Andrew Kelley didn’t mince words. The creator of the Zig programming language spent this week publicly grading the Rust code that Anthropic-owned Bun produced when it rewrote its JavaScript runtime from Zig to Rust in eleven days using a fleet of Claude agents running in parallel, at a reported cost of roughly $165,000 in API calls. His verdict: “unreviewed slop.” Kelley’s real complaint isn’t Rust versus Zig — it’s that a team let AI agents generate more code than any human could plausibly review, then shipped it anyway. Anthropic pushed back, but the underlying question, of who actually checks the checker’s work, isn’t going away now that AI-written code is the default rather than the novelty.
Google has the opposite problem this week: not too much confidence in a model, but not enough. Gemini 3.5 Pro, already weeks late, is now targeting a July 17 release after DeepMind reportedly scrapped its 2.5 Pro-derived base entirely and rebuilt it from scratch, following what sources describe as structural failures in recursive tool-calling and SVG generation. That’s an unusually candid admission for a lab racing OpenAI and Anthropic on release cadence — better to eat the delay than ship the failure mode. The timing is pointed: July 17 is also the opening day of Shanghai’s World AI Conference, where Xi Jinping will deliver the keynote in person for the first time since the event launched in 2018, with Beijing expected to use the stage to push a Shanghai-headquartered World AI Cooperation Organization. Three of the year’s most-watched model and policy events are now converging on the same week; Google would rather not walk into it with a broken product in hand.
Meanwhile the Apple-OpenAI trade secrets fight is intensifying rather than settling down. Apple’s suit accuses OpenAI of instructing hardware hires to bring confidential design documents on their way out the door and evade Apple’s own security procedures; this week’s filings suggest both sides are digging in for a long fight rather than a quiet settlement. It hasn’t stopped Sam Altman’s separate, more personal feud with Elon Musk from running in parallel — the two have spent the week trading jabs on X even as OpenAI tries to keep its hardware ambitions and its legal exposure as two different stories, which they increasingly aren’t.
None of these are new problems, exactly. AI coding tools have always needed review, model launches have always had deadlines, and OpenAI’s rivals have been suing and sniping at it for years. What’s different is the scale of what’s now riding on getting it wrong: a rewrite that touches a JavaScript runtime millions of developers depend on, a model launch timed against a geopolitical showcase, a lawsuit that could shape whether OpenAI ever ships hardware of its own. The industry spent the first half of 2026 selling confidence. This week, in four unrelated rooms, that confidence got checked.
