Fetching from the wire…
01
02
03
04
05
06
07
08
Top 5 · 2026-07-11 · source-backed
This is the case study I'll be pointing people to for months. Simon Willison shipped sqlite-utils 4.0rc2 on July 5, his first major version bump since 2020, with most of the code written by "Claude Fable" across 37 prompts and 34 commits, for about $149.25 total.
The number that made me sit up wasn't the cost. It was that Fable surfaced 5 issues Willison categorized as release blockers that he hadn't hit himself, including a silent transaction bug. On a backwards-compatibility-sensitive major release. From one of the most careful maintainers in the Python ecosystem.
That reframes the whole "agents are for greenfield scaffolding" assumption. Greenfield is the easy case. Nobody's callers break if your weekend prototype has a subtle bug. A SemVer major bump on a library thousands of projects depend on is the opposite, and it's exactly the unglamorous, high-stakes work people assumed agents couldn't touch. Willison proved them on it.
The reusable technique here is worth stealing directly. Before cutting a breaking release, run a single agent task scoped only to "enumerate everything that could break existing callers," completely separate from feature work. Not "review the code." Not "find bugs." Specifically: trace every path where an existing caller's behavior might change. Agents are willing to exhaustively walk edge cases a human maintainer skims past on the fiftieth read of their own code. That's the actual edge, and it's the opposite of the "vibe coding" reputation. This is agentic engineering with a review gate, and it caught a transaction bug that would've silently corrupted data.
$149 for a major release with a dedicated correctness audit built in. I've paid more for worse code reviews from humans. If you maintain anything with a public API, add a scoped "find the release blockers" pass to your release checklist before the next breaking change. It's the highest-leverage 20 minutes of agent time I've seen described this week.
Each link below shares sources, entities, or timing with this story.
Anthropic shipped Fable 5 on June 9. Willison spent ~5.5 hours stress-testing it: slow and expensive, but it handled everything he threw at it, including agentic coding. (Simon Willison) The tell that it's a real working model and not a benchmark queen: because it post-dated A...
He handed Claude Fable 5, in Claude Code for web, the job of evaluating whether smolvm 1.8.3 could safely run untrusted Python and JavaScript under strict constraints. Verdict: it fits, with hardware-isolated VMs rather than shared-kernel containers, enforced CPU/RAM limits, g...
Willison's June 13 post walks through publishing WebAssembly wheels to PyPI so Python packages run client-side under Pyodide, alongside Datasette 1.0a33 adding JSON "extras" to the API (Simon Willison). He built a throwaway extras-API explorer with Claude Fable 5 to understand...
Willison published his AI Engineer World's Fair conversation with Anthropic's Cat Wu and Thariq Shihipar, covering Claude Code, Claude Tag, and Fable. Primary-source practitioner conversation with the people who actually build the thing, rather than secondary coverage of a pre...
On June 6, Willison released micropython-wasm, an alpha package that runs untrusted Python inside a WebAssembly binary with enforced memory, CPU, and file/network limits. He wired it into a Datasette Agent plugin, put GPT-5.5 xhigh inside, and challenged it to break out. So fa...
Simon Willison has been writing software for over 25 years. He's one of the most disciplined, transparent engineers in the Python ecosystem. And yesterday he published an essay admitting he no longer reviews every line of code that Claude Code generates for his production proj...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.