An honest retrospective on 20 years of work — through the lens of AI
Warning: this is yet another post about AI. But I hope it comes from a slightly different angle, and is thus worth your time. If not, let me know and I'll try to make it up to you.
I recently attended a discussion between Hannah Fry and Amol Rajan on AI and the future of learning. It was fascinating and candid — two people who foresee certain changes but are humble enough to know they can't predict exactly where this is all going.
That conversation made me realize something uncomfortable: I haven't done enough thinking about the role AI should play in my own work and career. I'm fearful that I'm not adapting quickly enough. To be fair, I use AI tools daily — for travel planning, market research, marketing copy, and I've vibe coded a side project app. (The tools have gotten remarkably better in just the last six months.) But I know I've only scratched the surface of understanding the role AI plays in my industry now and will in the future.
I also know I can't let this change happen to me. I want to understand enough to adapt on my own terms — to know which parts of my work I can hand off to AI and which parts deserve my full, undivided human effort.
So I set out on a retrospective quest. I went back through the meaningful projects of my 20+ year career and asked: How would these have looked different with today's AI? Which ones wouldn't have happened at all? And what does that tell me about what I should double down on?
Here's what I found.
The projects I examined
I picked three projects that span very different chapters of my career:
A university research thesis on ancient climate change (2003). I published a peer-reviewed paper examining Atlantic Ocean temperature changes during the early Pliocene (~4.1 million years ago). It involved weeks in a windowless lab picking up plankton skeletons with tweezers, crunching data with clunky old software, and synthesizing mountains of prior research.
A gamification initiative at Skillshare (2023). My team and I designed and launched a badge and certificate system to drive learner engagement. We dug into research on learning motivation, explored user behavior data, brainstormed rewards, worked with an illustrator on artwork, and navigated a complex cross-functional rollout involving marketing, legal, engineering, and the executive team.
Founding the Adobe Education Exchange (2010–2024). The highlight of my career. I saw a clear user problem — educators needed a space to find, share, and learn about teaching resources — and led the charge to build a platform that grew to over a million members. It was a classic founder journey: I wore every hat from product manager to customer support rep to t-shirt designer.
Where AI would have made things faster, easier, and better
Across all three projects, the pattern was clear: AI dramatically accelerates execution.
In my research thesis, AI could have saved enormous amounts of time synthesizing relevant studies and analyzing climate data. I recently used AI research tools on a consulting project and, as long as I fact- and gut-check the output, it's remarkably efficient. The data analysis that took me weeks with old software could have been done in a fraction of the time — and probably yielded deeper insights. (Many scientists are already using AI for data-rich research.)
At Skillshare, the gains would have been even more dramatic. AI data analysis tools could have helped my data science colleagues find the user actions that causally impacted business metrics and user motivation much faster. Vibe coding and rapid prototyping could have solved one of our most maddening UX challenges — designing a gamification experience prominent enough to notice but not so distracting it got in the way of learning. AI coding tools would have sped up the technical implementation significantly. And here's what excites me most: generative AI could have created personalized badges on the fly, tailored to each user's goals and behaviors. Imagine users showing off one-of-a-kind rewards to their community. AI can help product managers build for a wider variety of users without adding an untenable amount of engineering work.
For the Adobe Education Exchange, the efficiency story is similar — every operational challenge, from handling customer support to managing vendor agreements to localizing the platform into German and Japanese — could have been dramatically easier. As a founder, I could have used AI agents to alleviate the pain points that came from muddling through areas well outside my expertise.
The takeaway: almost every execution-heavy task I did in my career could be done faster with AI. That's not a hypothetical — it's already happening.
What AI could not have done
This is where it gets interesting. Across all three projects, the things AI couldn't do fell into a few consistent categories:
Framing the right problem and designing the approach. In my research, AI could not have done a good job identifying the research question, forming a hypothesis, or designing the methodology. That came from my advisor's expertise, the paleoceanography community, and years of accumulated judgment about what was worth studying. The same was true at Skillshare — no one at the company had ever really dug into gamification. Someone had to recognize the opportunity, frame the problem, and design an approach that would work within the business context.
Leading people through complexity. The Skillshare gamification project required collaboration across marketing, content, legal, customer support, engineering, and the executive team (including the board). The PM superpower is leading a process that produces an output greater than the sum of its parts — and that the whole organization feels ownership of. That's something AI can't do.
Building and sustaining community. This was the biggest lesson from the Adobe Education Exchange. AI can now generate instructional materials and expedite professional learning for educators. But community — the culture of knowledge sharing, the trust that forms when educators help other educators — is the element AI can't replicate. If I were building the Education Exchange today, it would probably be more of a community layer on top of AI tools and less of a content repository. I love imagining educators using AI to remix materials informed by real-life data on how those materials worked in another educator's classroom. The community provides the context and trust that makes AI output meaningful.
Creating moments of delight. When done well, earning a badge should make you smile. At Skillshare we added animations; in a gamification project at Adobe, we showered the screen with confetti when users earned a certificate. The creativity and imagination to make an experience genuinely fun still requires a human touch.
The learning that comes from doing hard things. This one surprised me. At Adobe, the operational struggles of building a platform — the legal work, the vendor negotiations, the support requests, even the bad actors — taught me things that made me a fundamentally better product leader. If AI had handled all of that, I would have missed the intimate knowledge that led to better decisions and deeper conviction. I can't help but wonder: what is the impact of AI tools on the quality of products and the resilience of founders? Do we even know yet?
What this taught me about the future of my work
The exercise clarified something I'd been feeling but couldn't articulate: the PM's job isn't going away — it's getting harder. When everything is easier to build, the critical question shifts from "can we build this?" to "should we build this?" Mediating between user needs, UX design, engineering capacity, cross-functional alignment, and executive buy-in is still essential work. In fact, when the cost of building drops and the number of possible solutions explodes, the ability to make the right call on what to build becomes even more valuable.
Here's how I'd summarize the personal framework that emerged from this retrospective:
Hand off to AI: Research synthesis, data analysis, rapid prototyping, technical implementation, operational tasks, localization, and the many forms of grunt work that consume time but don't require judgment.
Double down as a human: Problem framing, cross-functional leadership, community building, creative vision, and the kind of learning that only comes from doing things the hard way — at least once.
I went into this exercise a bit nervous — worried that the answer might be "yes, AI will replace me." What I found instead is that AI will replace parts of what I do, and that's a gift. It frees me to spend more time on the parts that matter most, the parts that are distinctly and irreplaceably human.
The question was never really "will AI replace me?" It was "do I know which parts of me to keep?"
I think I'm getting closer to an answer.
I'd love to hear your perspective. What did I get wrong? What did I miss? Where is AI further along than I think? Let me know in the comments.