Miju Labs

← All buyers · Vercel (v0)

M

Max Leiter

Former v0 co-lead (now at Anthropic) · Vercel (v0)left

medium confidenceTier C · Context onlyDesign & UI productsUI

Leiter co-created v0 and the Vercel AI SDK and co-led the v0 team, and he authored the public posts describing v0's composite model family and its coding-agent architecture. He has since left Vercel: his own announcement says he departed after almost four years, and his LinkedIn headline now reads Claude Code at Anthropic. He should no longer be treated as the v0 contact.

Background

Long-running open-source developer (The Lounge IRC client, Drift) who joined Vercel around 2022-2023 and became one of the founding engineers of its AI products: he co-created v0.app with Shu Ding, shadcn and Jared Palmer, and started the AI SDK with Shu Ding and Lars Grammel. He then helped build and lead the v0 team alongside Gaspar Garcia, and became the public voice of v0's model work through the Vercel engineering blog and podcast appearances.

His departure post frames it as a difficult choice after almost four years. If the LinkedIn headline is current, he is now working on Claude Code at Anthropic — a move from the buyer side of UI-generation quality to a coding-agent team at a frontier lab.

What they run now

Historic rather than current for this index. At Vercel he owned and documented v0's composite pipeline: a frontier LLM plus retrieval plus custom reinforcement-fine-tuned 'autofixer' models (vercel-autofixer-01, trained with Fireworks AI), evaluated chiefly on error-free generation rate, with user feedback folded back into iterations. That is an automated, correctness-led quality regime rather than a human-taste one — v0 measures whether the generated app runs, not whether it looks good.

What they buy

Nothing relevant now. At Vercel he bought compute and RFT tooling (Fireworks AI) rather than human data; v0's published eval metric is error-free generation rate, with human input arriving free as user feedback. At Anthropic, if the headline is right, he is a consumer of internal human-data programmes rather than a buyer — Anthropic's human-data purchasing sits elsewhere.

Career

  1. 2026 – presentClaude Code (per LinkedIn headline), Anthropic — Start date not verified; headline only.
  2. 2022 – 2026Staff Software Engineer; co-creator and co-lead of v0, Vercel — 'Almost four years' per his own departure post; co-created the AI SDK and v0.app.
  3. 2016 – presentMaintainer, The Lounge (open source)

On the record

Leaving Vercel · 2026

Announced leaving Vercel after almost four years, from starting the AI SDK and v0 to co-leading the v0 team. source

Interviews & talks

Why it matters here

Little, now. His value to Julian is as a correction to the index and as market intelligence: v0, the most obvious UI-generation buyer at Vercel, deliberately optimises for functional correctness rather than design taste, which is a useful counter-example when arguing that taste data is underbought. The live Vercel contact is Gaspar Garcia, who co-led the v0 team with him.

How to reach

Do not approach as a Vercel buyer. If contacted at all, treat it as a market-intelligence conversation: he is publicly accessible on X and writes candidly about model quality, and he can explain why v0 never bought human design evaluation. Speculation: that answer ('we could not make design quality measurable cheaply enough') is exactly the objection Julian will hit everywhere, so it is worth hearing from someone with no stake in the sale.

What we could not establish
  • Exact departure date from Vercel and start date at Anthropic; the LinkedIn headline is the only evidence for the Anthropic role and I could not open the profile.
  • Who now leads the v0 model team — Gaspar Garcia is the most likely successor but this is unconfirmed.
  • Whether v0 has since added any human-graded design evaluation.

Why they are on the map

Authored v0 model family (2025) and v0 coding-agent (Jan 2026) posts. source — evidence dated 2026-01

Needs the data for a model; writes the spec.

How Vercel (v0) buys data

v0 composite model family (Jun 2025): trained vercel-autofixer-01 via reinforcement fine-tuning with Fireworks AI; eval metric is error-free generation rate; human feedback from v0 users informs iterations. Jan 2026 post details dynamic prompts, streaming fixes and autofixers. No human-data vendor named.

Small v0 engineering team authoring model posts (Leiter, Khandelwal, Garcia, Pesok); eval is automated success-rate rather than human taste.

Known vendors: Fireworks AI (RFT training infra, not human data)

Sources

  1. Max Leiter — personal site (v0 and AI SDK co-creation, Vercel 2023-2026)
  2. X — Leiter announcing his departure from Vercel
  3. LinkedIn search result headline — 'Claude Code @ Anthropic'
  4. Vercel — Introducing the v0 composite model family
  5. Vercel — How we made v0 an effective coding agent (2026-01)