Boris Cherny spent seven years at Meta (2017-2024), going from E4 to E8, before leaving to lead Claude Code at Anthropic. Along the way he shipped products that failed, products that scaled, and infrastructure that still runs today. These are the 16 patterns he learned, organized by the career stage where each one hit hardest.
Stage 1: Foundation (E4, 2017-2018) — “Earn Your Seat”
1. Always Start with Code
Boris’s most consistent belief: the world will conspire to pull you away from coding. Meetings, docs, presentations, consulting. If you let it happen, you lose intuition for your codebase, lose credibility with fellow engineers, and miss opportunities.
In his first month at Instagram in 2022, he wrote more code than in his previous six months at Facebook. By 2023, he shipped 868 diffs — his most productive year ever — and it led directly to his E8 promotion. The pattern held at every level.
The test: If you haven’t shipped a diff this week, ask yourself why.
2. Side Hustles Build Networks and Influence
Before joining Meta, Boris built Undux, an open-source state management framework. At Meta, he spent a month biking between MPK buildings doing 30-40 tech talks to pitch it. Through those talks he met engineers on React, Flow, and DevInfra — relationships that paid off years later when he needed cross-org support for IGWWW.
In parallel, he started the San Francisco TypeScript Meetup (the world’s largest at the time) and wrote O’Reilly’s first TypeScript book. That brought him to Joe Pamer, the creator of TypeScript and later Instagram’s Head of Infra — who became the key connector for IGWWW.
The pattern: Side projects introduce you to people you’d never meet through your job. Those relationships compound. Better Engineering is the easiest way to grow your network as an engineer.
3. Scrappy MVPs with Hypothesis-Driven Iteration
Chats in Groups had no PM, no designer, no UXR, and no DS. Boris served as TL and PM with 3 engineers. They designed in Sketch, did guerilla usability testing with cafeteria workers in MPK21, and taught themselves Daiquery for data.
They were religious about funnel logging — learning in days instead of waiting weeks for Deltoid. They built on one platform first to learn fast, then expanded what worked. Each week they shipped half a dozen experiments. In months, they drove 1 million Meaningful People. Everyone on the team got promoted.
The contrast: Communities as the New Organizations started with a pitch deck, a Zuck review, and 120 headcount. Almost every workstream was eventually canceled.
Stage 2: First Big Bet (E5→E6, 2019) — “Shape the Platform”
4. Get Ahead of Platform Shifts
Boris heard about Comet — Facebook’s secret full rewrite — and immediately pushed to make Groups the next surface on it. He got in early, and his technical decisions became the defaults for the entire platform:
- Actor Framework for permission management
- Oncall enforcement patterns
$keysuffix for Relay fragment props- CanViewerDo for GraphQL APIs
These conventions are still in the code today. Later adopters built within the framework Boris defined.
The principle: Platform shifts create a power vacuum. No best practices, no conventions, no “we’ve always done it this way.” The first people to write code define the rules everyone else follows.
5. Present Three Options (Anchor to the Middle)
When pitching Groups on Comet to VP Jenn Dulski, Boris asked for “6-14 engineers depending on how fast we want to get it done.” Jenn picked the middle option. He used the same technique for SEO — PM director first, then Fidji Simo.
The rule: In VP reviews, always present three options at different investment levels. 80% of the time they pick the middle one. Make the middle one what you actually want.
Stage 3: Scaling Impact (E6→E7, 2020-2021) — “Multiply Yourself”
6. Do What PMs Aren’t Doing
Meta has a low PM-to-engineer ratio. Many valuable product opportunities go undriven. Boris spotted this with Groups SEO: his PM Hasan was stretched thin and simply didn’t have bandwidth. Boris did the opportunity sizing, wrote the pitch deck, convinced the director, and eventually catalyzed the Meta SEO team in Tel Aviv.
The pattern: If you’re a product-minded engineer, look for impactful product opportunities that PMs don’t have time to drive. Own them. You create scope, drive impact, and make the team look good.
7. Delegate What You Love, Not What You Hate
Boris was generating so many infra ideas that he had to delegate almost everything. The key: he delegated work he understood deeply (Bloks migration, lint rules, large test objects), so he could monitor quality and course-correct.
If you delegate what you don’t want to do, you can’t tell if it’s going well.
From Andy Grove’s High Output Management: “You always delegate the thing you do want to do and that you know well because then you can monitor the progress.”
8. Know When to Say No
Lightweight Groups v2 was a product Boris knew was wrong from day one. Instead of simplifying the existing Groups product piece by piece, they built a brand new product. He saw the problem but focused on “what was expected of him” rather than tackling it directly.
A year later, the project was killed. His reflection: “I probably would not have taken on the project.”
The distinction: GroupSync failed because of skipped homework. LG v2 failed because he went against better judgment. The latter is more damaging to your career and your team.
9. Admit Mistakes, Correct Fast
When building Public Groups, Boris debated with Bob Baldwin about whether to split the membership data model into two assocs. Boris lost the debate. A year later, 99% of callsites used the two assocs identically. Boris proactively reversed his own decision and merged them back — a painful six-month migration, but the right call.
Eric Schmidt’s framework: First decide if a decision is reversible. If yes (99% of cases), speed matters more than correctness. Decide fast, correct later.
10. Follow Latent Demand
You can never get people to do something they don’t already do. Great products come from finding intent users already have and building a proper product around it.
- Facebook Marketplace: 40% of Groups posts were buying and selling
- Facebook Dating: 60% of profile views were opposite-gender non-friends
- GroupSync (failure): tried to create behavior (auto-syncing chats and groups) that users weren’t already doing
Boris’s words: “You can find the intent they have and steer it to let them better capitalize on that intent.”
Stage 4: New Org, Fresh Start (E7, 2022) — “Prove It Again”
11. Outsider’s Advantage
Within weeks of joining Instagram, Boris identified the Python stack as “10x worse than Meta’s Hack stack across pretty much every dimension.” Old-timers didn’t see it: they’d stopped coding, were habituated to the warts, or took pride in being productive with broken tools.
The window: When you join a new team, you can briefly see problems insiders are blind to. This advantage has an expiration date — by 2021 at Facebook Groups, Boris’s own outsider advantage had faded.
12. Street Cred Before Mentorship
In his first conversations at Instagram, Boris asked Tim and Yoshi about their growth plans and was “greeted with blank stares.” Levels aren’t public, and he was the new guy. A year later, an E8 engineer asked Boris the same questions, and his reaction was: “Who is this person to ask me that?”
The lesson: Don’t try to mentor people who don’t know you. Build credibility through visible, hands-on work first. Street cred is earned per-context — it doesn’t transfer across orgs.
13. Prove the Problem Before Proposing a Solution
For IGWWW, Boris didn’t start by proposing a solution. He built consensus that the status quo was broken:
- Quantitative: Pulled Phabricator and HR data on which stacks engineers use
- Qualitative: Interviewed 40 top ICs one-on-one about their pain points
- Broad survey: Confirmed the findings weren’t biased
Only after all three signals converged did he start discussing solutions. When he presented the data to senior engineers, they were surprised — their comfort with the codebase had made it hard to see the forest for the trees.
Boris’s words: “You can’t solve a problem if no one thinks it’s a problem.”
14. Step Back to Let Others Grow
After recruiting Jake Bolam to TL IGWWW, Boris noticed something: “The more I stepped back, the better Jake’s work got.” Jake used a completely different approach — formal teams, conferences for alignment, investing in dev tools first — that was better suited for scale than Boris’s volunteer model.
The lesson: When handing off a project, truly step back. Don’t lead from someone else’s shadow. Give space and let the new owner do things their way.
Stage 5: Breakthrough (E7→E8, 2023+) — “Build the Future”
15. Build Product and Infra in Parallel
Instagram Profile took 8 seconds to load. Instead of building a GraphQL framework first and finding adopters later, Boris fixed Profile performance (product) while writing a custom GraphQL executor (infra) at the same time. The A/B test showed -42% latency and one of Instagram’s biggest session wins that half.
Today, that executor processes a quarter of Instagram’s CPU and 40% of its TAO at peak.
Boris’s words: “That kind of close product-infra collaboration is rare at Meta these days.” The best infra isn’t designed in a vacuum — it grows alongside product needs.
16. Build for the Model Six Months From Now
At Anthropic, Boris’s manager Ben Mann pushed him: “Don’t build for the model of today, build for the model six months from now.” For months, Claude Code was mediocre — used for about 10% of Boris’s coding. Then Sonnet/Opus 4 dropped and usage exploded, because the product was already ready.
The counter-example: Competitors who optimized for current model capabilities (tab autocomplete, inline suggestions) had to rebuild when agentic coding became possible.
The Failure Taxonomy
Across seven years, Boris identified seven distinct ways projects fail:
| Type | Example | Root Cause |
|---|---|---|
| Skip homework | GroupSync | Built without checking the data first |
| Against better judgment | LG v2 | Knew it was wrong, did it anyway |
| Org friction | Chats in Groups | Had traction, but cross-org friction killed it |
| Cart before horse | Communities | Headcount first, then went looking for a problem |
| Scope inflation | LG v2 | Design vision executed literally as a new product |
| Zombie handoff | Oculus Japan | Sponsor left the company, project died instantly |
| Priority tsunami | Chats in Groups | Exec directive drowned a working project |
His biggest growth from 2017 to 2023: moving from “skip homework” to “prove the problem exists before building.”
Source: Boris Cherny’s two Workplace career posts (~30,000 words combined) and the Ryan Peterman podcast interview (April 2026).