What the Past Year Taught Meta

This podcast is a live conversation I hosted for Skip Coach, a community of senior product and tech leaders navigating career transitions together. Members get access to live sessions like this one, including real-time Q&A, before the content is shared publicly. If you’re a product or tech leader looking to maximize your career, apply to join Skip Coach — it’s free — to access this content and other exclusive resources.

Listen on YouTube, Spotify, and Apple Podcasts

Brought to you by:

Jira—Do the right work with Jira

---

When two billion people use your product every day, changing it is like a game of Jenga. Touch one piece and something else moves. Every change was measured dozens of times before it shipped. And PMs were the gatekeepers for nearly every one of those changes, which bred a quietly antagonistic culture: the PM fighting to serve their millions of users, against the PM chartered to protect the experience of the billions. That was true when I worked at Meta, and it shaped everything about how the place built software.

Our June conversation with Jagjit Chawla, the VP behind Facebook’s Feed and Reels, became one of the biggest episodes we’ve published, and this second visit was the chance to see how things have landed since. What surprised me, catching up with him, is how much of that equation is moving. The first wave was everyone building their own stack: personal agents, custom tooling, maxed tokens. That energy has since shifted into company-wide services available to everyone, and the focus has moved to true, lasting transformation.

Four changes stood out:

• A smaller, flatter org is building more, not less. Less process, and more time spent on the product.

• The senior IC track got real. A credible path to the top without managing anyone.

• The expert functions aren’t dying. Data science and design shed the routine work and kept the precision work.

• Facebook is experimenting faster and bigger than before. Purpose-built apps for sellers, group admins, and creators — ideas that had been failing the prioritization bar for a decade are now getting the green light.

Below, each one in depth. The full episode has dozens more.

---

Doing More With Less

Meta is a smaller, flatter company than it was a year ago. That part has been well covered. What almost nobody talks about is what the organization actually looks like a few months later.

The most concrete change is fewer PMs and fewer layers. On Jagjit’s own team, no IC is more than one layer away from him now — you’re the VP’s direct, or a direct’s report. Two years ago, that wasn’t close to true.

But the result shows up in the culture. Less process and layers has meant more time spent building — the coordination tax that made big-company product work feel like arm-wrestling is much of what came out.

Where the org became smaller was not just in the junior ranks. And they continue to invest in talent - the RPM program, Meta’s rotational entry path into product, is hiring at full speed with a new class arriving in September, learning the new way of building from day one.

And Jagjit doesn’t treat the smaller team as an end state. His view is that the productivity wins eventually buy headcount back:

“There will come a point where everyone in the industry will be hiring hand over fist because we eventually now have unlocked our productivity gain. […] Our ambition has been up-leveled because we can now build a lot more.”

A Golden Time to Be a Senior IC

You no longer have to be a manager to grow in the PM ladder. In fact, he is seeing many managers converting back to becoming IC builders. The test Jagjit applies is a sharp one: can this person operate as an IC at the next level, not the level they’re on? If yes, they trade the bench of PMs for a pod of engineers and designers and start building again. The result is a wave of very senior ICs — and, for the first time, a well-trodden path above them. Most companies support ICs up to their director level — at Meta, that’s Level 8, or IC8. What’s new is a credible path beyond it, to IC9 and IC10. And the IC10s are real. Jagjit described one: a PM of fifteen-plus years, at VP level, whose individual-contributor project is allocating tons of compute across the company — how much capacity goes to frontier model training versus inference, to Facebook versus Instagram. That PM, in his words, “runs the building.”

For years, product people have watched engineers and designers climb real senior-IC career ladders while product forced its best people into management. That asymmetry is closing, and the timing isn’t a coincidence: when leaders can read ground truth directly and pods ship without layers of coordination, the scarce thing is a person with strong judgment who can own an enormous problem alone. And with fewer manager seats, the ones that remain go to the best coaches — hands-on executives who can scale other people, not just projects. If coaching isn’t the job you want, staying hands-on is no longer a career ceiling.

Design and Data Science Are Upleveling, Not Shrinking

In the spring, it looked like the PM was swallowing the sister functions — running their own analysis, generating their own designs, shifting the need for these teams. What actually settled at Meta is a division of labor: the PM owns the first mile of a problem, and the expert owns the last. A PM querying an analytics agent to build intuition is a good outcome, and it stays. But precise experiment analysis at a scale where “we measure everything to the third place of decimal” still needs a data scientist who can spot a polluted experiment, a bad holdout, or a metric drop caused by something else entirely. Design is the same: a PM can generate a system-compliant mock, but making Facebook’s video player distinguishable from TikTok’s with the logos hidden is craft only a designer delivers.

The observation underneath is the one worth keeping. The same shift PMs went through is now happening to every function simultaneously. DS and design used to be mired in the basics — pulling routine queries, fixing padding on mocks — and AI absorbed that work. What each function gets asked now are the tougher, more strategic questions, the ones that require judgment. That’s why they’re producing more, not shrinking: a data scientist who completed three deep analyses a week now does five.

The pullback reached even PMs shipping production code: “I would say at least I have changed my mind on that.” Engineers ship to production; PMs illustrate their ideas through code, then hand to the right engineer. First mile, last mile — engineering leveled up, just like the other functions.

One quieter update from June: the energy people were pouring into building their own stacks has shifted into company-wide services available to everyone — even those who never started down that path. The personal assistants Jagjit and others were hand-building — the nightly briefs, the inbox triage — are now available as a set of corporate-standard tools anyone can switch on. As he put it, “your real work starts in the morning after that efficiency is baked in.”

Three New Apps in One Summer

The ideas weren’t new. Jagjit has been at Meta six years, and every few months someone would pitch a dedicated app for Facebook’s power users — the sellers, the creators, the group admins and their most active members. Tens of millions of people each. “It never passes the prioritization bar,” he told me. In an app serving two billion, a feature for tens of millions adds bloat and risk, and pulling ten engineers off the main app always lost the opportunity-cost argument.

What changed is the cost — of building, and just as important, of maintaining. His team pointed AI agents at Meta’s existing codebase and extracted a reusable scaffolding: Facebook login, notifications, the performance work, the ranking infrastructure, all inherited. A small pod built Forum, the groups app, in weeks, with no large dedicated team behind it and no new organization needed to keep it running. And the framework is a cookie cutter. Once Forum worked, Seller and Creator Studio came off the same line:

“The beauty of some of these tools now is that once we build, let’s say, the Android app, we can point to an AI agent to say, ‘Hey, this is fully built out. Can you give me a scaffolding and start the iOS app?’ So it’s like eighty percent done without a lot of additional effort.”

All three shipped this summer — Forum in May, Seller in July, Creator Studio in August. None of them are sure things. They're bets, and the point is that they're now cheap enough to run several at once, knowing some won't work. The cannibalization worry turned out to have a data answer: give these power users a surface of their own and they produce more, and that content feeds both apps. And distribution, the thing that kills every indie app, is the one thing Meta has in abundance — point a promo at two billion dailies, the way it grew Threads and Edits.

The bigger change may be the cultural one. The old way to ship anything for these users was influence: negotiate with the Feed team for a sliver of their surface, compromise your way through every stakeholder, then run the holdout math to prove your feature earned its keep. Now:

“I have my own app, my own canvas to paint. I can try different formats that are not constrained by what the overall Facebook feed format would constrain you, and I can go much faster.”

A sandbox, a clear set of users, and experiments you can simply run. The Jenga tower still stands — but now you own a smaller tower beside it, connected to the same ecosystem. Some of what you build there will move into the main app; some will stay standalone for good. Either way, it feeds the whole.

Evals Are a PM Skill, Not an AI Skill

Every future-of-PM conversation now ends at the same word: evals. For most product people the term is unclear, and the skill is genuinely hard to build. So I asked Jagjit to explain it in plain speak — and to say why it matters so much. The core problem: AI systems are non-deterministic. The same input produces different outputs, so the old way of judging software (”does it do the thing?”) stops working. An eval is how you judge a system like that anyway: pick the dimensions that define good, rate real examples against them, and turn a fuzzy, subjective call into a number a team can be held to.

For the real-world version, Jagjit walked through how his team uses evals to keep Reels showing people the right videos. Relevance is the biggest lever the product has — show people irrelevant videos and “eventually we’ll have nothing left to run.” A small percentage of users rate what they see against the dimensions that define a good video: does it match your interest, is it timely, is it fresh, was it worth your time. The ML team is then goaled on growing the five-star answers and shrinking the one-stars — read over months, not experiment by experiment, because a thousand experiments are running on any given day.

So how does someone build the skill — and how does he screen for it? His interview question is deliberately subjective:

“If I was to say, write an eval for how would you evaluate bath towels […] you can come up with dimensions of how quickly does this thing dry, how does this feel on my face, is it cotton or nylon, does it have any chemicals […] Given a set of hundred bath towels, you could rate them on all these five dimensions, and that becomes your eval set.”

That subjectivity is the point: it shows him whether a person can build a system around something that’s hard to measure — take a thing they’ve never quantified and break it into dimensions worth measuring. It’s a hard skill to build. It may also be the one that shapes the next generation of PMs: the people who can distill subjective judgment into a quantitative set of measures.

---

The State of AI and Product Management

Stepping back from Meta specifically, here’s the state of things as Jagjit tells it, a year into the transformation:

• What’s unlocking: the ideas that never passed the bar. Build costs dropped, so opportunities that used to fail the engineering-cost threshold deserve a second look. But cheaper building doesn’t mean build everything — the judgment call is which of those ideas truly move the top line. Meta’s standalone apps are the model: a handful chosen carefully, each adding to the ecosystem.

• What’s shifting: the shape of teams and careers. Flatter orgs, managers converting back to builders, a senior-IC path that finally reaches the top, and expert functions doing more expert work rather than disappearing. The spring’s role-collapse predictions mostly didn’t happen — instead, every role moved toward its judgment and away from its process.

• Where to pay attention: evals. The one genuinely new skill in this report, and a hard one to build. Judging non-deterministic systems is becoming the shared craft of product work, and the PMs who master it will define the next generation.

The thread under all of it: the first wave of AI made everyone a builder. The second is deciding what’s worth building — and the scarce asset is judgment, not code. That’s why Jagjit can look at a smaller, flatter Meta and say the glory days of being a PM in big tech are still ahead. Fewer people are doing more product work, and the work that’s left is the good part.

---

Have your own career question? Get personalized guidance at Nikhyl.AI. It’s where the questions keep coming, and where I’ll keep sharing what I’m learning.