Last summer I had major surgery and haemorrhaged 2.4 litres of blood. I only found out later that a person my weight has about 3.5 litres total — meaning I’d lost almost 70% of the blood in my body. For a few days, I fainted every time I tried to stand up.

About a month after that, I was lying on the sofa watching an AI tool tutorial on YouTube.

Not because I felt well enough to. Because I felt like I had to.

My LinkedIn feed was relentless — people shipping agents, automating entire workflows, building things that seemed impossibly sophisticated. The AI development pace felt like a treadmill someone had cranked up while I was in the hospital.

I tried to catch up. I bookmarked tutorials, saved podcasts, signed up for courses. But the longer that list grew, the less I wanted to touch any of it. My body was still recovering, but my mind was drowning in a backlog I’d created for myself.

But I didn’t always feel this way.

. . .

Ages ago, a few years before ChatGPT existed, I was already tinkering with GPT-2. I spent evenings and weekends playing with no-code tools, messing around with SEO automation, learning data models and analytics. Nobody told me to. There was no career pressure attached to it. I did it because I found it genuinely interesting — the kind of tinkering where you lose track of time without meaning to.

That version of me wasn’t “keeping up.” She was just curious. (Cue: She Used to Be Mine by Sara Bareilles)

Here’s what I think happened between then and now: the AI industry got very good at manufacturing urgency.

The people who benefit financially from our anxiety — poeple who sell online courses, AI consultants, or, God forbid, “AI thought leaders” — have turned AI education into a fear campaign. The message, dressed up in different ways, is always the same: [Your job] is dead, here’s how AI is revolutionizing it, and if you don’t know how to use [latest tool] you’re going to be left behind.

And it works. I’ve had enough private conversations with solid, experienced product folks to know I’m not alone in this. Perfectly competent product leaders feeling secretly inadequate because they haven’t tried OpenClaw yet, or because they are still wrapping their heads around Claude Code, or because someone on LinkedIn just raised $100M on a project they vibe-coded over the weekend.

The problem with fear-based learning isn’t just that it feels bad. It’s that it doesn’t actually work.

I’ve noticed that when I’m learning to avoid falling behind, I’m not really learning. I’m performing learning. I bookmark the video, subscribe to the webinar, save the thread — and then avoid opening any of it. Because the motivation was fear and anxiety, not love and curiosity. And that’s exhausting to sustain.

Contrast that with how I felt tinkering with GPT-2 back in the day. I wasn’t trying to cover everything. I was just following whatever seemed interesting that evening. Things stuck because I actually cared about them. I discovered new skills. I built something small and immediately wanted to build something else.

. . .

Yesterday, I rebuilt Thinkers and Tinkerers website using Claude.

I would normally have reached for Carrd or Framer — reliable no-code tools that get the job done. But this time, I just started tinkering. HTML and CSS, back to basics, with Claude helping me figure things out as I went. The result is nothing impressive. It’s just a simple, static site.

But here’s what it did: it beat the inertia.

I’d been stuck in that paralysed state for months — saving things to watch, not watching them, feeling vaguely guilty about it. And then one afternoon of low-stakes tinkering snapped me out of it. Not because of what I built, but because I remembered what it felt like to be interested in something rather than anxious about it.

The bar for “starting” is so much lower than the LinkedIn feed makes it look. It doesn’t have to be an autonomous agent. It doesn’t have to impress anyone. It just has to be something you’re actually curious about.

. . .

So here’s the question I’m sitting with, and maybe you are too:

If you stripped out all the fear and anxiety — what would you actually want to learn? What would you explore just because it seemed interesting?

That’s the thing worth starting with. Not the tutorial that promises to make you “AI-native.” Not the tool everyone’s talking about this week. The thing that makes you a little bit curious, even now, underneath the noise.

Start with the thing you’d do if you weren’t afraid.

Maybe it’s just me, but the last few months of 2025 felt so chaotic. Every week brought some “gamechanging” AI news. Latest models! New tools! AGI! Our LinkedIn feed was full of people launching yet another AI startup, showcasing what they built in 30 seconds, new workflows that would “completely revolutionize” our job, and reasons why we should rethink our career.

It’s hard not to feel FOMO these days. And it’s exhausting — trying to keep up with every AI development is now a full-time job. And I would argue that it’s actually making you worse at your actual job.

Your 2026 resolution shouldn’t be to try more AI tools. It should be to go on an AI diet.

“It’s like Christmas everyday!”

I remember Tobi Lütke, founder of Shopify, excitedly made that statement about the rapid AI development. Every new AI tool felt like a revelation. ChatGPT went from text to vision to voice to search. Replit and Lovable made prototyping instant. Cursor made coding accessible. Claude got Projects, artifacts, and extended context.

It’s tempting to drop everything and try the new tools immediately. After all, you don’t want to be the dinosaurs that adopt the tools a little too late. It was a sensible mindset when AI was new. Exploring these tools was super important — you needed to understand what was possible.

But guess what, Tobi? Nobody works at Christmas. Nobody gets anything done — there’s too much food to eat, too many presents to open, too many people to talk to.

I’m not thinking that the more AI tools you try, the less value you extract from any of them. When you chase every new release, you’re constantly in beginner mode, never developing mastery. You’re optimizing for novelty, not utility. You’re spending more time learning tools than solving problems. You’re reacting to what’s new instead of choosing what’s useful.

Every hour learning a tool you don’t need is an hour not getting better at something that actually matters for your work.

What you need in 2026: Careful curation, not consumption

Not “what’s new?” but “what’s essential for my actual work?” Not “what’s possible?” but “what’s worth the learning curve?”

Going on an AI diet means being deliberate about what you let into your workflow.

1. Audit what you actually need

Start with the job, not the tools. What problems am I solving repeatedly? What tasks genuinely benefit from AI acceleration? What capabilities would create leverage in my specific role?

For a PM, this might be research synthesis, data analysis, competitor research, prototyping for specific validation questions, drafting specs after thinking is done.

Unless it’s actually useful for you, there’s no point learning 17 different ways to generate images and edit videos.

2. Choose your core stack

Pick the tools you’ll master, and learn it deeply. Most PMs only need one general LLM, one prototyping tool, and maybe one specialized tool for your domain.

That’s it. Everything else is optional. Resist the urge to add more just because something new dropped.

The power isn’t in having access to 20 tools. It’s in deeply understanding 3 tools so you can use them without thinking.

(Sorry, can’t resist adding this meme)

3. Set a “time budget” for new tools

Literally ration how much time you spend evaluating new AI tools. Maybe it’s two hours per month. Maybe it’s one new tool per quarter. Whatever feels sustainable.

When something new launches, ask: Does this solve a problem I currently have? Is it meaningfully better than what I already use? Is the switching cost worth the marginal improvement?

If the answer to all three isn’t “yes,” it’s a no.

4. Stop following AI news daily

The AI news cycle is designed to create FOMO. Every release is “groundbreaking.” Every update is “game-changing.”

Most of it doesn’t matter for your actual work.

Check in monthly, not daily. Let other people be the early adopters. You’ll hear about the truly important developments anyway. By being more intentional about your learning, you actually reserve your brain space for things that are truly important.

Your job is product management, not AI journalism.

5. Measure by outcome, not output

The wrong metric: “How many AI workflows and agents have I built?”

The right metric: “How much better or faster am I at the work that matters?”

If adding a new tool doesn’t measurably improve your outcome, it’s noise. Quoting a classic from Peter Drucker: “There is nothing more useless than doing with great efficiency something that should not be done at all.”

Bonus point: Maybe don’t build that AI feature?

Just because you can add an AI feature doesn’t mean you should.

Every product is getting an AI button now. Most of them are solving problems users don’t have, adding complexity users don’t want, and competing on “AI-powered” as a feature rather than actual value.

Before building any AI feature, ask: What user problem is this solving? Be specific. Does AI actually make the solution better, or just different? Will this create real value, or just check the “LOOK, WE HAVE AI” box? Are we adding this because our users need it, or just because our competitors have it?

The best AI features are invisible. They make existing workflows faster or easier without making users roll their eyes and think, “Yay, another AI feature I don’t need.”

It’s time to tighten your belt

Going on an AI diet as a product builder means ruthlessly cutting AI workflows and features that don’t create clear, measurable value. Having the clarity to say “we don’t need AI here” when that’s true.

The PMs who thrive in 2026 won’t be the ones with the longest list of AI tools in their repertoire. They’ll be the ones who mastered a few selected tools deeply, ignored the noise, and spent their energy solving real problems instead of chasing shiny objects.

I often see the same pattern play out.

Somebody excitedly shares how they prototyped a new feature in two hours. They show their team a working demo. Everyone’s impressed by the speed. Three weeks later, the feature is quietly shelved because it solved the wrong problem.

AI gave them speed. But speed masked that they never stopped to think.

In my last article, I talked about how AI removes or automates many of the busywork of product development. But some shortcuts don’t save time. They skip essential work that looks slow but is actually the work.

This is the new trap for product managers. You might actually be atrophying your critical thinking muscle by outsourcing too much to AI. Some examples:

When AI prototypes replace real thinking

The shortcut is seductive: spin up a working prototype in hours. Show stakeholders something tangible. Watch them get excited. Feel like you’re making progress.

But you’ve just anchored everyone on a solution before confirming the problem is worth solving. You’ve optimized the easy part — making something — and skipped the hardest part: making sure it matters.

The best companies in the world like Meta, Shopify, and Amazon share one thing in common: a culture of writing. I’m a strong believer that writing is thinking — when you write a narrative document, you untangle a woolly set of insights and complaints into a clear problem definition. You sketch on whiteboards. You debate with engineers. You sit with the discomfort of not knowing the answer yet. This is how you actually build your product sense.

When you skip the heavylifting and go straight to prototyping, you never develop the muscle. I’m not saying you should skip AI prototyping entirely — it’s to be careful when you use it.

Use AI for research synthesis first — summarize user interviews, identify patterns in feedback, map out the problem space. Prototype on paper to keep it ugly enough that it invites real critique. Only reach for AI prototyping after you’ve validated two things: the problem is actually valid, and your solution direction is roughly right.

Before you prompt an AI prototyping tool, ask yourself: “What evidence do I have that this deserves a solution?” If the answer is thin, prototyping is the wrong tool. You’re using a screwdriver to hammer a nail.

When AI “customer interviews” replace real conversations

The five-minute shortcut: ask ChatGPT to role-play as your target customer. Generate persona descriptions. Get “insights” about their pain points without leaving your desk.

This is probably the biggest crime you can commit as a product manager.

AI generates statistically average responses. It smooths over the weird, specific, surprising details that actually matter. The marketing manager who prints out every dashboard on Friday afternoons because she doesn’t trust digital-only charts. The small business owner who uses your software in ways you never designed for because it solves a problem you didn’t know existed. The user who almost churned three times but stayed because of one specific feature you nearly deprecated.

These are the insights that shape great products. AI can’t give them to you because they’re not average. They’re outliers, edge cases, surprising behaviors that reveal what really matters.

You’re also training yourself to accept synthetic empathy instead of building real understanding. Real customer conversations have texture. You hear the hesitation in someone’s voice when they say something is “fine.” You notice what they don’t say. You pick up on the problem behind the problem because you asked a follow-up question to something unexpected.

The thoughtful alternative is to use AI as a multiplier, not a replacement. Use it to prepare for customer conversations — schedule interviews, draft interview questions, synthesize past research, identify gaps in your understanding. Use it to analyze interviews after you’ve done them: transcription, theme extraction, and clustering similar quotes across multiple conversations to spot patterns you might have missed.

When AI gives you an illusion of validation

AI can make any idea sound viable. Ask it if your concept makes sense and it will confidently articulate benefits, edge cases, risks, and user motivations. It will explain your idea back to you in a way that feels polished and persuasive.

But it’s doing is only pattern-matching from the entire internet and giving you a coherent narrative.

I once heard this saying: AI makes mediocre PMs faster at being mediocre. This is one sure way to achieve it. If you’re using AI to confirm rather than challenge, you’re just accelerating bad judgment. Great PMs use AI to stress-test their thinking, not rubber-stamp it.

The thoughtful alternative is to flip how you use AI in validation. Don’t ask it to confirm your hypothesis. Ask it to destroy it.

Use AI as a devil’s advocate: “What’s wrong with this idea? What am I missing? What would make this fail?” Use it to generate counter-hypotheses, not just validate yours. Use it to identify what you should validate with real users, not to replace that validation.

When AI shortcuts create shallow strategy

AI is very good at structure.It can generate crisp strategy documents, articulate goals, and outline roadmaps with almost eerie confidence.

Feed AI your team’s OKRs, last quarter’s retrospective, and some market context. Ask it to write your next quarter’s strategy and roadmap.

The output looks right. It reads like strategy. It contains the words we associate with strategy.

But real strategy is choice. It’s about what you won’t do, which users you won’t serve, and what success won’t look like. It’s about constraints and conscious trade-offs. It is about how the pieces fit together, not how the document reads.

The thoughtful alternative is to use AI for the inputs to strategy, not the strategy itself. Use it to research the competitive landscape, synthesize market trends, pull together data on user behavior.

And then, once you’ve written your first strategy draft, use it as a sparring partner to stress-test your thinking: “What would a skeptic say? What am I under-weighting?”

If you’re using AI to write your strategy, congrats, you’ve just abdicated the most important part of your job.

To tie it all together…

The best product managers I know are leaning into AI hard — but they’re doing it selectively. They’re using it to eliminate the parts of their job that don’t require judgment so they have more time for the parts that do. They’re getting faster at execution without getting sloppier at thinking.

It’s all about recognizing where shortcuts help you, and where they slowly erode your craft.

For the past decade, “product manager” has meant many things — strategist, roadmap owner, cross-functional diplomat, team glue, meeting facilitator. But as AI reshapes how products are built, a new archetype is now clearly emerging: the product builder.

The title change isn’t just semantics. It aligns with what I’ve argued for years: PMs need to step up, or risk becoming completely irrelevant. Many companies like LinkedIn, Zapier, and Prolific are redesigning roles around building, not managing.

AI has automated the mechanical side of product work, such as documentation, synthesis, and coordination between teams. This shift doesn’t erase the fundamentals of product craft; it amplifies them. Your job is still to find meaningful problems and deliver value — but now, your ability to think and execute with AI determines how far and fast you can go.

First principles still matter

The fundamentals of good product work haven’t changed:

• Deeply understanding your customer — their jobs-to-be-done and the gaps in completing them

• Deeply understanding the market you’re operating in, its competitive landscape, and how your product stands out

• Choosing and building a solution with the highest ROI

• And, as always, working with internal stakeholders and leading a team of designers and engineers to achieve the above

AI doesn’t alter these principles — it distinguishes PMs who apply their taste, critical thinking and judgment from those who merely facilitate. As AI makes the busywork disappear, there’s less room to hide.

So what does it actually look like to be a product builder? Among the many great PMs I’ve had the pleasure of working with or mentoring, I’ve noticed shared patterns — traits that define what it means to be an AI-fluent product builder:

1. Bias to ship, learn, and loop

AI shortens every feedback cycle. Prototypes that once took a week now take an afternoon. Summaries, analyses, and design variants appear in seconds.

Dharmesh Shah, HubSpot’s founder and one of the biggest proponents of AI, put it simply: “Never spend more time debating the priority of a product improvement than it would take to just do it.” HubSpot teams revived their “daily delights” ritual — shipping one small user-visible improvement every day. The compounding effect of that rhythm is enormous.

AI makes that cadence attainable for everyone. The learning loop tightens from weeks to hours. But here’s the catch: You have to treat this as a superpower for learning, not a license for shipping shitty features for the sake of speed.

Try this:

• Every week, pick one friction in your product workflow

• Prototype a fix using AI and test it

• Share what you learned, not what you shipped

2. Cross-craft fluency

I’ve touched on this in my last article: the days of staying in your lane are gone. AI makes it possible — and expected — for PMs to cross the old boundaries of design, research, data, and engineering.

Product Builders build prototypes to test flows, write SQL to analyze metrics, and tweak front-end code with AI assistance. We are seeing more and more roles converge. Instead of four specialized contributors — analyst, designer, PM, engineer — you now see pairs of hybrid builders who can execute together. You don’t need mastery of every craft, but you do need literacy, and build it up to fluency.

Try this:

• Pick one adjacent skill (design, data, or code)

• Use AI to complete one small task in that domain — e.g., generate a SQL query or design a wireframe

• Ask your expert peer to review and give feedback

PS: To be clear, this isn’t about taking anyone’s job. AI is a leveller. Designers, engineers, analysts, and PMs can all reach across the aisle to prototype, test, and ship. When you learn each other’s languages, you can move faster together.

3. Judgment and taste

AI can generate endless options. The new skill is knowing which one is worth doing.

Judgment now differentiates great builders from fast ones. When everything can be built, what should be built becomes the next hard question.

Dhanji Prasanna, Block’s CTO, warned that “we’ll need human taste to anchor these AIs so they don’t go off-script.” Taste isn’t aesthetics; it’s discernment — the ability to know when simplicity serves the user better than cleverness. In her latest blog, Elena Verna framed it nicely: “Taste is how a product earns trust. If it looks and feels like everything else, it gets forgotten like everything else.” As algorithms dominate distribution, the only moat left is emotional connection.

AI can help you brainstorm design, copy or flows, but it can’t sense when something feels off. That’s your job.

Try this:

• Build a “taste library”: five products you admire for clarity and emotional pull. Reverse-engineer why they feel good to use.

• Run “taste reviews” in your team — 15-minute critiques of copy, look and feel, or tone.

• Ask the AI to generate five versions of a message, then pick one and explain why to a peer. You’re training your discernment muscle.

4. Systems thinking

In product work, a system is any process or product loop that gets smarter or more efficient over time.

“Systems thinking” means you stop viewing your work as a chain of isolated tasks (launch this, fix that) and start seeing how each decision feeds back into the next one — through data, learning, or team habits.

For example:

• When you automate how user feedback is categorized, you’re not just saving time. You’re building a system that improves every week as more data flows through.

• When you design metrics and review cadences that teach the team something after every release, you’re building a learning system.

• When you set up shared AI tools that remember context across projects, you’re creating a knowledge system.

It’s the shift from “How do I deliver this feature?” to “How do I make the whole process of delivering features faster and smarter next time?”

AI supercharges this because it can connect and optimize across these loops — automating data capture, summarizing insights, or spotting patterns you’d miss manually.

Try this:

• Map one workflow you repeat often (research synthesis, roadmap reviews, launch tracking). Ask: what parts teach the next cycle nothing?

• Automate or document those dead ends so they feed back into learning next time.

• When you introduce an AI tool, define the loop upfront: what gets smarter each time this runs?

• Build small data habits: tag feedback, track usage patterns, and share them — you’re gradually building the system’s memory.

5. Ownership mindset

When anyone can automate reports, synthesize feedback, or test a prototype, there’s no reason to wait for permission. Yet many PMs still operate as if someone else owns the solution space. Product builders don’t. They run toward ambiguity and pick up loose ends without waiting around.

In companies where AI transformation worked effectively, it’s usually not because leadership handed down a plan, but because individuals took ownership. Teams built their own automations, shared results, and improved them collectively. Adoption spread bottom-up — a culture of builders solving their own problems.

Ownership mindset shows when you move past being annoyed at a bottleneck and actually fix it. It’s the moment you stop waiting for someone else to clear the path. (A bit of bragging — but this has been consistent feedback from my managers across different companies. At Meta, the note that came with my promotion and “redefines expectations” rating said: “When you find a problem, you don’t rest until you fix it.” That, to me, is the essence of ownership.)

Try this:

• Identify one process everyone complains about but no one owns. Try to automate at least part of it with AI tools.

• When something feels “not my job,” ask instead, “who’s better placed than me to start it?” — then start it anyway.

6. Hands-on leadership credibility

AI has flattened hierarchies. Many companies now care less about “how big of a team can you manage?” and more about “how hands-on can you be?” It’s now a common expectation for a product leader (GPM, Head of Product, or even Product Director) to do some IC work.

Hands-on product leaders gain credibility not through title or scope, but by staying close to the ground and being resourceful. You can actually make faster, sharper calls because you don’t rely on filtered updates. Staying close to the work doesn’t make you less strategic. It makes your strategy grounded.

If you manage a team, you are also best placed to drive meaningful AI adoption across their teams. When you model how to use AI in real workflows, adoption follows naturally. It signals that experimentation is safe and expected, and it grounds the AI conversation in real impact, not abstract enthusiasm.

Try this:

• Block 90 minutes each week to build something yourself

• In team meetings, share one concrete way you used AI to save time or improve quality

• Run a short “show and tell” session where everyone demos one AI-powered workflow they’ve built

• Ask your team where AI has not helped — these gaps often reveal process friction worth fixing

7. Last but not least: Ethical and sustainable judgment

Speed is seductive. AI can help you move 10x faster — but without guardrails, you risk breaking trust just as quickly. Ethical builders know when to slow down. They decide where humans must stay in the loop, how data should be used, and what responsible trade-offs look like for their product.

I’ve seen firsthand how “move fast and break things” turned dire at Meta. (I could speak for hours on this, so don’t tempt me.) Governance and guardrails shouldn’t be afterthoughts. Treat them as design principles, not constraints.

Try this:

• Create a three-point check before shipping anything AI-powered:

  1. Could this mislead, exclude, or disadvantage a group of users?

  2. What mechanism do I have in place to alert me when the automation goes wrong?

  3. Can a human audit or override the system easily?

• Add a “trust review” to your release checklist — just like a bug bash, but for user confidence.

Final thought

The shift from “manager” to “builder” isn’t a demotion — it’s a return to what made product work exciting in the first place.
AI has stripped away layers of process and admin, leaving us face to face with the real craft: solving problems, making good decisions, and shipping things that matter.

Those who embrace that shift — who lead with clarity and curiosity — will shape not just better products, but a stronger product culture built on first principles.

"Would you rather ship twice as many features — or let go of half your team?" That was a question a CEO asked me a few weeks ago. It's not just a purely hypothetical question. We were discussing how AI tools like Cursor have sped up software development massively, and the rate of change is staggering. In the past 6-12 months, we've seen countless new tools and updates that disrupt the product development cycle.

Picture this: AI can augment tech teams and double the speed of product development. What used to take a month now takes two weeks. What does it mean for your product team?

At first, this seems like good news. Every product person I know wants their engineering team to build faster. More features! Better products! Finally matching our grand vision!

But then what? Once you reach the sweet spot of your product vision, would you want to build more features? This is when you start saying, "Eeehh" unconvincingly. And your intuition is not wrong. There's a diminishing return or even a dip in user satisfaction once you pass a certain stage.

The paradox of AI-powered productivity

Here's the uncomfortable truth: AI enables us to build more, but building more could make your products worse. So once you've reached this happy user peak, you, or your manager, will probably be at a point where you realize you don't have to build as many features anymore. Sure, you have to keep up with or exceed your competitors and continuously improve the product, but do you need 100% of the people in your team to do that?

The harsh answer is: probably not. We're already seeing layoffs and fewer roles available. This is our new reality:

• There will be less demand for our roles

• The remaining roles are changing at a rapid rate

• We all face stiffer competition, either to keep our job, or to find a new job

There's no other option than to upskill and adapt.

The architecture of modern tech organizations

Let's go back to a typical organization structure where the product is software. It usually looks like this:

When ChatGPT was launched 2-3 years ago, many people predicted support functions would be hit first — especially customer service. They were right.

But what's happening to the layer above? There are two fundamental shifts happening:

First: Efficiency within each role

I’m sure you have seen this happened in your company, or to your friend. A marketing team used to have three people. One of them resigned or was let go, and wasn’t replaced. The other two remaining people are expected to cover the workload.

That marketing team still needs to hit the same lead generation numbers, launch the same number of campaigns, and maintain the same quality standards—just with fewer people.

This creates a brutal but clear dynamic: survive or get cut in the next round.

So if you're one of the people who remains in the organization, you naturally turn to AI to make your job more efficient and enable you to do what used to require 2-3 people.

AI isn’t a magic productivity wand, but it’s very good at amplifying your existing capabilities. The person who thrives isn't the one who can prompt ChatGPT the fastest — it's the one who can critically evaluate what comes out, refine it with expert knowledge, and sharpen it with years of hard-earned intuition.

You fire-proof yourself (aka staying above the firing line) by becoming the person who can wield AI like a precision instrument rather than a blunt tool. This requires being genuinely excellent at your core craft — having the taste and judgment to know when AI's output is brilliant, mediocre, or completely off-base.

Second: Consolidation of roles

This shift is less common, newer, and we are still grappling with it: AI has democratized access to the skills that were historically gatekept within specific roles and took years to learn. Now anyone with curiosity and critical thinking could access these skills: Copywriting, creating compelling value props, conducting user research and synthesizing the results, designing beautiful interface, coding, data analysis… you name it.

Think about what used to be impossible without years of training and experience:

• A PM couldn't create pixel-perfect mockups without years of learning and practicing design

• A designer couldn't code the front-end of their own design

• A user researcher couldn't analyze complex datasets without SQL knowledge and statistical training

Now? AI tools have compressed these learning curves from years to weeks, sometimes days. The barriers between roles are dissolving. What matters now isn't just your core specialty — it's your ability to recognize good work across disciplines and direct AI to produce it.

The result? Companies are starting to hire fewer specialists and more AI-savvy generalists who can wear multiple hats effectively.

Beyond T-shaped: The flower-shaped individual

You've heard of T-shape individuals — broad knowledge with one deep specialty. You might know pi-shaped (π) individuals — two deep specialties. But I want to introduce you to the flower-shaped individual:

Let’s break it down. In the middle is the core skills and traits that are non-negotiable. In the age of AI, these are:

• Human relationships: The ability to build strong connections and collaborate effectively

• Curiosity and hunger to learn: How else would you keep up with rapid developments?

• User empathy: Deep understanding of your users, what makes them tick, their pain points and their needs.

• Adaptability: The agility to evolve quickly. Think about the disruption we’ve experienced in just the last five years — it’s not slowing down any time soon.

• Humility: Recognizing AI's capabilities rather than dismissing it as "not good enough" to impact your role

Each petal represents specialization in different functional areas. You start by becoming literate in each area, then gradually develop fluency. The more petals you can cultivate, the more valuable you become to any organization.

Two paths to thrive in the AI world

My prediction is that there will be two types of super ICs who will thrive in the next few years:

1. The Specialist

You're the top 1% at one petal. You have very specialized skills from years of experience honing and refining them.

Think of the senior software engineer who can design systems that AI can't conceptualize, or the PM who has such deep industry knowledge and intuition about user behavior that they can spot the nuanced patterns AI misses.

The specialist path works because AI still struggles with edge cases, complex judgment calls, and situations requiring years of contextual experience.

2. The Orchestrator

You're a generalist with solid understanding of multiple petals, able to orchestrate and supervise AI across domains.

The orchestrator is like a conductor leading an AI symphony. They might not be the best designer, researcher, or analyst in the room, but they know enough about each discipline to effectively prompt AI tools, critique outputs, and weave everything together into an excellent result.

Picture a PM who uses AI to generate initial user personas, refine them with AI-powered research synthesis, create mockups with AI design tools, and analyze user feedback with AI — all while applying their judgment to guide the process and make final calls.

You'll probably thrive better in the startup environment where wearing multiple hats is expected, and where the ability to move fast across disciplines with AI as your force multiplier becomes a massive competitive advantage.

Both paths require one crucial skill: knowing when to trust AI and when to override it. That's where your human judgment becomes irreplaceable.

Burying your head in the sand isn’t an option anymore

The shift isn't coming — it's already here. Right now, while you're reading this, someone with your exact job title is learning to do twice the work with AI. Another person is figuring out how to eliminate your role entirely.

If you're a mediocre performer banking on one petal alone, you're already obsolete. You just don't know it yet.

Here's what separates the survivors from the casualties: survivors don't wait for permission to evolve. They don't wait for their company to train them or for the "perfect moment" to upskill. They're already growing their petals while others are still debating whether AI is a threat.

The question isn't whether AI will reshape our industry — it already has. The only question that matters is: Will you be shaping that future, or will you be shaped by it?

The rapid development of AI has been a source of anxiety for a lot of people. CEOs and founders are under pressure to incorporate AI into every aspect — from the website’s headline, sales or investor decks, user-facing features, to their company strategy. Executives and managerial levels are pressured for more efficiency by utilizing AI tools, whilst they often haven’t caught up with the latest developments of AI beyond using ChatGPT every now and then. Team members are anxious about losing their jobs to AI — the constant stream of layoffs doesn’t help. CEO memos like Shopify’s and Duolingo’s are flooding people’s timelines, some more controversial than others.

It’s an unsettling time in tech.

Some people would say that the hype will pass, that it’s just another fad like blockchain and web3. I fully disagree though — AI is truly changing the way we work, create, and develop products, and the sooner you embrace that, the better.

As the CPO of Zero Gravity, it’s my job to stay ahead of the game (and my CEO nagging me helps too), so I crystallized my thinking into this AI strategy. It’s nothing fancy — not the kind of strategy you’d get from McKinsey or Bain. This is a simple 3-prong approach to AI strategy that ensures we leverage AI in different aspects of the company. The nice thing about this framework? You don't need to be an AI expert or have a massive budget to get started. Let me walk you through each part.

Prong #1: Craft Optimization

This is the entry-level AI usage that you've probably been doing since the launch of ChatGPT. ChatGPT, Perplexity, and other general AI assistants are definitely super helpful for research, brainstorming, and writing. But don’t let yourself get stuck there.

The advancement of LLMs means that, more likely than not, there’s already a specialized AI tool out there that focuses on helping you do your job better and faster. Just some examples:

• Gong for the Sales team: It captures and analyses every interaction with the customers and provides insights to improve sales performance.

• Jasper for the Marketing team: An AI content generation tool that assists marketing teams in creating high-quality copy for blogs, ads, and emails.

• ChatPRD, Cursor, and Magic Patterns: These tools assist product and engineering teams in drafting product requirement documents, coding, and designing UI components efficiently.

I understand the hesitation about adopting these tools. I know in some cases AI takes away the fun part. But hey, I personally believe that resisting it only harms your own career. The now-cliché adage that “You won’t be replaced by AI but you will be replaced by somebody who knows how to use AI” is true.

As a leader, I argue that our job is to create a safe environment for your team to try out the tools. Make it fun — create a ‘show and tell’, and award prizes for the most innovative use of AI. Give them the budget and freedom to try out new tools. Bottom-up adoption will work much better than top-down — they are more likely to utilize the tool if they choose it themselves. Your team are the subject matter experts anyway, so you should trust them to evaluate the tool as such. If you put too many procurement and security red tapes before allowing them to try out new tools, you’re killing innovation in your company.

Prong #2: Product Innovation

This prong is probably the one the investors are most keen to see. You probably already have one or two AI-powered features in your product now.

The flood of AI-native tools out there might make you think that it’s too late, or the market is too saturated to incorporate yet another AI feature into your product. But the game has just started, and the competitive landscape is still being shaped. If you already have access to a certain group of users and proprietary data about their goals and behaviours, you’re already miles ahead. As long as it still aligns with your users’ needs and problems, an AI-based solution is definitely worth considering.

I used to hesitate — do we really need AI for this? Is it too gimmicky? Now I’m using this heuristic: Could I see my biggest competitor launch this AI feature? Would I be seething if I were reading it on TechCrunch? If the answer is yes, I might as well be the one building this feature.

Prong #3: Internal Operations

This is the one I would argue is relatively unexplored compared to the other two. This prong is about streamlining your company’s day-to-day collaboration and communication. I’m sure you’re already implementing tools like Zapier, Asana, and Notion in your company that do a decent job of holding the company operations together. (If you haven’t, then you have a bigger problem, so maybe start there first.)

The tools above are mainly still operated by humans. Your team did a 1-hour meeting, documented the notes on Notion, and wrote some action items on Asana. A few days later, a team member would take a ticket from the Asana board and execute it.

AI agents can now be utilized to run many operational steps automatically. However, the critical prerequisite for building effective AI agents is high-quality input data. While the inner workings of AI may remain a black box (though it's my mission through this newsletter to demystify it), one thing is non-negotiable: you must feed it high-quality input data.

In a company context, much of this vital input data exists only in your team's heads, shared primarily through meetings and discussions. The first step to capturing this knowledge pool is consistent meeting transcription. Not utilizing AI to transcribe, summarize, and tag every meeting is a missed opportunity. Fortunately, you have many AI transcription tool options – even Notion now has this built into their product, making it seamless if you're already using it for company knowledge management.

With rich meeting notes as input, you can then create AI workflows and agents to automatically handle next steps – whether that's sending follow-up emails, drafting social posts, or even fixing simple code bugs. While AI agents are still in their early stages, we can fully expect their capabilities to advance dramatically even within the next year. Regardless of current limitations, starting to capture high-quality input data now will position you to take full advantage as the technology matures.

Now, how are we going to operationalize this strategy?

A strategy is only as good as its execution. And I believe that buy-in from every layer is the key here — otherwise, you’re perceived as either the bad guy who’s trying to replace workers with AI, or the deluded leader who’s too optimistic about AI. This is how I operationalize AI strategy in my company:

• Set up the principles and the definition of success for each prong. I don’t dictate what people should do; I give them the guardrails.

• Appoint a champion (or ask people to volunteer as a champion) for each prong. If you have a big team, you might need one champion from each sub-team. The champion’s job is to lead the exploration of new AI tools and get their teammates to use them, collect feedback, and iterate. That way, it’s not “Debbie from the top” who insisted that the developers should use Cursor for coding; it’s “Alex the Principal Engineer” who advocated for this tool.

• Set up a meeting cadence with your champion — more frequent during the setup period, and less frequent for check-ins. Evaluate against the principles and definition of success you have created in the beginning.

Above is an example for one prong. There are three elements of each strategy:

1. Principle: It acts as a heuristic for the champions. Whenever they want to suggest implementing a new tool, they should ask themselves, “Is this removing admin, or adding admin?”

2. Areas: To give the champions a starting ground and some ideas. Not meant to be exhaustive.

3. Success definition: To allow us to evaluate whether the implementation has been successful. Note that I added some control metrics there as well, e.g. “without losing relationships or context”. The success metrics aren’t easily measurable, but many important things in life aren’t.

You can download this editable template if you want to adjust it for your company. (No catch, no email address required, I’m just nice like that.)

Download AI Strategy template

Hope this brings you a clear, actionable framework to implement AI strategy in your organization. As I said, it’s nothing fancy, but it works!

How are you approaching AI strategy in your company? I’d love to hear — drop me a note in the comments!

In my previous post, I discussed how AI makes product development cheaper and faster than ever. When a competitor or new entrant can clone your product in weeks or months, how do you stay ahead in the game?

This post explores the first pillar of the 3D’s: Data.

I know the idea of data as a competitive advantage isn’t new — it goes way back to the rise of the dot-coms. Your local supermarket or store used to know nothing about you: who you are, what you buy, whether you splurge on some items at certain times. But with the move to online shopping, where every purchase was tied to your user account, stores suddenly had the ability to analyse your behavior and predict what you might want or need next. You might have heard that old story where a father complained to an online retailer for promoting pregnancy products to his teenage daughter — only to discover she was indeed pregnant, and the store’s algorithm had figured it out before he did.

Despite how long this idea has been around, I think it’s still worth writing about. The newer generation of founders and tech leaders are *shudder* young enough to be digital natives — they’ve never lived in a purely offline world — and might take data collection for granted.

What kind of data are we talking about?

When I say "data," I’m not talking about big numbers on a dashboard. I mean the specific, actionable data about your users: who they are, how they behave, and how they use your product to get their jobs done. Broadly, there are three main types of data you should care about:

1. Self-reported data

This is the information users provide directly. It could be their age, gender, date of birth, job role, company size, goals, or preferences — whatever you ask them during sign-up, onboarding, or later down the line. It’s the simplest type of data to collect, but it comes with limitations.

First, no one loves filling out forms, especially if they’re trying to get into your product quickly. Overdoing it at onboarding can hurt your activation funnel. Second, this data becomes outdated fast. Many tools ask for someone’s role at sign-up and never check in again. Roles and goals change, and stale data doesn’t help anyone.

2. Behavioral data

This is where things get more interesting. Instead of asking users what they think they do, you observe what they actually do. Which features do they use? How long are they spending on a specific workflow? What’s their typical path through the product? This kind of data gives you a clearer picture of user behavior and intent.

As a leader, it’s easy to assume you’re collecting everything automatically through your product. Hate to break it to you, but you’re probably not. Engineers don’t usually log every single action by default — it costs time, storage, and can slow the system down. If you want this data, you have to be intentional about defining what to track and why. Focus on the behaviors that drive value: those linked to activation, retention, and loyalty.

3. Unstructured user-generated content

Here’s where the game has really changed. Think of the content your users create within your product — notes, tasks, messages, comments. Just as an example, in tools like Asana or Trello, a task might have a title, a description, and associated comments that reveal rich context about the user’s goals. Historically, this kind of data was hard to tap into. It’s messy, inconsistent, and requires a lot of manual parsing to make sense of.

But with advances in large language models (LLMs), this barrier is crumbling. LLMs can process and structure this data at scale, turning messy content into clear insights. They can summarise, categorise, and even extract user intent based on what’s written. This is where you can unlock a level of product intelligence that goes far beyond surface-level interactions.

Other types of data: Metadata and industry data

While these three are the core types, there are two more types of data worth mentioning. Metadata — like device type, time of use, location — can add useful context, especially for performance optimization. And industry data (external benchmarks, market trends) can complement your internal insights, helping you understand where your product fits in the broader landscape. Used wisely, these can round out your data strategy.

From data inputs to powerful outputs

Having data isn’t enough. The real advantage comes from how you use it to create a better product experience — one that’s smarter, more useful, and harder for competitors to copy. There are three big levers here:

1. Personalization

This is where data makes the user’s journey smoother. Based on what you know about them (inputted data) and what they actually do (behavioral data), you can recommend the next best action or suggest ways to improve their workflow. This can take many forms: suggesting the next best action, tailoring the interface to their habits, or offering recommendations on how to improve what they’re already doing.

Using Trello or Asana example again, it might suggest a template, pre-fill fields, or automate repetitive steps. These touches make the experience smoother and reduce friction — making your product more sticky.

2. Prediction

Prediction goes one step further. By analysing behavioral data at scale, you can identify patterns that signal what a user is likely to do next. Are they about to churn? Are they primed for an upsell? Are they stuck at a bottleneck?

With these insights, you can intervene proactively — offering help before frustration sets in, nudging users toward valuable features, or offering incentives at just the right moment. This kind of predictive capability not only improves the user experience but also drives key business outcomes like retention and expansion.

3. 10x better generative AI products

This is where the latest advances in AI come into play. While anyone can plug an off-the-shelf LLM into their product today, what makes your generative AI truly stand out is the context you can feed it.

A generic AI created by a new entrant will usually produce something too broad to be useful. But if your system already knows the user’s role, recent activity, preferred formats, and the specific project they’re working on, it can deliver a report that’s almost exactly what they need — without extra prompting from the user’s side.

Think of it like ordering at a restaurant. A user trying out a new AI product is like telling the waiter, “I’d like a meal, please.” You’ll get something, but it might not suit your preferences. A context-aware system is like the waiter knowing your dietary needs, favourite flavours, and portion size. In product terms, this means automating the context and providing it to the AI behind the scenes, creating a product experience that feels almost magical to the user.

The long game: Data flywheels and network effects

The magic of leveraging data isn’t just in the immediate improvements — it’s in the compounding advantage you build over time.

The data flywheel works like this: as your product collects more data, it learns and improves. Users then get more value, which encourages deeper engagement, generating even richer data. This creates a self-reinforcing loop that’s hard for competitors to catch up with. Competitors might copy features, but they won’t have the same depth of understanding about your users’ needs.

In collaborative or community-driven products, this advantage is amplified by network effects. As more users contribute data (structured or unstructured), the product becomes more valuable for everyone. Switching costs increase, not just because users are used to your interface, but because their data and workflows are embedded in your ecosystem — we’ll explore this a bit more in another post on Distribution.

Key takeaways

In a world where it’s easier than ever to spin up new products, data remains a key competitive advantage — if you know how to use it.

• Start with a clear understanding of the proprietary data you have — inputted, behavioral, and unstructured — and complement it with metadata and external datasets where useful.

• Use this data to deliver personalized experiences, predictive interventions, and context-rich generative AI to make your product frictionless.

• Remember that data compounds. The more value you deliver, the more data you collect, and the harder it becomes for competitors to catch up. Combine this with network effects where possible, and you’ll build not just a better product, but a more defensible one.

I had an interesting chat with a series A founder last month that got me thinking. Like many of us in tech, he was both excited and terrified about AI. "Every time I open Twitter, there's another AI tool that could basically replicate what we've spent years building," he told me. And honestly? I get it.

The pace of AI development is wild. What used to take months of design and coding can now be knocked out in days. It's both amazing and slightly terrifying — depending on which side of the disruption you're sitting on.

These days when I talk to PMs and tech leaders, I like to throw this question at them: "What happens when someone clones your entire product in 3 months? How do you win then?"

Usually there's an uncomfortable silence. Then some nervous laughter. Then the realization hits: in 2025, your product itself probably won't be enough of a competitive advantage. You need something else.

Having worked with both startups and tech giants, I've seen this pattern play out before. The winners aren't always the ones with the best product — they're the ones who build moats that others can't easily copy.

The moats could be many things, but in my view, the main ones are Data, Distribution, and Depth. I call this the 3D's Framework.

1. Data: The foundation of it all

Here's the thing about data that many PMs and tech leaders overlook — it's not just aggregated metrics in a spreadsheet, it's the story of every interaction your users have had with your product. If you've got a live product with real users, you're sitting on something incredibly valuable. You have insights about your users’ behaviors, goals, and pain points that can’t easily be bought or replicated by newcomers. You are in a better position to release value-adding features to your users.

This becomes even more powerful when we talk about AI. Everyone's rushing to bolt on AI features to their products (I get it, I've been there), but here's what separates the good from the great: context. Sure, generic AI can give you decent results — kind of like how a fortune cookie sometimes accidentally gives you relevant life advice. But when you feed AI years of real user behavior and specific use cases? That's when things get interesting.

Having proprietary data about your users enables you to build a far superior AI product compared to new entrants. The advancement of LLMs has also unlocked a type of data previously untapped: you can now make sense of that pile of unstructured data currently just sitting on your database. I wrote about it in more detail here.

2. Distribution: Leveraging your user base

Let’s face it: getting people to use your product has always been harder than building it. And guess what? AI hasn't changed this fundamental truth. If anything, it's made distribution even more crucial. Because when everyone can build decent products quickly, your ability to reach and retain users becomes your superpower.

Say you already have 1,000 active users in a niche — accountants, teachers, digital artists, whoever. When you launch an AI-powered tool tailored to their specific needs, you start with an immediate distribution advantage. Your competitors might come along with a technically superior tool, but they'll be stuck at the bottom of the hill, shouting into the void while you're already having conversations with your users.

Don’t let the saturation of AI tools deter you. You already have the advantage: existing relationships, user trust, and adoption pathways that competitors envy. Pair this with your proprietary data, and you’re already two steps ahead in the race.

3. Depth: Keep users coming back

Now, attracting users is just the start — keeping them loyal is the real battle. I'm not a branding guru, so I won’t preach about brand loyalty or community building —although both matter greatly. What I'm emphasising here is something more concrete: Depth.

Depth means you deeply understand and support your users’ entire workflow, not just one isolated task. Your AI-powered tool helps users accomplish task X brilliantly —great. But what comes next? Tasks Y and Z, probably. Ask them: Do they currently export results from your app only to move them into other tools? Why not close that gap yourself? Or at the very least, can you provide an integration with that other tool?

By extending your product deeper into your users’ workflow, you drastically increase the switching cost. Users stick around because you’ve not only solved their initial problem but also simplified their entire workflow. Each step you incorporate into your product strengthens your moat, and making your product harder to switch from.

Key takeaway: Your moat isn’t your tech anymore.

In 2025, competitors will clone, replicate, and probably improve your tech faster than ever. I've seen it happen repeatedly, and it's only getting faster. Your moat needs to be built on things that are harder to copy: proprietary Data that fuels unique features, superior Distribution through your existing user base, and strategic Depth that keeps users in your ecosystem.

Share The Shift by Debbie Widjaja