Confessions of a Silicon Valley PM: AI Lab Secrets, Rest-and-Vest Hangovers, and the Myth of the Mentor

A record of a conversation I had with a Silicon Valley product manager. She's worked at a company like Meta and later moved to startups, and is about to join a Series A B2B SaaS company. I've cleaned up the original loose transcript and organized it by topic, stripping out filler, but kept it as the back-and-forth it actually was. Names and companies are removed. Me is me; PM is her.

TL;DR

• Decision framework: lay every option out as a table, mark each dimension (revenue, growth, experience) red/yellow/green, and make the trade-offs visible — communication shifts from a personal argument to an objective resource-allocation problem.
• Side project validation: if you're your own ICP, AI has pushed the cost of a POC toward zero — put a rough-but-working thing in front of people instead of pitching slides.
• Inside the AI labs: DeepMind's core decisions sit in London; Anthropic's culture-fit bar is extremely high; OpenAI pays absurdly well but runs "Silicon Valley's high-paid 996," and everyone's betting on the token appreciating; Meta gets slowed down by compliance and PR review.
• Meta's rest-and-vest culture: during the pandemic, coasting while your stock vested was normal — now the company runs a hard PIP process that forces everyone to perform.
• Finding a mentor is the hardest part: your manager isn't your mentor, the truly great people don't have time for you, and paid coaches are a quality gamble — what actually works is peer mentorship, trading war stories with people at the same stage as you.

The art of PM decisions: the traffic-light framework

Me: You gave a talk once on how PMs should make decisions between options. Can you walk me through it?

PM: In practice, say you've laid out options A, B, and C — most people can tell the pros and cons by gut feel. But in meetings you'll constantly get challenged: "I get why you'd pick A, but why not B?" If the PM is also just going on instinct, cross-functional stakeholder management gets hard fast — especially with someone aggressive, where the meeting turns tense.

Me: How do you handle a stakeholder who comes in hot like that?

PM: I try to pull the conversation back to something objective. If I know beforehand someone's going to blow up over a particular option, I line up allies first — I'll talk to the designer or eng lead who supports my direction before the meeting even starts. There's also a technique I really recommend: a VP of Product at Meta came up with a "traffic-light framework" borrowed from how engineers do system design. For each option, you list a few key dimensions — revenue, growth, user experience — and mark each one green, yellow, or red. Then you lay the table out in front of everyone. It's basically applying software system design to product management: you make the trade-offs visible, then ask the room, "here's where things stand — whoever makes the call owns the risk — so how do we want to go?" The conversation becomes an objective resource-allocation problem instead of a fight over opinions.

Validating a side project: interviews first, or ship a POC?

Me: How's the side project going?

PM: A few friends and I are building a tool around product lifecycle and roadmapping. We've hit a disagreement recently: AI is moving so fast, and YC's usual doctrine says you should interview your ICP and validate demand before you write a line of code. If we walked into a room with YC people right now with zero product, they'd tell us we haven't even done basic market validation.

Me: So are you just going to build the POC first?

PM: Exactly. The pain point we're solving is our own — we're our own ICP. Since we already have the need, and build cost is basically zero now, we decided to ship the POC first and put a prototype in front of people. With AI tools you can write two hundred scrappy lines of code and something real comes out — you don't need to spend money shipping to an app store, you don't need polished UI. A simple webpage that proves the value is enough. Handing people something rough but working and asking if they'd use it beats interviewing them with slides.

Inside the AI giants: DeepMind, Anthropic, OpenAI, Meta

Me: Are you still looking at roles at the frontier AI labs?

PM: I am, but the culture at each one is wildly different.

Start with DeepMind. I interviewed for a PM role there and ran into a real dealbreaker: the core power and decision-making sits entirely in London. If you're based in California, the projects you can actually touch are extremely limited. The interviewer told me flat out — if I wanted to work on anything that actually mattered, the only way was to move to London, otherwise they wouldn't recommend joining.

Me: What about Anthropic?

PM: A great company, but the bar is brutally high and they're extremely selective about who they hire. Internally the culture is shaped a lot by effective altruism, and the culture-fit standard is strict. Some genuinely excellent friends of mine applied and got cut.

Me: And OpenAI — everyone talks about it, huge pay but also huge pressure?

PM: It's genuinely absurd. Comp packages routinely start at $900K to $1M. But people I know there work twelve to fourteen hours a day, basically every day — it's Silicon Valley's high-paid version of 996. Everyone who goes there knows exactly what it is: a meat grinder, and you're burning your life to be in it. But people stick it out because they're all betting the company's internal equity token doubles or more — everyone's chasing financial freedom a few years out. It's not a pressure most people can actually sustain.

Me: And Meta's AI org?

PM: Meta's problem is that it's a legacy big-tech company facing far more regulatory and public scrutiny. They're extremely focused on "AI Safety" — which in practice means compliance and PR risk. If Meta's AI says one wrong thing, it's a headline the next day. Internal AI projects constantly get stuck behind legal and safety review. For an engineer or PM who wants to move fast, that layer of red tape is exhausting.

Meta's "rest and vest" culture and the layoff wave

Me: A lot of your friends are still at Meta — what's the mood like now? I've heard a few years back a lot of people were basically coasting.

PM: During the pandemic, that was true — it was the golden age of resting and vesting. People I know at Meta were working maybe one or two hours a day and spending the rest of the time on their own stuff, just waiting for stock to vest. But the mood has completely flipped. Leadership caught on and started enforcing performance strictly, rolling out a genuinely hard PIP process. If you're still trying to coast, your review drops fast and then you're managed out. That's a big part of why so many people inside big tech are anxious right now.

Me: Is that part of why you left big tech?

PM: Somewhat. What I found is that the pay at a big company is good, but the sense of accomplishment is low. A lot of projects are handed down top-down — you're just the executor with little say, and that burns people out. For me, I need to see that my own decisions produce real impact. That's ultimately why I moved to a startup.

What's next, and the pain of finding a mentor

Me: What's your plan for what's next?

PM: I'm joining a Series A B2B SaaS startup. My goal is clear: I want to personally live through a company growing from $8M ARR to $100M ARR, and understand what breaks — for both the product and the team — at that speed, and how to fix it.

Me: That sounds like a huge challenge. Do you have a mentor or coach guiding you through it?

PM: Honestly, that's the biggest pain point I have right now. Finding a genuinely useful mentor in Silicon Valley is really hard. Your manager usually isn't your mentor — that relationship is more about reporting lines and incentives. And the people who are truly excellent, who've made it to the top, have extremely limited time — they don't have room to coach you one-on-one.

Me: What about paying for a professional life coach or career coach?

PM: I've tried that, and I'm willing to pay — a few hundred dollars an hour. But the problem is quality is a gamble. A lot of coaches simply don't have your depth of experience, or work in a different domain, and the advice ends up surface-level. It's hard to find someone whose cadence fits, who has real operating experience, and who's actually willing to invest in you specifically. What's actually worked best for me is peer mentorship. I have a group of friends who are PMs or engineers at different startups, and we get together over coffee or drinks and trade notes on whatever we just got burned by and how we solved it. That kind of real-world exchange has been far more useful than chasing an unreachable mentor.

Me: What about an MBA? Some people use that stage to build credentials and a network.

PM: A friend of mine did an MBA at Wharton, and his complaint was that if your goal is founding a company or joining an early-stage startup, an MBA doesn't help much. At a top business school, most people are aiming for consulting or investment banking — safe, high-paying tracks. In that environment, if you say you're going to join a high-risk AI startup, people think you're strange. If what you actually want is operating skill and a startup network, you're better off skipping two years and a few hundred thousand dollars on an MBA and just going straight into the market to fight the real fight.