Simon Willison — 2026-07-20#

Highlight#

Today’s standout post explores how AI coding agents have fundamentally altered the return on investment for reverse-engineering undocumented APIs. By drastically reducing the cost and effort of writing code, agents remove the psychological burden of throwing away or maintaining fragile custom integrations for smart home devices.

Posts#

Reverse-engineering is cheap now · Source Simon reflects on anecdotes of developers using coding agents to reverse-engineer and automate smart home devices. Previously, the high effort and future maintenance burden of undocumented APIs made these integrations hard to justify. Now, because AI drops the cost of trying, failing, and writing code so significantly, the psychological baggage of abandoning or maintaining fragile code is largely eliminated.

Who’s Afraid of Chinese Models? · Source Highlighting a piece by Ben Thompson, Simon notes a compelling proposal to address the hypocrisy of major AI labs outlawing model distillation while training on unlicensed data. Thompson suggests the U.S. should lean into a new copyright policy that explicitly legalizes training data collection as fair use, while barring terms of service that forbid distillation for U.S. companies. The post also points out that Alibaba’s recent pivot to release Qwen 3.8 Max as open weights may have been spurred by a recent speech from Xi Jinping encouraging open-source collaboration.

Quoting Sam Altman · Source Simon highlights a revealing October 2022 email from Sam Altman to the OpenAI board, which surfaced during the 2026 Musk v. Altman lawsuit. In the message, Altman discusses releasing a local GPT-3 equivalent model to preempt competitors like Stability. The stated goal was an aggressive strategic move to discourage others from releasing similarly powerful models and to make it harder for new competitive efforts to secure funding.

Project Pulse#

Today’s updates share a common thread exploring the shifting strategic landscape of open-weights AI, touching on corporate moats, international competition, and the practical impacts of LLM coding agents on daily developer workflows.