<- all tokdocs

Jason Lemkin on AI Code Review: What the 20VC Clip Leaves Out

By , founder of Mavric and Apso

View on TikTok ->

In a 20VC clip, SaaStr founder Jason Lemkin argues that code review is now the most important category in software, because AI writes code far faster than teams can make it production-ready. The bottleneck he describes is real. The clip treats code review as an open field, but by the time it was posted Cursor already owned a code review company and Anthropic's Code Review product was generally available.

What does Jason Lemkin say about AI code review?

A TikTok from @20vc_tok shows Lemkin saying that everyone is "building 10, 12, 14 hours a day in Claude Code or Cursor," and that turning generated code into shipped code takes "an order of magnitude more" time than generating it. He calls the goal "production intent." He says that if he could find "the best next generation true code review and the best team, I'd make 10 bets because we need it," and adds, "And they may all get killed by Anthropic. We'll see."

He also says, "It does not mean there'll be a 600 billion code review cursor." The sentence is a negation, but it uses Cursor as the yardstick.

How much is Cursor actually worth?

Nowhere near $600 billion. Cursor's maker, Anysphere, raised a $2.3 billion Series D at a $29.3 billion valuation on 13 November 2025. TechCrunch reported a further round in progress at $50 billion before SpaceX announced on 16 June 2026 that it would buy Anysphere for $60 billion in stock. The deal closed on 14 August 2026, according to Wikipedia, and Cursor is now a wholly owned SpaceX subsidiary.

The "$600BN" figure is ten times Cursor's sale price. Lemkin's point survives, because $60 billion is still an outcome code review tooling may never match.

Has Anthropic already built a code review tool?

Yes. Anthropic launched Code Review on 9 March 2026 as a research preview for Team and Enterprise plans. When a pull request opens, it "dispatches a team of agents" that look for bugs in parallel, verify findings to remove false positives, rank them by severity and post comments. Anthropic's preview figures:

  • On pull requests over 1,000 changed lines, 84 percent get findings, averaging 7.5 issues.
  • On pull requests under 50 lines, 31 percent get findings, averaging 0.5 issues.
  • Fewer than 1 percent of findings are marked incorrect.
  • Reviews cost roughly $15 to $25 each.
  • Inside Anthropic, pull requests with substantive review comments went from 16 percent to 54 percent.

Code Review became generally available on 15 October 2026 for Team and Enterprise organizations. Anthropic says "most reviews now finish about 3x faster (~20 minutes down to 6-10 minutes)," and admins can choose Quick, Standard or Deep review per repository.

Anthropic also offers ultrareview, a research preview run with /code-review ultra. It runs "a fleet of reviewer agents in a cloud sandbox," takes about 15 to 25 minutes, reviews up to 500 changed files and 8,000 changed lines by default, and costs roughly $5 to $25 per review after three free runs on Pro and Max.

The TikTok was posted on 23 October 2026, eight days after general availability. The recording date of the conversation is unknown, so Lemkin may have spoken earlier.

Did Cursor buy a code review company?

Yes. Cursor announced it was acquiring Graphite on 19 December 2025, describing it as "a code review platform utilized by hundreds of thousands of engineers at leading organizations." Cursor's post makes Lemkin's argument too: "reviewing changes, merging them safely, and collaborating effectively have increasingly become the bottlenecks to building production-grade software." Graphite continues to run independently. Cursor did not disclose a price, and third-party reports put it well above Graphite's roughly $290 million prior valuation.

That supports Lemkin's thesis that review matters. It also means the two companies he names were already competing in the category, which is a harder setting for the ten seed bets he describes.

What are Astra, Fable and Kimi?

Lemkin says he can "rotate between Astra and Fable" and maybe Kimi "if it catches up." All three are real models from different labs:

  • Astra is OpenAI's model, launched on 3 September 2026. OpenAI called it its "best model for software engineering to date."
  • Fable is Anthropic's. The models overview lists Claude Fable 5.1 (claude-fable-5-1) with a 1M-token context window, 128K max output and pricing of $10 per million input tokens and $50 per million output tokens.
  • Kimi is from Moonshot AI, a Chinese lab founded in March 2023. Its releases include K2.7 Code in June 2026 and K3 on 16 July 2026.

His point is that the code generation models are interchangeable and the review bottleneck is not.

Is review really 10x slower than writing code?

The clip gives no source for the "order of magnitude" claim, and I found none. Anthropic's figures (7.5 average findings on large pull requests, and substantive comments rising from 16 to 54 percent) measure how much review work exists. They do not measure review hours against build hours.

Bottom line

Lemkin is right that review, not generation, is the slow part of shipping AI-written code. The category is already contested: Cursor, which sold for $60 billion, bought Graphite in December 2025, and Anthropic's Code Review was generally available before the clip was posted. Treat the 10x time ratio as his opinion.

Sources

Published October 23, 2026. Based on a TikTok by @20vc_tok. Full video notes (markdown)