Reading Notes

Jenny Wen on Designing Claude, the Legibility Framework, and the Three Designer Archetypes

Source: Jenny Wen on Designing Claude, the Legibility Framework, and the Three Designer Archetypes

Notes — Jenny Wen on Designing Claude, the Legibility Framework, and the Three Designer Archetypes

Four questions [Adler frame]

Q1 — What is it about? How the design role changes when engineers can ship faster than any design process can track. Wen, who leads design for Claude Cowork at Anthropic and previously directed design at Figma, argues that the taught design process — research, diverge, converge, polished mocks — is dead, and that design work has split into two activities: supporting execution and creating direction. Secondary threads: hiring (three designer archetypes), spotting early ideas (the legibility framework), running teams (psychological safety, low-leverage-as-high-leverage), and the IC-versus-management question for design leaders.

Q2 — How is it argued? From Wen’s own working week rather than from theory. She gives proportions — mocking fell from 60–70% of her time to 30–40%, with the recovered time going to pairing with engineers and to writing production code herself. She narrates concrete episodes: the Cowork ‘ten days’ myth, the internal Claude Studio prototype that seeded the skills framework, a team that mimicked her crit phrases. The legibility framework is borrowed (from VC Evan Tana) and re-applied to design; the archetypes are her own hiring heuristic.

Q3 — Is it true? The core claim — that engineering’s acceleration forces design to change — is well-grounded in her direct observation, though it is a frontier-lab view that she herself flags may not hold at slower-moving companies. [?] The prediction that AI will get good at taste and judgment is asserted from trajectory (coding fell faster than anyone expected) rather than demonstrated; she is careful to separate taste from accountability for the decision, and rests the durable human role on the latter. The archetypes are a useful hiring lens but are not validated against outcomes — they describe what she finds interesting to hire, not what provably succeeds.

Q4 — What of it? Directly actionable for design leaders: shift time from mocks to pairing and implementation; treat coding tools as table-stakes vocabulary, not a specialisation; hire for resilience and one of the three archetypes; read the team’s internal channels to spot illegible-but-energetic ideas. The ‘build trust through speed’ discipline — ship a labelled research preview, then visibly iterate on feedback — is a concrete brand-protection rule for shipping unfinished AI products.

Glossary

Design stratification — Wen’s claim that design work has split into two distinct activities: execution support (implementation, polish, keeping parallel engineering work coherent) and vision creation (deciding what should exist and what the interaction model is). AI compresses the first; the premium moves to the second.

Legibility framework — a two-by-two for evaluating early ideas, originated by VC Evan Tana (a partner at SPC) for founders and ideas, and re-applied by Wen to product ideas. An idea is legible if observers immediately grasp it, illegible if it sits on the frontier or is told in a way that does not yet land. Illegible ideas carrying energy (excitement, curiosity that people cannot fully articulate) are where novelty lives and deserve the most investment.

Three designer archetypes — Wen’s hiring lens: the strong generalist (block-shaped — roughly 80th-percentile across several core skills, not merely broad), the deep specialist (deep-T — top ~10% in one domain, e.g. near-engineer technical depth or icon craft), and the craft new-grad (early-career, humble, fast-learning, unburdened by ingrained process).

Build trust through speed — shipping an explicitly labelled research preview (‘this is the worst it will ever be’), then visibly responding to feedback and shipping fixes fast. Trust is lost not by shipping early but by shipping early and then going quiet.

Low-leverage-as-high-leverage — Wen’s rejection of the manager’s two-by-two that delegates away ‘low-leverage’ nitty-gritty. A senior leader who dogfoods, repros bugs, files PRs, or vibe-codes an anniversary card creates familiarity and signals that nothing is beneath them — which is high leverage precisely because it is them doing it.

Research preview — Anthropic’s term for a product released early with explicit caveats (Cowork is the example), on a promise to iterate publicly on user feedback.

Cowork — Anthropic’s agentic product Wen now designs; ‘Claude with hands’ (Lenny’s phrase, which she adopts). Took ten days to move from a working internal build to an external release — but rested on a long prior history of prototypes and explorations.

Block-shaped vs T-shaped — in the standard T-shaped model, one deep skill (the stem) plus broad awareness (the bar). Wen’s block-shaped generalist has several stems at once; the deep-T has an unusually long single stem.

Key frameworks

Design stratification [§ How the design process is changing]

Engineering changed first; design is forced to follow. With engineers running ‘seven Claudes’ in parallel, the designer can no longer gate work behind beautiful mocks. Two kinds of work remain:

  1. Execution support. Anyone can voice an idea; an engineer makes a scrappy version; the designer polishes in code and keeps the parallel streams coherent. Wen does ‘last-mile’ implementation herself in VS Code.
  2. Vision creation. Pointing the parallel work at one cause so it adds up. The horizon shrank from two-to-ten-year visions to three-to-six months, and the artefact is often a prototype rather than a storytold deck.

Time shifted: mocking 60–70% → 30–40%; the rest now pairing with engineers and writing production code. Figma survives for what code does badly — throwing 8–10 divergent directions at the wall, and fine typography/interaction exploration.

The legibility framework [§ The legibility framework]

Borrowed from Evan Tana’s founder × idea two-by-two and re-applied to ideas:

  • Legible idea + legible founder — probably not novel; someone is already doing it.
  • Illegible idea + energy — the interesting quadrant. On the frontier, or told in a way that does not yet land, but generating excitement people cannot fully articulate.

Wen’s design application: scan internal channels for illegible-but-energetic ideas and do the work of making them legible — through storytelling, UX, or form factor. Worked example: the internal Claude Studio prototype looked incomprehensible to her (‘I don’t know what’s going on’) but carried obvious energy among researchers; the skills framework and instruction-markdown files came out of it, and its display patterns (Claude’s plan, to-dos, context, files) were pulled into Cowork. ‘Designers can be more like VCs.‘

The three designer archetypes [§ Hiring designers]

What Wen looks for now, on top of resilience and willingness to adapt:

  1. Strong generalist (block-shaped) — 80th-percentile across several core skills. Flexes easily as the role stretches toward PM- and engineer-shaped work. Rare and hard to hire.
  2. Deep specialist (deep-T) — top ~10% in one domain. Examples: a designer technical enough to be half a software engineer (valuable when working directly against the model), or exceptional visual/icon craft (differentiating when ‘anybody can make anything’).
  3. Craft new-grad — early-career, wise beyond their years, humble, eager, fast-learning. Overlooked because most companies hire senior; valuable precisely because they lack ingrained processes and assume anything is possible. Advice to break in: build and ship real things, share them, find a building community (she cites Socratica).

Claude itself is not yet hireable as a designer — good at a first pass and at presenting options, but not yet a generalist, specialist, or craft new-grad.

Psychological safety and high standards [§ Running teams]

Two traits that look opposed but compound:

  • Roasting as a safety signal. Not mandated. When teammates poke fun at each other — and at the manager (her team mimicked ‘okay, what are next steps?’) — it signals they do not fear being fired for speaking.
  • High standards on top of safety. Once safety exists, demanding standards land without fear. Wen frames it as tough-parent leadership: always there, never firing on a whim, but visibly wanting the best work. Lenny ties this to Radical Candor — care personally, challenge directly.

IC versus management for design leaders [§ Going back to IC]

Wen returned from directing a 12–15-person Figma org to full-time IC work at Anthropic, to stay close to a craft changing too fast to manage from a distance. The lesson: hands-on time bought hard skills and the empathy to manage the new way of working later. She draws the analogy to engineering orgs that rotate new EMs through real tasks before they manage. Her view of the durable manager: not pure people-management, but direction-giving plus creating the environment for the team’s best work.

Connections