Dr. Productlove, or: How I Learned to Stop Worrying and Love the AI

Picture this (and if you are a PM in 2026, you probably don't have to picture very hard): you predict a bug in writing, 3 weeks before it ships. (Supplier records, of all things. Boring, I know.) Archive a supplier, and it vanishes forever, no restore, which means somebody creates five suppliers called "test," and the whole company gets to live with them for the rest of eternity. You named the failure mode. You named the fix. You put it in a document and handed it over like a responsible adult.

It ships anyway.

I have lived some version of this more times than I care to admit, and so has nearly every PM who has come through my cohorts in the last two years. The sheer impotent rage that follows is real (believe me), and it is also a complete dead end, because it's the last artifact of a job that quietly stopped existing while we were all busy writing longer PRDs.

So what is the job now? How does a product person actually add value when the machine builds faster than any of us can read? That is what this piece is about, and after 16 years of shipping AI products, here is my humble answer (at least at this point on the space-time continuum).

Quality was always a byproduct

We told ourselves software got good because we cared. That was a collective hallucination.

Software got good because changing it was expensive. When a feature cost three engineers six weeks, everybody thought first — spec, review, design, the whole ritual — and every one of those rituals was a hedge against the cost of being wrong later. We called it craft. It was mostly economics. (Alan Cooper caught the inmates running the asylum back in 1998, but his cure assumed the inmates were slow, and they are definitely not slow anymore.)

Generation is now close to free, so the hedge costs more than the mistake, and rational teams stop hedging. Cheap bugs get fixed, expensive bugs stay, and the feature is marked shipped either way. Do that across a hundred features and you get a permanent sediment layer of things that technically work and are quietly bad. Every fast product has this layer. Yours does. Mine most definitely does.

The old PM job was preventing that layer from existing. That job is over. Here is what replaced it.

1. Aim the rocket

AI is a rocket, and it is blasting away regardless of what any of us write in a document. If you are lucky, and senior, and actually proved useful, you get to aim it. You do not get to control it, and you most certainly do not get to choose what color it's painted.

Yes, this is a demotion, and you should feel it in your bones: you are trading control over craft for influence over direction. It's a worse situation in every way we collectively use the UX/PM job as an identity, and a much better one in every way that actually moves a company. Your value is in deciding where the thing lands — which customer, which problem, which bet — and not in arguing about the shade of blue the rocket is as it's streaking past.

2. Ration your f*#$s

As Mark Manson so eloquently said in The Subtle Art of Not Giving a Fck: A Counterintuitive Approach to Living a Good Life:

Most of us were trained to maximize caring about quality, and we have zero vocabulary for rationing it.

Time to learn.

That supplier admin screen you are getting yourself enraged about? (And trust me, YOU are absolutely the one doing the enraging. It's an inside job. I know.) Somebody configures it once at onboarding and never opens it again, and the correct amount of bespoke f*#$s you should give about it is approximately zero.

Conversely, you should give all your care to the first ten minutes of your product: pages like search and item detail, which are used 80% of the time; mission-critical dashboards; strategic data visualizations; the Sales demo; and the AI surface, where a buyer stakes their internal reputation on you. Those places should never get a sediment layer. That's where you spend all the f*#$ you have and then some.

Knowing the difference between the two is that magical intuitive space between "taste" and metrics -- it's where AI (for now) holds limited sway, and where true professionals stand out from amateurs. It's where you earn your keep.

3. Polish the patterns, not the turds

Okay, but that supplier bug still stinks. Here's the fix, and it's the part I suggest you should care about.

Archive with no restore isn't a one-off mistake. It's what you get when every team invents the archive action from scratch. Somewhere else in that same product there is a different delete, a third kind of table, a search that sorts its own special way, and an audit log that covers half the objects. Each one built fresh, independently, using AI, and each one is wrong in its own unique and creative way. That is how ensh*ttification actually spreads: the same undifferentiated thing, built a hundred slightly different sh*tty ways.

So stop polishing one turd at a time, and start polishing the patterns. Pick the boring things every product needs and make each one a standard, built once and built right:

• CRUD with soft delete, archive and restore

• Tables with sort, filter, bulk actions, and real empty states

• Search that puts recent and exact matches first (and specify the sort, always, please, I beg of you, for the love of gods above and below!!)

• Audit trails on every object that matters

• Undo and rollback for anything destructive

None of these are unimportant (get rollback wrong and you will be on the news). They are undifferentiated — critical, and exactly the same everywhere, which is precisely why they should never be custom.

And AI makes this urgent, because AI writes code by copying what it sees. Give it one way to build a table, and it builds that table every time. Give it five ways, and it will cheerfully invent a sixth. Patterns flip that around: the AI codes faster against a known shape, tests get written once, a bug gets fixed in one place, and your customers learn the right UI once. You can't out-specify a system that generates faster than you can read, one feature at a time — but you can specify once and let the pattern apply a hundred times.

Two warnings from the trenches. One bad pattern breaks everywhere at once, so the patterns are where your craft goes now. And a pattern that lives in a wiki is dead in a week, so put it where the builders, human and AI, actually look: scaffolds, component libraries, shared tests, and the instructions your coding agents follow.

The only way to get in front of ensh*ttification is to stop propagating it in the first place.

4. Defend one gate

If I keep a single piece of old-world rigor, it's this one. Showing a customer one version and asking "do you want this, or nothing?" is not a test. It returns "yes" every single time, no matter how bad the thing is, because people are grateful for anything from a team that usually ships nothing (and gratitude, sadly, is not a signal).

Two versions. This way or that way. That is the whole gate, and generating the second version now takes a few minutes and some small amount of caring. In the old world it was expensive, so some people sometimes skipped it. (It was some people, not you or me, ever, right?) In the new AI normal, qualitative testing with multiple versions is cheap, but some people skip it anyway, out of pure habit. Don't be some people.

5. Detached Engagement

Twenty years of pattern recognition means you see the mistake before it exists, and seeing it first makes you feel like you own it. So when it ships, it doesn't land as a minor defect in a rarely used screen. It lands as subtly as a sledgehammer to the teeth and a final referendum on your judgment before everything fades to black.

Folks... None of that is real.

Nobody else is scoring that game; you are the only player, and you are losing only to yourself. (But please don't conclude that you should stop doing the analysis — the analysis is why you knew, and it's what makes you worth having in the room.)

Practice "Detached Engagement": Predict it, write it down, tell your team the best way you know how, and release the outcome like Arjuna releasing the arrow the moment it leaves his bow. You have complete control over your actions, your preparation, and your intent. The final result is shaped by countless external forces, time, and cosmic order. It belongs to the Universe, not to you.

6. Mind the vertigo

You spend six weeks designing for a product persona while the real people in that role are being laid off by the tens of thousands, sometimes by your own product! I don't have a comfortable resolution for this, and I distrust anyone who offers one quickly. So file the bug, fix the pattern, and stop grieving about the screen an admin opens once a year, if ever.

That is, if that admin still has a job a week from now.

Which brings me to the only part of all this I am actually sure about: the humans.

The Humanity! What I personally do and recommend

1. Play and experiment.

(There, I said it. Happy? Now go create some cool sh*t.)

2. Write and teach.

In a world increasingly filled with AI slop, produce original content that leans into humanity and reflects your values (such as this humble missive you're currently reading... You are still reading this? Don't you have some AI cat videos to watch?). Then teach — help others learn what you figured out. (Our UX for AI Professional Certification is one such program. That is my own humble contribution — Daria Siegel Kempka, and I will be publishing news about the next cohort soon; meanwhile, get on the waitlist. Our first cohort filled up in 11 days.)

3. Connect and learn.

Connect with like-minded people who will help you learn and sharpen your craft. Last night I attended an awesome AI get-together at the Okta headquarters in SF hosted by Enrich, where I ran into Vivek Chopra, who is planning his next AI book and who just happens to be a 2-time attendee of my SXSW workshop — truly a sucker for punishment, but it clearly takes one to know one -- because I will be making my 6th! appearance at SXSW, again with Daria Siegel Kempka, teaching Designing Multi-Agent Systems next year at SXSW 2027. (But seriously, Vivek is amazing, and you should follow him for all the no-nonsense, down-to-earth technical explanations of the latest technology in AI. I'm selfishly hoping he'll allow me to write a small perspective for his upcoming book.) Human connection is increasingly providing the anchor we all need in this increasingly busy world rocked by winds of AI-driven change and marked by ever-increasing busyness, anxiety, FOMO, and confusion.

(Now, I realize not everyone lives in the Bay Area and can casually throw on their white fashion sneakers and saunter down to an AI soirée with light snacks at the Okta SF downtown headquarters. My suggestion is to organize: make your own chapter of AI humanists where you live in the world and act as the organizer and the catalyst of your own. Silicon Valley can be a very exciting place, but unbound technology should not be writing the fate of humanity in the age of AI. Everyone needs to get actively involved. Like right the f*#$ now. Or we will all get the fate we deserve.)

4. Surf the ocean of change.

Change is no longer something that happens once in a while: it is a continuous massive rolling tsunami that is turning our world topsy-turvy as we are living in it. In order to stay afloat, it is no longer an option to remain a rock and not to roll — the only thing we can do is to surf the wave of change, continuously learning, inventing, adapting, and steering all at the same time. All we have today is each other, and our humble surfboards: the small chunk of not-really-seaworthy white triangle-thingy made of solidified core principles and our humanity we do actually get to control, as we alternatively glide and push to stay above water as the wave is gathering speed.

And speaking of change!

After an incredible run at Kindo, I am exploring full-time opportunities in AI Product leadership, targeting the companies and teams working on AI control plane and agent management. If that's you or someone you know, send me an email — I promise to read it this time.