.

🎙️ Just Interviewed: Dave Boyce on How AI Is Changing Go-To-Market Strategies & Business Models

“When I first started, we were hosting very large machine learning models in colo facilities, which eventually became known as the Cloud, inside a model that eventually became known as SaaS.”

Dave Boyce has been present at several moments that only get named clearly in hindsight.

He’s built and sold multiple SaaS companies across very different eras. One of them sold to Oracle. Another to Amazon. In between, he went from running businesses with average annual contracts of $1.2 million to building a company where the median ACV was closer to $1,200.

It was quite the whiplash.

Today, Dave works at Winning by Design, where he operates at a different altitude. He describes it as a “klick up” in perspective: instead of being heads-down inside one company, he now sees patterns across dozens of growth-stage businesses, many of them sitting comfortably at $50M to $2B in revenue… but also still worried about what comes next.

What makes his perspective especially useful right now is that he doesn’t talk about AI as a sudden rupture, but as another transition layered onto earlier ones. From sales-led to product-led. From high-touch to self-serve. From humans executing predictable work to humans focusing on judgment, coordination, and trust.

This episode pulls together those threads. Dave’s background gives him a long memory, and that memory shapes how he evaluates what is happening now.

If you want to hear Dave walk through this arc in his own words, from early SaaS to product-led growth to what AI unlocks next, the full interview is worth your time.

💡 One Core Insight: AI is the Continuation of the SaaS Revolution

One line from this conversation stayed with me while editing. Dave said, almost in passing:

“Everything’s changing again.”

Early in his career, Dave was building software before “the cloud” or “SaaS” were common language. Later, he had to reorient his thinking entirely when he moved from million-dollar enterprise contracts to a business where self-service was the only way the economics could work. Each shift forced a rethinking of how customers found products, how value was delivered, and where humans actually mattered.

Dave pointed out that over half of B2B purchases now begin with an LLM inquiry, and that most B2B buyers are millennials or Gen Z. The implication is structural, beyond behavioral: discovery, evaluation… even upgrading are increasingly happening without human intermediaries.

What struck me is how closely this mirrors earlier transitions. Each wave of software maturity pushed more work into the product itself. Each time, humans moved upstream, toward less execution, but more judgment and more responsibility for edge cases and trust.

Dave is careful not to overclaim. Growth curves are real, but we’re still early in the game. We do not yet know which AI-native companies will compound sustainably and which will hit familiar headwinds. Like he says, the laws of physics have not disappeared. What has changed is where friction lives.

🎙 Listen to the full episode now!

🔑 One Key Clip: What AI Can’t Replace

“AI is not gonna take our jobs as long as we are not trying to hold onto a job that AI should have.”

This clip stood out to me because it names the fear many people are carrying right now without dismissing it.

Dave started by acknowledging how reasonable that anxiety is. If your livelihood supports other people, it makes sense to feel unsettled when the ground starts shifting.

But, he did something important: he separated jobs from being human.

Order processing, routine execution, and predictable workflows are exactly the kinds of things AI will absorb. And I’d argue, are exactly the kinds of things AI should absorb.

Holding onto those tasks as the core of your identity is where the existential angst is coming from.

Simultaneously, Dave drew a clear line around the work that doesn’t transfer so easily: managing uncertainty, aligning stakeholders, building confidence, helping people navigate ambiguity... making connections between humans. These aren’t fallback skills—they’re the work that remains once the predictable parts are automated.

I really appreciated how practical this framing was, and how honest it was. It doesn’t deny displacement, but encourages us all to “Lean into the upper half realm of your job” and let AI take the rest. In that version of the future, automation is just a narrowing toward the work only humans can do.

If this question is already close to home for you or your team, the clip is worth a listen.

👂 Listen to the bonus short now!

🥡 One Takeaway: What Work Are You Protecting?

What this conversation surfaced for me is how often fear shows up around tasks, not purpose.

The problem, as it turns out, is that we keep defining our value around work that was already on its way out.

That raises an uncomfortable question, especially for leaders and operators:

Which parts of your role are you actually trying to protect?

Not your title. Not your team size.

The day-to-day work you spend the most energy defending.

Is it the reports you always write yourself? The approvals that always run through you? The processes that make you feel indispensable because they are slow, manual, or hard to explain? Those are often the first places AI fits, and the hardest places for people to release.

Dave recommended that we let go of the predictable work, and instead step into judgment, alignment, and responsibility for outcomes. It’s uncomfortable and new, but where growth can actually happen.

So here’s the question I’m sitting with after this episode:

💬 Which parts of your work feel hardest to let go of right now, even if you know they probably should be automated?

I’d love to hear how you’re thinking about that in the comments.

Leave a comment

🥁 Up Next: Julie Trelstad on AI and Book Publishing

Julie Trelstad has spent more than 30 years in book publishing, riding wave after wave of technology… from desktop publishing to eBooks to print-on-demand. Now she’s working on what comes next: how creators prove ownership, license “AI rights,” and stay discoverable in a world where copying and imitation can happen at internet speed.

Up next, and will be having a conversation about the infrastructure layer in publishing most people don’t get the chance to see: content fingerprinting, rights markets, and the emerging economics of training, research, and generative use. It also gets practical about the tension creators are living in right now: protecting work without locking it away.

✨ When Business Models Shift

One of the threads running through this episode is that business models don’t change simply because people get excited… they change because friction moves.

The web changed distribution.

SaaS changed delivery and pricing.

Product-led growth changed how customers enter the system.

AI is pressurizing the next set of assumptions, especially what gets handled by software, what gets handled by humans, and what work is worth paying for.

Personally, I’m pretty excited to see where things go!

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Today, we’re looking back exactly one year, to our Season 2 opener with Pat Belliveau, Managing Partner at GambitCo.

Let’s goooo! 💃

The conversation was recorded during a different AI moment, before “agents” became everyone’s favorite word, and while most teams were still trying to decide whether AI was a toy or a threat.

Pat was already living in the part of the curve we’re all catching up to now: building something real, putting it in people’s hands, learning fast, and staying close to the problem.

🎙️ From the Archives: Pat Belliveau

Pat is an entrepreneur and technologist whose AI journey started before the rest of us had language for it. He built AskEllyn an AI companion based on a breast cancer survivor’s memoir, so patients and families could access shared lived experience without being dropped into the chaos of unmoderated groups.

He did it with no coding background, learning by trial, YouTube, and ChatGPT. The result is now used in over 15 countries. That one project became the seed of GambitCo, and a repeatable way of building tools that people actually trust enough to use.

Pat is unusually grounded for an AI optimist and builder, and also refreshingly focused on building a real business.

“…if you can’t package something to sell to a client… you don’t have a business… because… the most dangerous thing you could do with AI is raise a bunch of money to go heads down, build something for a year…  it’s, like, the number one way to light money on fire.”

Unlike most AI tech founders, he talks about refusing VC money to stay focused on building what customers will pay for and calls out the game that most VC-funded tech startups and entrepreneurs play—they like to say it’s looking for the elusive Product-Market Fit… but at the end of the day, really, it’s looking for their next successful round.

The incentives are just totally misaligned.

💡 Core Insight: Build for “Dignity,” Not Just “Safety”

This interview had so many fun stories that I still refer to today. No joke.

One of my favorites was when Pat told of a story about a failure mode they couldn’t accept—if someone asked AskEllyn, “Can I take Tylenol?”, the system would sometimes answer:

“Yes, I took Tylenol on my journey.”

Which was true.

But, that “yes” was doing too much work. It sounded too much like medical advice, even if it’s meant as a personal anecdote.

And Gambit learned pretty fast that you can’t just write a giant list of things not to say—you just can’t and won’t be able to account for every single edge case that might come up.

You simply can’t put your arms around all the ways a person might ask something risky.

So they shifted the target from “What must it never say?” to “What does a dignified answer sound like?”

“…we got very good at… instead of worrying about the things that it shouldn’t say, worried about what is a dignified answer…”

Now that’s a weird engineering constraint.

“No medical advice” is still there, but “dignity” becomes the design standard for tone, framing, and how the system handles ambiguity.

Most AI teams treat safety as a checklist, then ship something that is technically compliant and emotionally clumsy. Users then bounce, or worse, they feel unseen. Pat’s insight is that for certain domains, trust is about how the tool behaves when the user is vulnerable—facts matter, but delivery is also important.

He gave a few really cool examples that show what “dignified answers” unlocked:

• Hospitals wanted to put AskEllyn QR codes in waiting rooms once they saw it was supporting the human side without stepping into medical advice.

• Husbands used it to learn how to support a spouse through chemo. No body thought husbands would be an ideal user, but there you go.

• A friend used AskEllyn to build a gift basket that actually matched what chemo can do to taste and appetite, and it created a moment where everyone felt understood.

If you’re building AI that interacts with people at an emotionally loaded moment, “helpful” is just not enough. The output needs to feel like it came from someone who understands the stakes.

Listen to the full interview if you want to hear how Patrick got there, and how his team thinks about product, business, and responsibility as one system.

🎧 Check out the full episode here!

🔑 Key Clip: Education Is Pointing in the Wrong Direction

The bonus conversation takes a sharp turn into education.

Pat zeroes in on AI detection tools being used in schools and universities and calls out the downstream harm of treating students as suspects.

“You are now putting kids on trial having used AI… and that thing is wrong.”

To make the case concrete, he describes running the same text through an AI checker, then asking a model to lightly rewrite it with small mistakes.

“100 percent plagiarized… then 0 percent likelihood of AI. Thanks.”

So, now we have to ask, what is the education system is actually teaching?

Instead of helping students learn how to use tools they will be expected to use at work, schools are penalizing them for touching the very technology employers already assume is table stakes.

“If you don’t use AI, employers are not going to hire you. That’s, like, a real thing when you graduate.”

Accusing students with tools that openly admit they are unreliable is reckless, especially when layered onto an existing mental health crisis.

“We can’t do this. It’s not good for them long term, and it’s definitely not good for them near term.”

He closes the thread with a deeper reframing: if intelligence is still being measured primarily as information retention, we have already lost the plot.

“If we measure intelligence on information retention… then we’re just dumber than an AI.”

The sad thing is, a whole year later… and as far as we can tell, nothing much has really changed. We are still hearing stories from students whose marks and futures are in jeopardy because their teachers are being forced to use AI plagiarism checkers that are unreliable.

It begs the question:

Are we educating for the world students are entering, or just protecting systems that no longer accurately describe reality?

👂 Check out the short here!

🧭 One Year Later

Pat ends this conversation with a line that feels even more relevant now:

“You have not missed the boat… you’re probably still early… AI is not very old and AI is not going anywhere.”

A year later, that reads less like encouragement and more like a challenge. If we are still early, then we still get to choose the habits we build into our products, our teams, and our definition of “good enough” in AI.

The question is, are we being intentional about it?

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Let’s go!

This might be dating myself, but I grew up using the internet by Googling things like “French Revolution wiki” or “large language model wiki” and clicking whatever showed up first, assuming it would be Wikipedia. That was my default.

Google was really just the easy way to get to a specific Wikipedia page.

(Anyone else?)

While editing this conversation, I realized I haven’t done that in a while. 🤔

🎙️ Just Interviewed: Bill Beutler on AI’s secret dependence on Wikipedia

At any given time… of all of the citations that appear in ChatGPT, between 5 and 15% of them go to Wikipedia.

Bill Beutler runs Beutler Ink, a digital agency focused on a problem most teams never realize they have: what happens when the public record about you lives on Wikipedia, and you’re not allowed to touch it.

If a company’s Wikipedia page is incomplete, inaccurate, or distorted by old news coverage, the people closest to the truth are also the ones least allowed to edit it. Wikipedia requires disclosure, distance, and patience. Changes must be requested, not made, and approved by third-party volunteers who are trying to protect the integrity of the project. Chances are, they don’t really care about your company.

Bill has worked inside those constraints since 2010, long before AI made Wikipedia newly visible. What changed, he says, is how often businesses now run into Wikipedia indirectly. As AI systems added search and citations to reduce hallucinations, Wikipedia became both a training source and a live reference point in answers. ChatGPT, in particular, relies on it heavily.

Now, many of us no longer visit Wikipedia, even as AI checks it constantly on our behalf. That makes the accuracy and balance of those pages more consequential than they’ve ever been.

💡 One Core Insight: AI Is Eating Its Own Sources

There’s a feedback loop that most of us are already part of, whether we intend to be or not.

Bill describes Wikipedia as a reference layer, not a primary source. It exists by summarizing and pointing to journalism, research, and reporting produced elsewhere. And, that model worked when people regularly clicked through search results, read articles, and landed on Wikipedia pages directly.

But, now, AI changes that behavior.

When answers show up instantly in a chat window, fewer people click through to the underlying sources. That means fewer page views for publishers. Fewer page views mean less revenue. Less revenue means fewer journalists. And fewer journalists mean fewer reliable sources for Wikipedia to cite.

Bill is careful not to claim we are already at the end of that road. But, the direction is clear enough to worry him, and it should probably worry anyone building with AI.

To me, it’s interesting how indirect this problem is. I mean, no one wakes up trying to undermine journalism or Wikipedia. (Well, most people don’t.) We’re just choosing convenience. But the systems we rely on for “accurate answers” are downstream of economic incentives that erode the very sources they rely on.

If Wikipedia loses sources, it cannot verify facts. If it cannot verify facts, it becomes less useful. And if it becomes less useful, the AI systems that depend on it lose one of their most stabilizing inputs.

That makes Wikipedia less like a website you can take or leave, and more like shared infrastructure whose health affects everything built on top of it.

🎧 Listen to the full episode now!

🔑 One Key Clip: Wikipedia’s first competitor… is an AI?

For most of its life, Wikipedia has occupied a strange position online. It has critics, but it has never had a real alternative. Bill’s been saying “Wikipedia has no competitors” for more than a decade, and until recently, that was simply true.

That changed this fall.

In the bonus clip, Bill talks through the launch of an AI-generated Wikipedia alternative backed by Elon Musk. It arrived all at once, with hundreds of thousands of articles, many of them longer and more coherent sounding than their Wikipedia counterparts. For the first time, there is something that looks like a parallel encyclopedia rather than a footnote experiment.

The first question we ask is, “Well, but is it better?”

… Does it matter?

If an AI system can generate encyclopedia-style entries faster and at greater scale, what becomes the role of a human-governed knowledge project? And if multiple AI companies need a shared body of knowledge, do they trust one owned by a competitor, or one that belongs to no one?

Optimistically, Wikipedia still functions as a kind of lingua franca. It is open, freely licensed, and not controlled by a single tech giant. That neutrality makes it usable by everyone. But, it is starting to look fragile.

👂 Listen to the bonus episode now!

🥡 One Takeaway: What Are We Letting AI Replace, Exactly?

When AI gives us an answer… what do we think it’s replacing?

Let me explain.

For most of us, it feels like it is replacing search. You know, fewer tabs, less scanning, less clicking around, less reading… That feels like progress. It’s certainly faster.

But, we can’t forget that AI answers aren’t created out of thin air (I mean, even hallucinations technically have a source). They’re assembled from systems that depend on journalism, reference works, and human-maintained knowledge projects like Wikipedia. If those systems weaken, the answers do too, even if that decay is slow and hard to see.

If we stop visiting sources, stop reading references, and stop supporting the places AI learns from, what responsibility do we have for what those systems become, or for whether they slowly fade away?

Leave a comment

🥁 Up Next: Dave Boyce on Going to Market with AI

Next week, Daniel sits down with , one of the people who has shaped how modern SaaS teams think about growth, sales, and renewal.

Dave has built and sold five SaaS companies, but his deeper influence comes from pattern recognition at scale. Through his work at Winning by Design, he studies how revenue systems behave across hundreds of companies, from global platforms like Canva to teams still searching for product-market fit.

If you care about AI, revenue, and durable growth, this is a conversation worth your attention!

✨ When Shared Knowledge Starts to Fray

This conversation with Bill really made me think about the systems I’ve taken for granted, and Wikipedia is one of those systems. It sits beneath search, beneath AI answers, and beneath how many of us come to understand the world. It’s almost invisible.

That makes it easy to forget that shared knowledge doesn’t maintain itself. It requires incentives, attention, and people who care enough to keep it accurate.

I think, if AI changes how we access information, it also changes who is responsible for keeping that information alive.

Do we really want an AI maintaining that body of knowledge for us?

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show 5 stars. Every little bit helps! ⭐

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

(It’s the dead of winter here in Canada. It’s cold. And snowy.)

Welcome! This is TL;DL where I write about what stood out to me in each episode, share some food for thought, and do a roundup of what happened and what’s next for those of us who prefer to read.

As we get ready to release the next season of Artificial Insights, I wanted to spend a bit of time to talk about the thesis of this new season.

This time, we’re not spending time on models, benchmarks, or demos. Those conversations matter… but they’re no longer the hardest questions to answer.

This season is about what happens after the tools start working well.

When systems can remember, act, and carry context across time, the center of gravity shifts. Less effort goes into recall and routine execution. More weight lands on judgment, coordination, and trust.

🎙️ This Season’s Focus

Across the next set of conversations, we are paying attention to how AI changes the shape of work, not just its speed. In particular, we’ll be investigating how AI is changing the landscape of economics and… maybe most importantly, how it’s changing us as people and as a society.

We will hear from builders, operators, researchers, and leaders who are wrestling with questions like:

• What does good judgment look like when context is automated?

• How does trust change when machines act on our behalf?

• Where does human value live when execution becomes cheap?

We’ll be looking at these questions and moving them out of the abstract, looking at how they show up in pricing decisions, org design, product scope, and how teams relate to the systems they depend on.

🎧 Listen to the season trailer now!

🎧 Start Here!

If you are new to Artificial Insights, or want a few entry points that capture the show’s core questions, these past bonus episodes are good places to begin. They’re shorts—so, shorter than full episodes 😉—and, usually, talk about a specific idea or topic we care about.

✨ Foundational Conversations

• How Is AI Anxiety Changing the World? with Jennifer Moss

• Why the AI Revolution Feels Different with Mike Kirkup

• What Happens to Jobs When AI Scales Output 35x? with Nicolas Tobis

🧐 Product & Founder Perspectives

• After 0→1: I Built a Product, Now What? with Alexander Millar

• Is AI a New Medium? with Ahmad Iqbal

• When Is the Right Time to Start a Company in AI? with Adeel Zaman

Each of these shorts touches a different angle of the same underlying question: how do we make wise decisions when the tools change faster than our institutions?

✨ What We’re Trying to Do Here

This podcast exists to stand at the gate of businesses using AI, separating hype from lasting impact.

As AI systems get better, they remove certain kinds of effort from the equation. While that can often feel like relief, it can also feel very destabilizing.

But, what’s important to remember is that what remains is not nothing—it’s actually judgment.

Knowing when to trust a system.
Knowing when to slow it down.
Knowing how to price work that looks effortless from the outside.

And, these are all human skills. They are harder to automate, and more important than ever.

👉 Have a guest in mind or a story worth sharing? Please reach out to Daniel at daniel@manary.haus!

As always, thanks for listening. 🙏

P.S. If you’ve found these conversations helpful, subscribing, sharing, or leaving a comment genuinely helps more people find the show. ⭐

(For, you know, the ones that really thought the whole season was too long, and so couldn’t listen. 🫣)

This season, I heard a lot of things twice. Sometimes thrice. Sometimes six, from different people who don’t know each other and are working in totally different industries.

By the time I was cutting together the Season Four wrap-up, I had that feeling you get when a pattern is obvious but no one has named it yet.

So that’s what this recap episode does. It doesn’t introduce new ideas. It traces the ones that kept showing up episode after episode.

(No, this isn’t a listicle.)

(I think.)

1. Everyone Is Trying to Fix the Context Problem

Early in the season, Aydin Mirzaee talked about building what he calls an “AI chief of staff” at Fellow.

No, not as a replacement for leadership or a productivity trick. But, as a system that actually remembers things like meetings, one-on-ones, and customer calls. All the context and information that inevitably leaks out of human memory over time.

Management usually runs on partial recall and anecdotes because that’s all we have. A system that can accumulate organizational memory changes how decisions get made.

🎧 Listen to Aydin's episode now!

That same idea showed up again later in the season, this time in healthcare.

David Proulx described how HoloMD supports psychiatrists between appointments with daily check-ins, medication adherence, mood tracking over time…

The important part wasn’t really the AI. It was the continuity.

Instead of asking patients to summarize weeks or months of lived experience in one appointment, and then trying to piece together a hypothetical story… doctors see the full picture.

Different domains, but the same problem: humans are bad at holding long-term context. AI is, on the other hand, really good at it.

🎧 Listen to David's episode now!

2. Time Keeps Collapsing

Then there was the time compression theme.

Josh Müller talked about a nonprofit translation workflow that used to take six months of volunteer audio editing. After forced alignment and automation, it took about sixteen minutes of compute time.

Six months to sixteen minutes is not just an optimization or efficiency gain… it’s an entire category change. Projects that were previously unrealistic suddenly weren’t anymore.

AI wasn’t just accelerating work. It was removing time as the main constraint on what teams could attempt.

🎧 Listen to Josh's episode now!

3. Software Is Starting to Act, Not Just Inform

Another thread that kept resurfacing was agency.

Alex Maier described what he calls AI-native systems. Products that don’t just show you information, but help you decide what to do next.

You don’t navigate a maze of features. You ask a question. The system responds with an answer you can act on.

🎧 Listen to Alex's episode now!

That idea moved into the physical world with Adeel Zaman. He talked about embodied AI in construction equipment. The system reasons out loud. Explains what it’s doing. Accepts feedback mid-task.

This is the part where my editor brain went, “Oh. That’s new.”

Humans are no longer the only bridge between software decisions and real-world action, and that has consequences. Some exciting consequences... and some that deserve caution.

🎧 Listen to Adeel's episode now!

4. The Barrier Between Imagination and Output Is Getting Smaller and Smaller

Ahmad Iqbal used the example of his young son trying to draw something he can clearly imagine but can’t yet physically produce.

AI fills the gap between intention and execution.

Suddenly, the conversation about AI turned to access. Who gets to create? Who gets to participate? Whose ideas make it past the first attempt?

🎧 Listen to Ahmad's episode now!

🤓 The Thing I Can’t Unsee

By the time I finished editing the wrap-up episode, the pattern felt obvious.

Across healthcare, nonprofits, robotics, and consumer software, AI was doing the same thing over and over again. Removing friction around time, memory, coordination, and execution.

When those barriers drop, more people can participate.

Which is pretty amazing, isn’t it?

More ideas survive. More dreams become realistic instead of hypothetical.

Season 4 convinced me that the most interesting AI work is about expanding what people and organizations can take on… not about replacing people.

Which, honestly, is a lot more interesting.

👂 Listen to the Season Wrap-up now!

If you want to revisit any of the conversations, the links above will take you back to the full episode posts.

And if you know someone building something that makes previously impossible work routine, please tell us. Those are my favorite edits to make.

See you in Season 5!

As always, thanks for listening! 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and/or give the show a review on your favorite podcast player. Every little bit goes a long way. 💖

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Allons-y!

Artificial support is better than no support at all.

It’s a thought that’s stayed with me since editing this episode. Especially given the context. At first blush, it sounds wrong.

Doesn’t it?

But, in mental health care especially, waiting is often the unspoken default.

Waiting weeks or months for an appointment. Waiting for symptoms to escalate enough to “count.” Waiting alone, without context, feedback, or reassurance that what you are experiencing even makes sense.

In the worst case, help only comes when it’s too late.

So, maybe, AI has a place here after all?

🎙️ Just Interviewed: David Proulx on AI that bridges gaps in mental health care

 It would be impossible to replace a psychiatrist with the models that we have now… Also, to put the health of someone into the hands of a machine… People are not ready for that… Society is not ready.

David Proulx has spent most of his career building at the edge of technology. He launched one of the first mobile e-commerce sites before the iPhone existed. He built and scaled a social network for mothers to more than 600,000 downloads across Canada and the U.S. And when that company collapsed during COVID, he locked himself in a room and taught himself AI from the ground up.

Today, he is Chief AI Officer at HoloMD, a psychiatric support platform designed to work alongside clinicians, not replace them. The system checks in with patients daily, remembers past conversations, and surfaces patterns that would otherwise be invisible in occasional appointments. But it is never positioned as a substitute for care. Every interaction is reviewed. Every escalation pathway is tested. And responsibility remains firmly with the psychiatrist.

🎧 Listen to the full episode now!

💡 One Core Insight: AI sheds light on what happens in the in-between

One of the constraints of mental health care is that it relies on snapshots.

A patient shows up every few weeks or months, and a clinician has to make sense of everything that happened in between based on memory, mood the day of the visit, and whatever feels safe to say in the moment. Important details inevitably get lost. Patterns flatten into anecdotes.

HoloMD checks in daily. It remembers past conversations, notices when mood drops for weeks at a time and can surface why, note that a pet died, that medication was skipped, and that sleep was bad. These are not dramatic revelations, but they are the texture of real life, and they rarely appear clearly in a single appointment.

Medication adherence is a good example. Patients often stop taking meds when they feel better, not realizing that the medication is the reason they feel better. Nightly check-ins make that pattern visible early, before things spiral.

This changes the role of the psychiatrist. Instead of relying on a single moment in time, they walk into an appointment with a longitudinal view of how someone has actually been living.

And this is only now possible because of AI.

🔑 One Key Clip: Why banning AI is like banning electricity

In the bonus clip, David compares restrictive AI policy to the moment when the Ottoman Empire refused to adopt the Gutenberg press. At the time, the decision was framed as protection, but the outcome was centuries of lost ground… all because others learned how to use it while they did not.

The truth is, AI is closer to infrastructure than to a single product.

David calls it the new electricity.

Something that reshapes everything it touches. Trying to ban it outright does not remove risk. It just guarantees that innovation happens somewhere else.

The clip is short, but it captures a tension many leaders are feeling right now. How do you slow down recklessness without freezing progress entirely?

👂 Listen to the bonus episode now!

🥡 One Takeaway: Hallucinations are not a bug. They are a design consequence.

We often talk about hallucinations as something models will eventually grow out of. As if more data, bigger context windows, or better tuning will make the problem disappear.

Daniel pointed out on LinkedIn that hallucinations are not just a temporary limitation. They are a consequence of how these systems relate to truth in the first place.

Language models memorize words and patterns. They return those words probabilistically. Even when trained on truthful material, they are not verifying statements against reality. They are producing language that sounds right.

What makes HoloMD interesting is that it does not ask the AI to discover truth. It defines truth explicitly. The system operates inside a constrained world where reality is what the patient has actually said, session after session. Statements can be checked against that record. Scope is limited. Context is controlled.

This suggests that progress here doesn’t necessarily come from smarter models, but from building systems that acknowledge what AI is and is not doing.

👉 Join the conversation now!

🥁 Up Next: A Season Wrap-Up

This conversation with David closes out the season.

Over the past stretch of episodes, we’ve talked with builders, researchers, and operators working in very different domains, but wrestling with many of the same questions. Where does AI actually help? Where does it introduce new risks? And what does it look like to use these tools without giving up responsibility or judgment?

Instead of jumping straight into the next interview cycle, we’re going to pause and take stock.

The next newsletter will be a special season wrap-up where I’ll attempt to shed light on what appears to be an emerging pattern. Stay tuned!

✨ Why restraint keeps coming up

As we close this season, I keep coming back to a pattern that shows up in almost every conversation.

The most thoughtful builders are not asking how far AI can go. They are asking where it should stop, slow down, or stay deliberately boring.

David’s work is a good example of that because it is specific about what is at stake and who carries responsibility.

As always, thanks for listening! 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and/or give the show a review on your favorite podcast player. Every little bit goes a long way. 💖

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Merry Christmas!

AI is really exciting.

And, also, AI can be pretty dangerous.

In this conversation, Alex Maier put words to that balance in a way I liked: he talked about putting your “kid eyes” on again. You know. That posture of curiosity where you start asking “what if?” and then actually, fearlessly, mean it enough to do something about it.

Then, he immediately paired it with the reality that you still need your “parent eyes” on too, because the systems that feel playful and empowering can also be leaky, persuasive, and hard to audit.

This episode is full of product thinking, grounded in lived constraints. After all, outdoor recreation is one of those domains where a wrong answer could be more than just annoying… it could be fatal.

(The outdoors is a dangerous place.)

🎙️ Just Interviewed: Alex Maier on AI meeting people where they’re at

I didn’t have to sort, didn’t have to filter. You met me where I was at. And, that was really the thesis behind this thing.

Alex Maier comes to AI product work from a place that is surprisingly physical.

Before leading marketing and product at onWater, he worked in textiles, shipped prototypes overseas, helped scale Nike’s swimwear category, and brought human-powered outdoor apps to market. His career has consistently lived at the intersection of bodies, gear, and environments that don’t forgive bad assumptions.

Now, he’s at onWater and it is not a novelty app. It pulls together messy, high-stakes data from weather systems, government agencies, conservation partners, and user communities to help people decide where, when, and how to get on the water safely… Fishing conditions, river flows, blue-green algae blooms, permits… regulations.

(I don’t spend very much time outside. This is all Greek to me.)

When Alex talks about “meeting users where they’re at,” he is not talking about channels or funnels. He is talking about mental state, experience level, and intent in the moment someone asks a question like, “Should I go paddle this river today?”

The goal is to simultaneously surface information faster and remove friction—a system that knows when the right answer is “No, not today!” and is opinionated enough to say so.

🎧 Listen to the full episode now!

💡 One Core Insight: What if users don’t need your product anymore?

Alex described the question that excited his team the most.

What if users didn’t actually need the app? 🤔

No, not because the product failed… but because AI changes the starting point entirely. Instead of opening an app, learning its structure, filtering, sorting, and triangulating information, users can now begin by asking a question in plain language. Where should I go fishing this weekend? Are conditions safe on this river today? What do I need to know before I head out?

Alex framed this as a shift from AI-assisted to AI-native thinking. AI-assisted helps you search faster. AI-native reshapes the experience so discovery starts with intent, not interfaces.

Most products are designed around the (sometimes unintentional) assumption that users will adapt to them. Alex flips that around. If your system already has the data, the maps, the regulations, and the context, why make people hunt for it?

The harder part is what comes next. Once the system gives an answer, it has to help make that answer actionable. Directions, gear considerations, safety warnings, and sometimes a clear “don’t go.”

🔑 One Key Clip: Why “personalization” no longer covers it

In the bonus clip, Alex goes after a word we use constantly and rarely define carefully: personalization.

He argues that what’s emerging now does not fit the old meaning. This is not about segmenting users or tailoring messages based on past behavior. It is about understanding someone’s intent, experience level, and confidence in real time, simply from how they ask a question and what they do next.

He talks about working trade shows early in his career, watching how people approached a booth. Did they linger? Did they avoid eye contact? Did they come in hot with questions, or circle back three times before speaking? All of that information shaped how a sales person ought to respond.

What AI does well is formalize that pattern. And, it does it so well because it is formulaic. It reads language, tone, and follow-up actions, then adjusts how it responds. That is why people experience it as “meeting them where they’re at.”

I think that raises the bar. If your system can’t adapt to who someone is in the moment, calling it personalized misses the point.

👂 Listen to the bonus episode now!

🥡 One Takeaway: AI-native reduces uncertainty

Daniel shared a line on LinkedIn that helped crystallize this episode for me: being AI-native is not just about speed. It is about reducing uncertainty.

What I appreciated about Alex’s framing is how grounded it is. He is not describing AI as a magic layer that makes everything faster. He is describing systems that automatically do what good humans have always done well.

They read the room. They notice hesitation. They adjust based on tone, language, and follow-up behavior.

They do all that many, many, many, many times.

Automation scales scripts, while AI-native systems scale room-reading.

But, this begs the question! If AI is mediating context and reducing uncertainty, then it is also making judgments that used to belong only to people. Some of those judgments are safe to automate. Some probably are not.

How can we tell which are which?

💬 Join the conversation now!

🥁 Up Next: David Proulx on AI that never hallucinates

The next conversation takes the themes of this episode into a very different, much higher-stakes domain.

David Proulx is the Chief AI Officer at HoloMD, where he helped build an AI agent that has handled more than 100,000 mental health conversations without a single hallucination.

That’s right.

No hallucinations.

If questions of trust, context, and responsibility plague you at night, you will want to listen to this one when it drops.

✨ Wonder, With Guardrails

This episode reminded me that we live in a very exciting time.

AI has cracked open a space where imagination feels permitted again. Where asking better questions matters more than memorizing interfaces. Where creativity shows up not just in art, but in how products meet people in the middle of real life.

We should all have our kid eyes on!

As always, thanks for listening! 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and/or give the show a review on your favorite podcast player. Every little bit goes a long way. 🚀

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Surf’s up!

(No this episode has nothing to do with surfing. I’m just feeling nautical today. Maybe because our guest is based in Australia?)

Lately I’ve been noticing how often “AI” gets blamed for things. You know: a confusing email, a weird slide deck, a design that feels off but no one wants to own.

Somewhere along the line, “the AI did it” has become a convenient shrug like it’s acceptable or something.

(It isn’t.)

This interview went in the opposite direction.

Instead of treating AI as something to hide behind, Ahmad talked about what it makes possible when it’s used with intention. After all, at its best, AI should actually help someone express what they were trying to say in the first place.

It should enable creativity. Not inhibit it.

🎙️ Just Interviewed: Ahmad Iqbal on how AI can make things delightful

I was actually approaching it like, this could be really delightful, where they just click this Magic Write button.

Ahmad Iqbal was AI product lead at Canva, now head of MENAP. He was Canva’s first generative AI product manager and led the company’s earliest gen AI features before most teams had even decided what “AI strategy” meant.

I really loved how unceremonious the origin story was.

He didn’t start with a roadmap, or a productivity metric, or a mandate to ship AI features. He started with a feeling. What would it be like if a user clicked a button and felt… well, delighted?

That instinct makes more sense once you hear how Ahmad talks about Canva’s product philosophy. Accessibility not as a layer on top, but as the entire point. If a tool feels intimidating or confusing, it doesn’t matter how powerful it is, it’s out.

Throughout the conversation, Ahmad kept returning to that idea. AI, in his mind, earns its place when it lowers the barrier between what someone wants to say and what they’re able to create.

That perspective shaped how Canva approached generative AI from the beginning, and it shows up in how Ahmad thinks about everything from design tools to education.

🎧 Listen to the full episode now!

💡 One Core Insight: Accessibility is the real AI strategy

Canva has always been built around the idea that creative tools should invite people in, not screen them out. If a product requires specialized training before someone feels competent, most people will never get far enough to enjoy it. Simplicity is about removing unnecessary friction.

Ahmad described AI as the next step in that same arc.

 We see AI as decreasing the level of competency and capacity you need to design.

He talked about his young son trying to draw a Ninja Turtle. The desire is there, the imagination is there, and the story is there. What’s missing is the fine motor control and vocabulary.

(Hey, I relate!)

AI, in that framing, isn’t about replacing creativity by making things easier to do or make, it’s about helping someone get past the mechanical barriers that keep ideas stuck in their head.

Instead of asking how fast something can be produced, the better question becomes: who can now participate?

For product teams, this has real implications. If AI lowers the barrier to creation, then the work shifts from teaching users how to operate tools to helping them communicate intent. That changes onboarding, guardrails, and what success actually looks like.

🔑 One Key Clip: When the medium changes, behavior follows

In this bonus episode, Ahmad frames AI through the work of Marshall McLuhan, the Canadian scholar known for the line “the medium is the message”.

See, today, we’re still treating AI like a container for old formats.

Radio started as someone reading the newspaper. Television started as filmed radio. Only later did each medium find what it was actually good at. Music, call-ins, sitcoms, new genres… new habits.

(Could anyone really have predicted the present-day TV-at-the-center living room?)

Ahmad thinks AI is still in that early phase.

People are using AI to make [things] better or faster, but they’re still making the same thing[s].

The real shift will happen when we stop forcing AI to imitate existing workflows and start paying attention to what new behaviors it enables. Then, it’ll really be magic. 🧙‍♂️

👂 Listen to the bonus episode now!

🥡 One Takeaway: If AI is the medium, what happens to work?

On LinkedIn, Daniel pointed out that AI is different in one important way. Unlike television, AI is a two-way medium. We don’t just consume it, we talk to it. And most of what we talk about is work.

If AI becomes the medium through which we understand, decide, and delegate work, then work itself may become conversation-first. Less about producing artifacts, and more about judgment, coordination, and meaning.

Or, we could use the same medium to strip humans out of work entirely.

What do you think will happen?

👉 Join the conversation now!

🥁 Up Next: Alex Maier on using AI to get you on the water

Most AI products stop at “ask the bot a question”. onWater is built around something more practical: helping people act.

In the next episode, Daniel talks with Alex Maier, President of onWater, about what it takes to design AI for the physical world. onWater helps people decide whether to go out on the water, understand real conditions, and move from information to action without digging through charts, maps, and disconnected tools.

If you’re thinking about AI-native design, or building systems where trust and context matter more than clever prompts, this is a conversation worth spending time with.

✨ When AI Lowers the Barrier, Responsibility Goes Up

When AI lowers the barrier to creating something, more people get to participate. That’s a gift! But, it also means we need to be more thoughtful about what we are inviting people into, and what kind of outcomes we’re normalizing.

Doing AI right looks less dramatic than the headlines suggest.

I think it looks like care, personal taste, and a humble willingness to ask whether a tool actually helps someone express what they were trying to do.

As a designer myself, I think it’s awesome! Many of the most important AI decisions aren’t actually technical ones.

They’re design ones.

As always, thanks for listening! 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and/or give the show a review on your favorite podcast player! Every little bit goes a long way. 😁

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Let’s go!

This week, Daniel spoke with someone that I found truly inspiring and one particular story hit me in a way that was quite personal.

In this interview, Josh shared that the idea that unlocked his biggest automation project dawned on him while he was reading his Bible, praying, and enjoying the Lord.

Not during a late-night grind session.

Not during a flow state while working away at a problem.

But, during a moment that felt unhurried. Maybe during a Sabbath.

When was the last time you had a Sabbath?

🎙️ Just Interviewed: Josh Müller on how AI changes what’s possible

So we were able to feed that to our system and now, literally, that six months happens in about 16 minutes of compute time on my server.

Josh Müller is one of the directors at Waha, a tech nonprofit building a Bible study app and curriculum designed to make the Great Commission simple and actionable. Their work right now is pretty intense: they are at 43 languages and trying to reach 100 by the end of next year.

The part of Josh’s story that was so inspiring was how absolute the bottleneck was… it was a limitation that seemed impossible to overcome… until it wasn’t. 🤯

Back in 2019, shipping a full curriculum meant months of manual audio editing. Someone would open an editor, find where each verse started and ended, copy and paste segments, and stitch everything together. It was slow and tedious and it sat in the critical path for getting content into new languages.

Then, Josh had his “Aha!” moment (see above), prototyped an automated workflow, then eventually used forced alignment, feeding text and audio into a model to generate timestamps.

Six months became minutes. A process that used to require sustained volunteer effort became a repeatable compute job on a server.

Think of what that means! 37 new languages in a single year, rather than 20 years. That’s a whole other level of enablement and possibility.

🎧 Listen to the full episode now!

💡 One Core Insight: Margin creates breakthroughs

Josh described the moment the whole solution came together for him while he was using the YouVersion app.

He was listening to the Bible and watching the text highlight verse by verse in real time. Then the thought hit him: if YouVersion can identify where each verse starts and ends, then Waha could, too!

Crucially, it was a reminder about how people work.

This was something that Aydin Mirzaee also spoke about in a previous interview and I think it points to something a lot of teams forget when work (and life?) gets busy:

The breakthroughs you are hoping for, more often than not, need margin.

And for Christians, specifically, it’s a reminder that when we slow down, we are more able to hear God speak and more able to recognize what is actually being placed in front of us.

Yes, AI reduced six months of audio editing to roughly 16 minutes of compute time. But, the deeper win is what that time returns. Josh talked about it as giving people their lives back from work that didn’t need a human in the loop, so they can do the work that actually matters and accomplishes the org’s mission.

That’s a helpful lens for anyone exploring AI.

It pushes you to ask a better question than, “what can we automate?”

You start asking, “what kind of work do we want humans to have time for?”

🔑 One Key Clip: Data stewardship is not optional anymore

For a long time, Josh’s instinct was to keep everything off the cloud. Emails, files, messages. Anything that reduced how much Google or Meta knew about him felt like a win. AI has complicated that stance. The tools become far more powerful when they can interface with your data, but the risks become more real at the same time.

Josh frames data management as a form of stewardship. More than just technical hygiene—it’s also moral responsibility, especially for mission-driven organizations. If you get your data organized well and apply AI thoughtfully, the upside is enormous. If you do it carelessly, the consequences can be just as large.

For Christian non-profits, this stewardship is non-negotiable.

👂 Listen to the bonus episode now!

🥡 One Takeaway: Speed changes the rules

Here’s a really cool thought:

Speed changes what is possible and what is viable.

Josh’s example makes that concrete. Reducing a translation workflow from six months to sixteen minutes doesn’t just make the same work cheaper, it breaks the assumptions the organization was operating under.

Goals that once sounded unrealistic suddenly become operational.

A lot of AI conversations get stuck at “AI as a tool” level. How much faster can I write this email? How much better can I summarize this document? But, that’s just scratching the surface! And, most of the time, not even all that net positive.

(Anyone else AI-summarize an email they received that was obviously AI-written?)

What Josh’s work shows is something more foundational. When the entire job fits into the time it takes to make coffee, you can redesign the entire system around it.

That is where discernment becomes important. Speed forces decisions and it should push leaders to re-evaluate where human effort belongs, what constraints are real, and which ones dissolve.

💬 Join the conversation now!

🥁 Up Next: Ahmad Iqbal on who gets to create

The next episode looks at capacity and access through the lens of product design.

Daniel is speaking with Ahmad Iqbal, product lead and Head of MENAP at Canva, about what accessibility really means in an AI-enabled creative platform. Ahmad helped launch Canva’s first generative AI features and shares how simplicity acts as a hard product constraint, not a nice-to-have.

If you’re building products that shape who gets access to creative power, this episode is worth your time.

✨ What we give time to shapes what we see

AI is changing where attention, creativity, and responsibility hang out. When routine effort collapses into minutes, what remains is judgment and discernment. We need to decide what matters enough to act on.

And more than ever, it looks like it’s getting more and more important.

That’s why this podcast exists. We started it to stand at the gate of businesses using AI, separating hype from lasting impact. Josh’s story is a good reminder that the most meaningful uses of these tools aren’t usually flashy.

They’re shifts that return time, restore focus, and create space for work that actually requires people.

As always, thanks for listening! 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and or give the show a review on your favorite podcast player. Every little bit goes a long way. 😍

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Let’s dig into it! 🦺

This episode was a really fun one and it made me (briefly) wonder if we’re in a sci-fi world now.

Don’t you just love it when interviews stretch your imagination?

See, when we talk about AI progress, we usually stay in the realm of screens, text, and software. But, what does intelligence look like when it has to act in the physical world, not just describe it?

Did you know that it’s possible?

🎙️ Just Interviewed: Adeel Zaman on giving AI bodies

 You’d be surprised, if we walked into manufacturing facilities, into construction job sites, in any kind of physical space, and see how much of it is done by machine learning.

And it’s a very, very small percentage.

And I think, arguably, it’s probably one of the biggest unlocks we can do to improve our standard of living as a society.

Adeel Zaman entered the University of Waterloo at sixteen, earned national recognition in math and programming, and later co-founded DOZR, scaling it into a $40M+ construction marketplace across Canada. After years in deep learning research and startup building, he stepped back to start again. Something he admitted was difficult to do!

And this time, his focus wasn’t software tools or dashboards… it was excavators.

In the interview, Adeel described working on what he calls “embodied intelligence”. AI systems that can see, reason, and act in the physical world. He started with construction equipment like excavators and skid steers, machines that already sit at the foundation of real-world work.

What made this conversation compelling was how grounded Adeel’s thinking stayed (haha, see what I did there?). He spoke openly about cost, learning constraints, safety, and the long road between today’s models and anything resembling general intelligence.

He also talked about real-life, talking mechs. Seriously. Check it out.

🎧 Listen to the full episode now!

💡 One Core Insight: Embodied intelligence may be a prerequisite for AGI

Adeel questioned something that a lot of AI enthusiasts and optimists take for granted—can we actually reach something like a capable “white-collar worker” AI using the knowledge-only training we do today.

Like, actually.

Today’s models excel at question answering and short tasks. But long-horizon work, whether physical or digital, requires understanding context, constraints, and consequences over time.

Think about it: humans don’t learn about the world only through language. We learn by acting in it, by bumping into limits, by adjusting after mistakes, by potentially crying a little when it hurts.

Adeel suggested that grounding AI in physical interaction may not just help robotics, it might help intelligence itself generalize more reliably.

I admit, the thought made me think of The Terminator and may have gave me goosebumps. Just maybe.

He pointed to research showing that mixing very different kinds of data can improve performance across tasks. Poetry and code. Vision and language. In the same way, learning to move objects, navigate space, and respond to feedback could shape internal representations that pure text cannot.

Don’t get me wrong. The takeaway isn’t so much that humanoid robots are imminent. It’s more that embodiment might be the missing ingredient for the kind of reasoning we keep asking current AI models to do.

🔑 One Key Clip: Why “why now” matters more than vision

In the bonus clip, Adeel explained how he evaluates whether an idea is worth building now by sharing a story about Jeff Bezos in 1995.

Bezos saw that the Internet existed, and he saw that usage growth was exploding. And he acted on it before anyone knew where it was going.

He was able to see that usage growth made a new category of companies viable, even if much of the infrastructure was still immature. Books weren’t necessarily the end goal, but they were the right shape for that moment in time.

Check out the clip to be inspired!

👂 Listen to the bonus episode now!

🥡 One Takeaway: Intelligence needs a body, but safety needs intelligence

AI systems are powerful, but disembodied.

Robots are embodied… but, technically, unaware.

That gap is why industrial robots remain isolated and unsafe to work alongside.

Yes, the labor shortages Adeel and Daniel discussed weren’t abstract—they’re already visible across construction and skilled trades in North America, and more severe in places like South Korea.

We know robots are going to be helpful, probably inevitable, but are we willing to trust them around people? Should we be? And, what would it take?

🦜 Join the conversation now!

🥁 Up Next: Josh Müller on how AI can change what’s possible

In the next episode, Daniel is speaking with Josh Müller, director at the tech nonprofit Waha, about what it looks like to apply AI under real operational, ethical, and theological constraints.

If you’re working in nonprofits, missions, or any environment where people’s safety and security matter more than convenience, this episode offered a real look at how AI can be used carefully, practically, and well.

✨ When intelligence meets the real world

This interview made me realize that the next frontier of intelligence isn’t going to come about simply because we’ve fed more human knowledge and data into our AI models.

We’re going to need to place that intelligence in the physical world eventually.

I don’t know about you, but that leaves me feeling simultaneously excited and uncomfortable.

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, Daniel Manary sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and/or give the show a review on your favorite podcast player. Every little bit goes a long way. 😁

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!

Thanks for joining me!

Can’t believe we’re in Season 4! It’s been a ride.

One of the things I love about AI is how it raises the floor.

This podcast and newsletter is an example of a project that never would have been possible without AI—we just wouldn’t have had the capacity or the ability to produce, edit, and create everything we do to make it happen.

We love doing what we do, but there really is only 24 hours in a day, and only two of us working on everything right now. And, we have four little ones.

AI has allowed us to do what would, quite literally, been impossible.

🎙️ Just Interviewed: Aydin Mirzaee on how AI can make everyone seem organized

Everything could be aided by AI and all of a sudden you could get the results without putting in the effort, which sounds like too good to be true, but like we’ve seen what AI can do, right?

AI makes a higher baseline of output feel normal, even for people who don’t naturally work in a disciplined, highly structured way.

In meetings, that’s easy to see. The organized 10-15% used to be the ones who wrote the agenda, captured decisions, and followed up. Now a system can do a lot of that. The result is a different minimum standard for clarity and follow-through.

And that’s pretty neat.

Aydin Mirzaee is the CEO and co-founder of Fellow, an AI meeting assistant designed to support teams before, during, and after meetings.

He has been building Fellow since 2019… and that matters because he lived through the pre-AI limitation of meeting tools—they actually work best only for people who are already organized.

The earliest version of Fellow worked extremely well, but only the most structured slice of a company used it. Everyone else wanted the outcomes of organization without the friction of doing the work.

AI changed the math. It made it possible to raise that baseline across the company, instead of relying on a few people dragging, sometimes reluctantly, everyone forward.

🎧 Listen to the full episode now!

💡 One Core Insight: AI raises the floor, then changes what “good” looks like

Aydin gave us a recruiting example that, I think, really embodies that idea of “raising the floor”.

He described a recruiter who meets with candidates, decides someone is a fit, then uses Fellow’s “Ask Fellow” feature to generate a high-quality write-up to send to the customer.

He framed it this way: if you were placing the CEO of PepsiCo, you would write an exceptionally thoughtful and detailed email. For a smaller role, you wouldn’t be able to spend the same amount of time. With AI, you can write an exceptionally thoughtful and detailed email for a smaller role, too.

The floor is rising! 🌊

Not that AI makes people brilliant by default… But, when baseline work gets easier and better, teams and individuals can redirect effort toward judgment, prioritization, and accountability. You know, the important human stuff.

🔑 One Key Clip: Why most teams never get past “basic” AI

In the bonus episode, Aydin named what he sees as the real bottleneck to AI adoption: workflow change.

At the end of the day, success with AI isn’t about model quality or even access to the tech… it’s actually just change management.

He explained that most people can get surface-level value from AI tools quickly, but the deeper value requires rethinking how work gets done, and that’s hard to do under deadline pressure.

Does the following sound familiar?

You could finish the task the old way by Friday… or you could experiment with AI, spend more time upfront, and risk frustration or failure.

Most people usually say, “I’ll try it when I have more time.”

Organizations that want real gains have to intentionally create space without deadlines. That’s the only way new workflows emerge and teams (and people!) can grow.

The clip is short, but it explains why many AI tools stall at novelty. Raising the floor requires patience (sometimes lots of it) before it pays off.

👂 Listen to the bonus episode now!

🥡 One Takeaway: When AI removes the blank page

Daniel shared a thought after the interview that helped me name what felt different about this conversation.

Before Fellow became AI-first, running good meetings required a series of small creative decisions. What should the agenda look like? What mattered most in that call? What do we need to follow up on next time?

That kind of work rewards people who are comfortable with a blank page… while excluding the ones who may not be so comfortable.

What AI changes, as Daniel put it, is the type of decision being asked of us. Instead of inventing from scratch, we become editors. Instead of asking “What should this be?”, we’re asked “Is this right?”

I think it’s interesting.

Editing is cognitively easier than creating, after all. It lowers the barrier to participation. You can reject the agenda more easily than create one. You can tweak the follow-up more quickly than write one. You’re still making the decisions… just not starting from a blank page.

That feels like another version of the floor rising.

🦜 Join the conversation on LinkedIn!

🥁 Up Next: Adeel Zaman on when AI leaves the screen

The next conversation pushes AI beyond meetings, documents, and software interfaces.

Daniel will be sitting down with Adeel Zaman, CTO and Co-Founder of DOZR, to explore what happens when AI begins to reason and act in the physical world. Adeel has moved from deep learning research to scaling startups, and now focuses on what he calls embodied intelligence. Real sci-fi stuff.

✨ Raising the floor without losing the human

I think when AI raises the baseline quality of everyday work, it changes who can participate, how decisions get made, and where human effort is best spent.

AI removes unnecessary friction.

That means fewer blank pages, fewer dropped threads, and more generally organized people. In the best case, this also means more space and room for more interesting human work.

And I think that’s pretty exciting.

As always, thanks for listening. 🙏

P.S. Artificial Insights is a podcast on how AI is changing work, life—and us. Every other Friday, sits down with leaders, thinkers, and builders in AI to have candid conversations on what they’re doing right now and how they think the world will change. If you’re a podcast listener, we’d love for you to check us out!

P.P.S. If you liked the episode, please subscribe, share, and/or give the show a review on your favorite podcast player! Every little bit goes a long way. 😁

Artificial Insights is a part of the Manary.haus family ❤️ Come say hi!