Product lessons from Facebook, Instagram, Uber, and OpenAI with Peter Deng

Executive overview

Peter Deng spent two decades building some of the most-used products in history — Facebook Newsfeed, the standalone Messenger app, Uber Reserve, ChatGPT Enterprise — yet has never spoken publicly until now. The core tension he returns to repeatedly: product craft matters enormously, but knowing which details don't matter is equally important.

He offers a set of durable frameworks for scaling products, hiring the right people, and leading teams — distilled from going through hyper-growth many times across very different companies.

Great product leaders obsess over the right details and have the wisdom to ignore the rest.

When the product doesn't matter

• At Uber, the price and ETA were the product — not the pixels on screen.
• Fixing UI bugs has far less impact than fixing something users fundamentally care about.
• Many of the most valuable tech companies weren't built on technological breakthroughs — they applied existing tech (GPS, databases, cameras) with superior execution.
• For AI startups, two things create defensibility: proprietary data flywheels and a deeply crafted workflow that fits into how people actually work.
• Product craft can overcome distribution advantages — Cursor, Windsurf, and Lovable broke through despite Microsoft Copilot's head start.

Going from one to a hundred

• Zero-to-one is finding product-market fit. One-to-100 is building systems that let you scale sustainably.
• You have to plan chess moves in advance — sometimes you need to go slow to go fast.
• The Facebook Newsfeed architecture has barely changed in 12 years because the team thought carefully about the full sharing loop before building.
• At Uber, re-architecting pickups and drop-offs (venues, airports, international edge cases) was unglamorous but enabled global scale.
• Build a growth team early — not primarily to drive growth, but because it forces rigor: proper logging, data analysis, and hypothesis-driven experimentation across the whole product.
• Maintain a portfolio approach: allocate resources across core product, growth, and craft simultaneously, adjusting the ratio as you scale.
• Keep a healthy tension between growth-focused and craft-focused people on the team — both are necessary, neither should dominate alone.

The five PM archetypes

1. Consumer PM — half designer, half product. Obsesses over details, craft, and delight. Driven by vibe and feel.
2. Growth PM — half data scientist, half product. Starts with numbers, runs tests, skeptical until proven.
3. Business/GM PM — half MBA, half product. Thinks in incentives, margins, and marketplace dynamics.
4. Platform PM — wired to build tools and systems for other builders. Often overlooked but critical for scale.
5. Research/AI PM — half researcher, half engineer, half product. Combines deep technical understanding with product taste to shape model behavior.

Everyone has a primary and secondary archetype. Knowing this helps you hire for what the team is missing, not for a generic "PM" shape.

Hiring and building teams

• Think of your team as a product. Identify what the company needs and hire people who spike at each distinct thing.
• Build a team of Avengers with different superpowers — the leader's job is to adjudicate tension, not eliminate it.
• Hiring razor: "In six months, if I'm telling you what to do, I hired the wrong person." This sets a high bar in selection, communicates expectations on day one, and reframes every 1-on-1 as calibration toward autonomy.
• The final interview question: describe the biggest mistake you've made and how it changed the way you work. Genuine vulnerability and reflection signals growth mindset.
• Growth mindset is the single most important attribute. Without it, feedback can't land, skills can't develop, and the org can't improve.
• Joanne Jang at OpenAI is an example: spotting a rare combination of deep technical skill and product taste, then asking her to write down what made her role unique, led to the creation of the model designer function — a key reason for ChatGPT's distinct feel and vibe.

Management and operating principles

• Say you're going to do the thing. Say you're doing the thing. Say you did the thing. Repetition isn't noise — it keeps goals visible, enables course-correction, and ensures work gets credited.
• Fit is a two-way street: find the role and company where your natural strengths are the thing that's needed, not a liability.
• Language shapes thought — choose words for docs, decks, and OKRs deliberately. The wrong word has downstream effects on how teams interpret and execute.
• User research is irreplaceable. Reading a summary is not the same as being in the room. You can only empathize with the full color — intonation, context, emotion — by being present.
• The IDEO design thinking framework still holds: empathize, define, ideate, prototype, test. The first two steps are where most teams cut corners.

On AI, education, and what's changing

• AGI is necessary but not sufficient. Value still requires builders to channel that intelligence into things humans want to use.
• Education will change more than most people expect — the skill that matters is learning to ask the right questions, not memorizing facts or writing code.
• For companies building on LLMs: proprietary data and the flywheel to keep generating it is the moat. Workflow fit is the other. Windsurf collecting accept/reject signals to train their own models is the canonical example.
• The closer product and post-training are, the better the model gets at the things users actually want. PMs embedded in research teams produce more leverage than PMs on the product surface.

Career decision-making

• Optimize for learning over prestige or safety. Jump when you feel you've exhausted what you can learn in a role.
• Join founders who have a strong, specific insight about how humans are wired and what they fundamentally need — not just a good idea.
• "We may not be right, but at least we're not confused." Conviction and directional clarity beat perfect decision-making.
• Failure is just data: Instagram's camera app Bolt launched in Australia, retention didn't asymptote, and the team moved on. The lesson — even the best team with the best taste can't always predict what will hit.