Engineer PM Convergence

Sources#

• Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next
• Designing with Claude: From prompt to production
• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)
• State of AI 2026: The Builder's Economy
• Thread by @AndrewYNg
• Why Netflix is betting on systems thinkers—not specialists—in the AI era | Elizabeth Stone (CPTO)

Summary#

Both Boris Cherny (Sequoia AI Ascent 2026) and Cat Wu (Lenny's Podcast, April 2026) report the same trend from inside Anthropic: roles are merging. Engineers do PM work, PMs ship code, designers ship code, every functional role on the Claude Code team writes code. The convergent skill that matters across the merged role is product taste — deciding what to build, given that building is now cheap. This challenges the conventional model of cross-functional teams (one PM, one designer, several engineers, separate eval/QA) and points toward smaller groups of high-context generalists who individually own end-to-end delivery.

What's actually happening at Anthropic#

From Cat Wu:

• Everyone on the Claude Code team codes — engineering manager, product manager, designers, data scientist, finance, user researcher.
• Designers were formerly front-end engineers.
• "There are many engineers on our team who are fully able to end-to-end go from see user feedback on Twitter through to ship a product at the end of the week with almost no product involvement."
• Hiring bias: "We're pretty focused on hiring engineers with great product taste."
• Boris ↔ Cat split is "80% mind-meld, 20% domain-driven."

From Boris Cherny:

• "Generalists across disciplines" — engineers who are also great at design, or product + data science.
• "Everyone on our team codes."
• The "product engineer" archetype (iOS + web + server) is the baseline generalist; the new pattern is cross-disciplinary generalist.

From Dan Carey (Anthropic Labs, building Claude Design):

• The team was three people for most of development; "everyone on the team does everything — the engineers talk to users, PMs write code, designers do data analysis."
• "The lines between the roles on this team have essentially dissolved." You keep your specialization and unique perspective, but at any moment one person can talk to 10 users, find the underlying problem, design the fix, ship it, and keep iterating — "most things on the team are totally solo."
• "Claude is a pretty good team member" — the third teammate that lets a 3-person team behave like a larger one.

Outside Anthropic: the same convergence, and its failure mode#

Andrew Ng reports the trend from outside the frontier labs entirely (June 2026, practitioner-opinion) — the third independent vantage:

"With coding agents speeding up software development, more engineers are starting to play a partial product management role… the hardest part is shaping the product vision and striking a balance between building (bridging the gap between vision and spec) and getting user feedback to evolve the vision. It is important to do both!"

Ng adds what the Anthropic accounts don't: a named failure mode. Engineers newly holding the product role over-invest in the half they already know how to run. Building is fast, legible, and fun; the external feedback loop is slow, unpleasant, and unautomatable — so it gets skipped, and the vision stops updating. He is symmetric about the direction of travel: "engineers are playing an expanded role (just as product managers and designers now do more engineering)."

Ng also declines the word this page leans on. What Cat Wu calls taste, he calls a context advantage — not a faculty the generalist cultivates but knowledge they happen to hold. If he is right, "hire for taste" is really "hire for proximity to the user," and it is perishable.

Surveyed beyond the frontier labs (ICONIQ, Q2 2026)#

The Anthropic/OpenAI accounts are frontier-lab self-report; ICONIQ's State of AI 2026 survey (~305 AI-building software companies, empirical) shows the convergence is broader. AI-heavy companies are more cross-functional (42% at 50%+ AI-revenue vs 35% below) and flatter (72% vs 56% run on 1–4 management layers at $100M+ scale) — the structural precondition for merged roles. The convergence appears in the network anecdotes too: one operator "collapsed PM and designer into single-person product ownership," and RevOps teams are "replacing non-AI-native RevOps hires with AI-native ones." Whether this validates Cat Wu's small-teams-of-generalists model or degrades into the discipline-loss Ambrosino warns about is exactly the open tension — ICONIQ measures that the collapse is happening at population scale, not whether it preserves the specialties.

The enterprise-scale vantage: fluidity with guardrails (Netflix)#

Elizabeth Stone (Netflix CPTO, July 2026, practitioner-opinion) confirms the trend from the largest non-lab org in the wiki — and is the most explicit about its boundary conditions. She hears the role-confusion frustration inside Netflix ("what is my job anymore?") and reads it as "a storming phase before the forming phase." PMs, designers, and data scientists now "get farther in the product development life cycle before engineering really needs to be front of the line." But the fluidity is sanctioned only when the business problem is clear and the engineering partner is aware — "it's not throwing a bunch of spaghetti at the wall" — and she draws the line the lab accounts don't: "Do I believe that means anyone should be shipping code to production? Probably not."

Her craft claim is the counterweight to full convergence, aligned with Ambrosino: comparative advantages persist — the data scientist knows whether the data can be trusted, the PM whether the what is framed right, the engineer the how (scale, quality, deployment consequences). "I still see a craft excellence that's really important… I still find great engineering to be scarce, great data science to be scarce, great creativity to be scarce." Roles speak more languages; the craft nucleus stays. Organizationally her enablers are infrastructure, not coordination: source-of-truth data, production guardrails, and reiterated human accountability ("it can be that an agent wrote the code… it doesn't make people not have the responsibility") — see Systems Thinking Over Specialization.

The taste argument#

"As code becomes much cheaper to write, the thing that becomes more valuable is deciding what to write."

— Cat Wu

Code generation cost is dropping. The bottleneck shifts upstream:

• Picking which feature to ship from 10K GitHub issues
• Designing the UX so the model's strengths are exposed and weaknesses patched
• Knowing when to push something out as research preview vs full product
• Distinguishing the 90 features people say they want from the 10 they'll actually use

This is product taste, and it's not bound to a job title — engineers can have it, designers can have it, PMs can have it. Whoever has it ships well; whoever doesn't ships drift.

Why now (Lenny's contrasting take)#

Lenny pushes back during Cat's interview, citing Anthropic Head of Growth Amol Avasare's prior episode: Amol says he needs more PMs because engineers are shipping so fast that designers and PMs can't keep up — every day there's a new feature.

Cat agrees this can be the case, but on Claude Code prefers the engineer-with-taste path: hire fewer of those, scale leverage, fewer coordination handoffs.

The two answers aren't contradictory — they describe different team shapes:

• Cat's model: small teams of generalists with high autonomy (Claude Code, ~couple dozen people)
• Amol's model: larger orgs where engineers move fast and need PM/design support to keep parity (Growth, GTM, Enterprise)

The asymmetry: human EQ remains#

Cat names what isn't merging: tacit, common-sense, EQ-heavy work — knowing the right venue to communicate with stakeholders, sensing when a launch is ready, knowing what counts as a fair trade-off. The model is improving here but isn't there yet, and humans still provide the connective tissue across a launch.

"Just do things"#

Cat's life motto:

"Jobs are fake. If you understand the constraints, you can figure out what you can do and then just like try to do it quickly, learn from the mistakes and apologize or fix them if you did something wrong."

This is the cultural substrate that makes role merging work. If roles are bounded by JD, no one acts cross-functionally. If "do what needs doing" is the norm, the merging happens organically.

Mission alignment is what makes "just do things" not chaotic — Cat: "If there's two competing priorities, we'll talk about which one is more important for Anthropic's mission. And it makes it a lot easier to decide which of the two we prioritize. And then everyone will stand behind the one that we decide."

Implications#

1. Hire for taste, regardless of role. Cat's stated bar: anyone with strong product taste they can demonstrate.
2. Cross-train aggressively. If your designers don't ship code, that's an organizational choice, not a constraint.
3. Smaller teams. A 5-person team where every person owns end-to-end ships faster than a 15-person team with handoffs.
4. PRDs become lighter. Cat: short bullet PRDs for ambiguous features; metrics readouts and team principles do most of the alignment work; full PRDs only for heavy-infra projects.
5. Career ladders fragment. Cat: "We're sacrificing product consistency" as a tradeoff for shipping speed; career-ladder consistency is a quieter casualty.

Connections#

• Task Crossover — the convergence measured at population scale and outside tech: 43.5% of occupation-specific ChatGPT work use is another occupation's work, and crossover is highest in the smallest workspaces — role merging tracks organizational thinness, not just AI capability
• Implementation Abundance Inverts Product Work — the OpenAI-side, process-level statement of the same "deciding what to build is the bottleneck" shift
• Role Averaging, Not Role Elimination — Ambrosino's counter-caution: welcome the convergence, but don't eliminate roles as specialties with knowable best practices
• Outsource Your Thinking, Not Your Understanding — the converged generalist must keep understanding while delegating execution
• Seven Powers Applied to AI — process power erodes as roles converge into generalists
• Cat Wu — primary articulator
• Boris Cherny — converging report from same team
• AI Native Product Cadence — the cadence that makes role merging viable
• Claude Code Best Practices — engineer-with-taste is the user persona Claude Code targets
• Harness Shrinkage as Models Improve — as harness shrinks, the surface area of a "PM" role shrinks; the surface area of a "person who ships product" expands
• Printing Press Software Democratization — same direction, different timescale: software literacy expands, role distinctions blur
• Agent Loop Pattern — at the limit, individual contributors run dozens of agents and ship like teams used to
• Human-AI Accountability Redesign — workforce-wide mirror of this convergence: as agents take execution, human roles concentrate on supervision/judgment/oversight quality
• AI Employee Framing — counter-evidence for the cross-functional generalist: in HR/finance contexts, framing AI as colleague (rather than tool) erodes individual accountability rather than expanding it
• Compute Allocator — "deciding what to write" as the bottleneck skill, restated at the level of individual model invocations (Thariq Shihipar)
• Living Design System — tooling that lets non-engineers self-serve high-fidelity assets is the artifact side of roles blurring across the org
• Founder as Agent Orchestrator — within-company role merging extrapolated to the founder/company-of-one scale; same direction, smaller unit
• AI-Native Startup Lifecycle — the founder/startup version of this convergence: stage-by-stage compression of function-specific gates
• Evals as Product Spec — the canonical hybrid-role activity: PMs writing evals, engineers writing evals, both converging on "ten great evals" as a shared definition-of-done artifact
• Managers as ICs — Fiona Fung extends "everyone codes" up the org chart: every Claude Code manager starts as an IC
• Dogfooding as Product Discipline — "creative builder with product sense" is the dogfooding-fed hiring profile this convergence produces
• Dan Carey / Anthropic Labs — the 3-person Claude Design team as a second case study: "everyone does everything," roles dissolved
• Compounding Loop Optimization — the small, role-dissolved team is what makes "build your own tooling, run the loop solo" viable
• Returns to Expertise in Agentic Coding — "deciding what to build, given building is cheap" as the bottleneck skill, measured at population scale: domain/product understanding (not coding) predicts who succeeds with an agent, and managers (delegation taste) post the highest verified success
• Conversation-to-Delegation Shift — OpenAI's Codex study sees the same convergence in usage: intensive users "manage a portfolio of agentic work," shifting their own effort toward delegation, supervision, and integration
• Parallel Agent Orchestration — the role convergence made literal: running a fleet of concurrent agents is the IC-becomes-orchestrator shift, with adoption numbers
• Organizational Complements to AI — the job-redesign complement: realizing AI's value requires roles that direct/monitor/integrate agent output rather than execute tasks directly
• The Three Loops of AI-Native Building — Andrew Ng's third-vantage report, plus the failure mode: engineers-turned-PMs over-run the inner loop they enjoy and skip the outer one that updates the vision
• Context Advantage, Not Taste — Ng's reframing of the convergent skill: not taste but proximity to the user, which makes "hire for taste" a perishable strategy
• AI-Native Organization — the whole-company version: ICONIQ's cross-functional/flatter-org data and the PM+designer-into-single-person-ownership anecdote are this convergence measured across ~305 AI-builders (see that page's restructuring section)
• Systems Thinking Over Specialization — Stone's org-level answer to the convergence: staff the paved paths and design systems that let fluid roles ship coherently, and hire the systems thinkers who build them
• Excellence as an Operating System — the Netflix cultural substrate for fluidity, parallel to "just do things": trust + accountability instead of role boundaries
• Elizabeth Stone — the enterprise-scale vantage: fluidity sanctioned, production-shipping line held, craft scarce

Open Questions#

• Does this scale beyond ~50-person Claude Code-style teams? Boris hedges: "I think this is going to be a question for years."
• What happens to formal PM career ladders in companies where engineers do PM work? Open at Anthropic per Cat.
• Cross-disciplinary generalist is a hiring bar — where does the supply come from? Career changers, or new-grad bias toward AI-native education?

Derived#

• Learning to Co-Work with AI: A Software Engineer's Field Guide — engineer-PM convergence framed as a personal career-skills directive (product taste as bottleneck skill cluster)
• Opinions on Using AI Tools & the Future of the Software Engineering Role — role convergence as a core pillar of the future-of-the-SWE-role synthesis

Sources#

• How Anthropic's product team moves faster than anyone else | Cat Wu (Head of Product, Claude Code)
• Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next
• Thread by @AndrewYNg — Andrew Ng, The Batch (2026-06-30), practitioner-opinion: engineers into partial PM roles, and the balance failure
• State of AI 2026: The Builder's Economy — ICONIQ Growth, State of AI 2026: The Builder's Economy (2026-07-08, empirical): §"Talent & Organization" cross-functional/flatter-org cuts and the PM+designer-into-single-person-product-ownership anecdote — the convergence at population scale beyond the frontier labs
• Why Netflix is betting on systems thinkers—not specialists—in the AI era | Elizabeth Stone (CPTO) — Lenny's Podcast (2026-07-19, practitioner-opinion): the Netflix-scale confirmation with guardrails — storming-before-forming, farther-before-engineering, the production-shipping line, craft scarcity