Best of your X follows: May 19–27

Gemini 3.5 Flash is now generally available — Google skipped the usual preview phase and shipped Gemini 3.5 Flash straight to GA at Google I/O on May 19. The model is priced higher than its predecessor but Google is deploying it across a wide range of their core products. Simon Willison notes it arrives with updated tooling support (llm-gemini 0.32) and tested it by generating a pelican riding a bicycle.

OpenAI and Anthropic have both found product-market fit — and it's enterprise coding agents. Simon Willison's May 27 analysis draws on concrete numbers: he personally consumed $2,180 in API tokens over 30 days while paying just $200 for his subscriptions, but enterprise customers no longer get that cushion. Both companies quietly switched enterprise pricing to raw API rates (Anthropic in November 2025, OpenAI in April 2026), while simultaneously releasing pricier frontier models (GPT-5.5 at 2× its predecessor, Opus 4.7 at ~1.4×). Uber blew through its full-year AI budget months early; Microsoft cancelled Claude Code licenses to push its own Copilot CLI. Willison reads these not as failure stories but as the classic "suck air through their teeth, then say yes" pricing signal.

Anthropic is paying SpaceX $1.25 billion per month for compute. The figure surfaced in SpaceX's S-1 filing: Anthropic signed Cloud Services Agreements covering access to both Colossus and Colossus II data centers through May 2029. The deal appears to be for inference capacity, not training — consistent with Anthropic announcing higher usage limits for Claude Code and the Claude API alongside the announcement. The data centers have a troubled environmental record; the xAI side also added a clause allowing Elon Musk to revoke access unilaterally if the AI is deemed harmful to humanity.

Datasette Agent launched: Simon Willison's LLM library and Datasette finally merged. The May 21 launch brings an extensible AI assistant directly into Datasette, letting users run natural-language queries against their data via the agent.datasette.io interface. Plugins already exist for sandboxed code execution (Fly Sprites), chart rendering with SQL transparency, and per-user LLM usage limits. Willison described it as the moment three years of LLM library work converged with three years of Datasette work.

SQLite officially banned AI-generated code — then tightened the ban. A new AGENTS.md file appeared in the SQLite repo on May 22, aimed at agents pointed at the codebase. It accepts AI-generated bug reports with reproducible test cases and proof-of-concept patch PRs, but explicitly rejects agentic code submissions. Five days later, D. Richard Hipp removed the word "currently" from the rule, making it unconditional. The SQLite bug forum was also split off into a separate venue to manage the flood of AI-generated reports.

The Vatican published an AI ethics encyclical. Pope Leo XIV's Magnifica Humanitas, released May 15 and analyzed by Willison on May 25, addresses safeguarding the human person in the age of AI. Willison called it "some of the clearest writing I've seen on the ethics of integrating AI into modern society" — notably without hedging. The document covers both technical risks and social effects.