Nikhyl Singhal from Skip on Product Management in the AI Era

ELI5/TLDR

A veteran product exec (Google, Meta, Credit Karma) talks to Stanford CS students about what product management is actually becoming. Short version: the old PM job — sitting in meetings and moving information between teams — is dead. AI does that now. The new job is having strong opinions on what to build, being hands-on with the tools, and making good calls fast. Middle managers in their mid-30s who only know how to organize people are in deep trouble. New grads who live inside Claude Code and similar tools are in the best position the industry has seen in a while.

The Full Story

What product management actually was, before all this

Every tech company has people who build things and people who sell things. A product manager sits in between, gluing what to build to how to build it. That’s the whole job description. For the last decade, PMs mostly showed up after a founder found product-market fit — when the company needed to stop experimenting and start being consistent. PM as a role was quiet, process-oriented, and mostly about coordinating multiple teams that don’t talk to each other.

Nikhyl walks through four stages of a company’s S-curve: finding product-market fit (no PMs, just founders rubbing two sticks together), scaling after product-market fit (first wave of PMs), hypergrowth (armies of PMs to scale and expand simultaneously), and late-stage big tech (PMs fighting innovator’s dilemma, trying to do zero-to-one inside a giant). Each stage needs a different kind of PM, but the work was always a mix of building and what he calls “movement of information.”

The part that got killed

The movement-of-information part. That’s the AI roadkill everyone’s talking about.

Product in general was essentially a movement of information. No matter how big or small the company was, your job was to package information for some other decider.

He calls this bureaucratic layer the reason nobody liked their PM job three years ago, even when the pay was great and layoffs weren’t a thing. The work was packaging — survey results, sales call summaries, customer complaints — into a slide deck for a VP who packaged it again for a director who packaged it again. Theatrics, he calls it. All the way down until some IC4 at the bottom ran an experiment, and the person who actually knew what happened never made it into the room.

Agents do this now. A PM wakes up to a prioritized summary of every customer chat, every sales call, every survey, ranked by potential revenue or strategic fit. The thing that sounded like science fiction a year ago is just Tuesday. So the layer of people whose whole job was shuttling that information around — gone.

The part that got better

Everything else. Judgment, decisioning, talking to customers, hands-on building, picking what to make. Nikhyl says leaders in his network are having more fun than he’s ever seen.

There is more joy for the leaders and people that are in tech than ever before because they don’t have to depend on an engineer, on a designer, on a founder, on their boss in order to get something done. They can all build.

The old silos — PM, designer, engineer, data scientist — are merging. Designers who only pushed pixels are struggling; designers who have opinions about what to build are ascendant. Engineers who only write code will get squeezed by the models; engineers who have product opinions win. And PMs who only managed are obsolete; PMs who build are in demand. He calls this emerging role a “product builder.”

Two things that shouldn’t be true at the same time

Big tech is doing 30-70% layoffs this year. Salaries for the top 1% of product people have doubled in 18 months. Nikhyl says he’s personally helped negotiate four product leadership contracts north of $10M annual comp. Open PM roles are at an all-time high.

How does that square? Companies over-hired during the zero-interest-rate era. They added PMs whose job was organizing other people, not building. Those people are being cut. At the same time, companies can build so much more, so much faster, that they desperately need people with judgment about what to build. The pool is getting squeezed in the middle and expanded at both ends — more hands-on ICs, more senior builders, fewer pure managers.

Who’s actually screwed

Not the students in the room. The people in deepest trouble are middle managers 8-15 years into their careers who got promoted during the cheap-money years because they could “talk,” never developed strong building instincts, and now have kids, mortgages, and aging parents — no time to reinvent themselves. They see Claude Code and know intellectually they should learn it, but can’t carve out the hours. Those are the people who won’t find a next role.

Companies are a lot denser, meaning that there’s not 12 layers between yourself and say the CEO, and so there’s a lot more connection, there’s a lot more camaraderie, but the pace is extraordinary.

The interview has changed

Top employers — he names Anthropic and OpenAI — have stopped caring about brand pedigree. They’re not impressed that you worked at Google for six years. In fact, Google for six years might make you less relevant than a current student, because Google tenure means sitting in back-to-back meetings on an internal stack that works a certain way. What they check in an interview now is how modern you are — whether you live inside the tools, whether you have opinions, whether you’re hands-on. They can tell when someone is learning the tools during the interview itself.

The Meta metaverse detour

A student asks why Meta sank a decade into the metaverse with no apparent payoff. Nikhyl, who worked there, gives a careful answer. Mark’s view, as he reads it, was that Meta had never been the innovator of a computing platform — it leveraged mobile and the web but didn’t build them. To get to the next level, Meta needed to own the next platform. Mark believed it was the metaverse and, unlike Google with Hangouts, he refused to iterate his way there. He committed ten years of capital. It didn’t work. AI came in as the next platform instead. Meta is now making a strong play there.

Nikhyl’s point isn’t that Mark was wrong. His point is that big, unobvious bets can’t be made by consensus — they have to be founder-led. Apple runs this way. Google doesn’t. Google is reactive. Both models produce real companies. What startups have over incumbents is the ability to play in spaces that look too small to matter, until they don’t. Waymo looked like the metaverse for its first five years.

Careers as chapters, not periods

Most tech workers stay at a company 2-3 years. If you work for 50 years, that’s 15-18 jobs. He frames this as chapters in a book, not periods in a hockey game. His company Skip is built on the idea that the best career advice is about your next job, not your current one — how does chapter N set up chapter N+1. He says most adults he talks to have five or six of their jobs turn out suboptimal because they weren’t intentional. Skip is his attempt to run a talent agency for product people, the way Hollywood reps actors or CAA reps athletes.

What to actually do in college

He gets asked what a student should focus on. Three answers:

Be modern. “Brand is at an all-time low.” Being in this classroom with a Max Claude Code account probably makes you more useful to a hiring manager than someone who spent six years inside Google’s walled stack. Hiring is about how current you are, not where you’ve been.

Network. He regrets not being more social in college. Twenty years later, the 25 undergrad classmates he still talks to brought him more opportunity than any course he took. He specifically regrets not drinking — not because he wants to drink, but because it cut him out of social pools.

Systems mindset. In a world where you can express anything and have it built, the hard problem isn’t “how do I construct this?” It’s “should I construct this, does it fit, how will I know if it’s working?” That’s engineering abstraction — understanding how platforms layer on each other, where the stack is going next. Assembly became compiled languages became scripting languages became prompted languages. Every few years the stack gets smarter. If you understand how platforms evolve, you know where to stand.

On meetings, theatrics, and the gig being up

He describes the old rhythm: enter the industry, and within three years you’re in back-to-back Zoom calls all day. VPs working late to polish decks that get handed up to other VPs who polish them further, until the actual person who ran the experiment is nowhere in the room. He sees a strong trend toward no-meetings-at-all cultures — a half-day of gathering per week, and AI doing the context-setting the rest of the time.

I think the gig’s up.

Coaching, communities, and ChatGPT

He’s direct: most paid coaching is run by people who weren’t particularly successful as operators and find coaching a more pleasant way to pay bills. ChatGPT will eat that industry. Most online product communities are the same — designed to scale and monetize, which means optimizing for people learning the craft rather than practicing it. None of the 125 executives in his private group participate in public communities.

When to leave a job

Simple rule: you want the environment around you to be growing slightly faster than you are. If you’re the one pulling everyone else forward, your own learning stalls. Mike adds a corollary at the end — when the job gets comfortable, leave. Comfort is the signal that growth has stopped.

Key Takeaways

Product management splits into “product” (opinions, judgment, building) and “manager” (moving information). AI kills the manager half. The product half gets paid more than ever.
Open PM roles are at an all-time high right now. Top-1% product salaries have doubled in 18 months. Four product leadership contracts in his network crossed $10M annual comp.
Big tech is running 30-70% layoffs this calendar year and hiring aggressively at the same time. The cut is mostly middle managers promoted during the ZIRP era for coordination skills.
The most dangerous career position is middle manager 8-15 years in, with a young family, who only knows how to organize others. No time to reinvent, skills getting obsoleted.
Company brand pedigree matters less than “how modern you are” — whether you live inside the AI tools, whether you have opinions, whether you’re hands-on.
Career arithmetic: 50 working years ÷ 2-3 years per job = 15-18 chapters. Optimize each chapter to set up the next one, not to be the best version of itself.
S-curve stages of a company: product-market fit (no PMs), scaling (first PMs), hypergrowth (armies of PMs), late-stage innovator’s dilemma. Each stage needs different people.
Google shipped Chrome every 6 weeks while Firefox shipped quarterly and IE shipped yearly. Speed of iteration was the whole game. Android vs. iOS was the same story. Start ugly, iterate fast.
A forward-deployed engineer is basically a rebranded professional services role. With AI, the function of extracting customer insight can increasingly happen without sending a human into the field.
The three roles — designer, engineer, PM — are merging into a hybrid “product builder.” Pure pixel-pushing designers and pure code-writing engineers without product opinions are being squeezed.
Big, unobvious innovation can’t be done by consensus. Meta’s metaverse bet is a founder-led ten-year commitment. Google, by contrast, is reactive — they tested Hangouts, it didn’t work, they moved on.
At scale, a new $1B business is rounding error. Meta making another billion was “a four-line change to a ranking algorithm.” This is why big companies need moonshots (Waymo, metaverse) and startups get to play in spaces that look too small to matter.
Agents that summarize every customer call, sales meeting, and survey in prioritized order — with revenue impact and implementation complexity already scored — are now standard tooling for product leaders.
“Theatrics” — slide decks flowing up through layers of VPs — is the form of work AI is most obviously replacing.
The three things that matter in college for a tech career: be modern (live in the tools), network (the 25 people you stay in touch with), systems mindset (understand how platforms layer and evolve).
When to leave a job: when the company stops growing faster than you, or when the work gets comfortable.

Claude’s Take

This is one of the better reads on where PM is heading, mainly because Nikhyl isn’t selling a narrative — he’s reporting from inside a network of 125 heads of product at the companies that actually matter. His core move is separating the word “product” from the word “manager” and showing that AI is coming for one of them, not both. That distinction clears up the apparent contradiction everyone keeps tripping over: how can PMs be getting laid off and PM hiring be at an all-time high? Because two different jobs have been living under one title.

The strongest prediction, and the one that will actually be testable, is his claim about the middle manager. Eight-to-fifteen-year tenured managers who got promoted for coordination skills during cheap money, who now have families and time-starved lives — he thinks they’re in terminal trouble. That’s a clean, falsifiable bet. Score it in two years.

The weakest part is the Skip pitch. He’s careful not to hard-sell, but the framing of “talent agency for product people” is clearly in its infancy and may or may not become a thing. Ignore that part. The strong parts of the talk are the structural ones: the S-curve, the three skills for students, the “grow slightly slower than the environment” rule, the observation about brand collapse. His swipe at coaching as an industry is bold and probably correct.

Score: 8/10. Not quite a 9 because some of this is now conventional wisdom in tech circles — the death of the information-mover PM, the merging of roles. But the specificity (comp numbers, company examples, the Meta/Google contrast, the three-skill framework) lifts it well above the usual take.