Kelley Ignites the Bun Drama, Codex Becomes ChatGPT & GPT-5.6's Price Shock

The Zig Post Fallout

Yesterday Jarred Sumner shipped the "Rewriting Bun in Rust" post; today Zig creator Andrew Kelley published his response — "My thoughts on Bun's Rust rewrite" — and it became the day's angriest thread-storm.

  • Theo read it live on stream and detonated (post, 83K views): "I've never been so pissed off on camera. Andrew is a petty, evil man, and you should not trust any software he writes." Pushed on whether that's a stretch, he doubled down: "If he goes after his users in such a vile way, I don't think you should use his software. He is a ticking time bomb." The replies carry useful context: the resentment reads as being about Jarred taking the VC path, Zig banning AI contributions (Bun was Zig's flagship project and its biggest donor — reportedly $60k/yr, now ended), and the post repeatedly questioning Jarred's engineering judgment. Plenty of pushback too, from "petty sure, evil?? why?" to a pointed callback to Theo's own ffmpeg-maintainer bounty episode.

  • Charlie Marsh's cooler-headed version (post, RT'd by theo): "I find this post to be really disappointing... It's explicitly framed as 'not a personal criticism', then goes on to include a bunch of very personal criticisms about Jarred."

  • Armin Ronacher, notably not a Bun fan, defended Jarred (post): "what @jarredsumner is doing is pretty fucking awesome and I can confirm that he's great to hang out with... The responses to his blog post by some people are uncalled for." Best replies: Jamie Turner's "It's reprehensible and condescending. Call me crazy, but if I lose my biggest customer, I'm gonna consider it my fault", and Seb Insua calling it "one of the most poisonous posts I've ever seen... It's sad that some people's reaction to AI makes them treat other humans like this" — because underneath the personal feud, this is really a proxy war over AI-assisted rewrites (Zig forbids AI contributions; Jarred used Claude to port the entire runtime).

  • The contrarian minority is worth reading too: Ben Vinegar's "we've gone soft" (72 likes) and a reply collecting the Zig side's actual grievances — correctness advice ignored, and AI-generated changes that allegedly made the Zig compiler non-deterministic.

Codex Becomes ChatGPT

The other half of Wednesday's OpenAI launch — the Codex desktop app being folded into a unified "ChatGPT" app alongside the new ChatGPT Work — aged badly overnight.

  • Theo: "turning Codex into ChatGPT Desktop is a generational fumble" (post, 304K views, 4.4K likes, 483 replies). The replies are a catalog of migration pain: the Codex app vanishing from people's Macs, accounts not rolled over so threads and workspaces are unreachable, Shift+Tab planning mode gone, and kitze summarizing the brand damage as "aura loss of biblical proportions". Recurring take: "It is Cowork/Code all over again" — several people note OpenAI copied the one Anthropic decision devs disliked. OpenAI's dkundel offered the defense: "We just made this setting a dropdown. It's the same Codex".

  • Simon Willison speaks for everyone (post): "I am so confused by ChatGPT v. ChatGPT Codex v. ChatGPT Work v. Claude v. Claude Code v. Claude Cowork right now!" He read OpenAI's own help page on the split and reported back: "This did not help".

  • The contrarian case, from Jerry Liu (post): he actually tested both and prefers the ChatGPT Work/Codex split to Claude Cowork/Code — the interface is more unified, chat history is shared between modes, and "the team put more care into the specific toggles that should feel different." He'd still rather OpenAI went all the way to one app surface for all work.

  • Theo's bigger lament (post): "Codex just rebranded to ChatGPT. Claude is spamming me with Excel plugins. xAI is bragging about how good Grok is at PowerPoint. I knew the 'AI labs focus 100% of their effort on software devs' thing wouldn't last, but it's sad having it end all at once." Corey Quinn adds the money angle (post): nobody can pin down when GPT-5.6's price jumps 50% or when it stops being included in subscriptions — docs, blog posts, and employees all disagree.

  • Practical scrap for the migration: keep a "Codex" alias so Spotlight/Raycast searches still work.

GPT-5.6 Day Two — Benchmarks & Economics

The launch-day dust settled into numbers, and the story everyone converged on is price-performance, not raw capability.

Claude Code & Anthropic Updates

Agentic Coding & Craft

Quick Hits

  • swyx on AI-era SEO (post): "whoever does AEO for @resend needs to get a raise, all the leading frontier models keep trying to use resend for emails even when I have existing transactional infrastructure already set up." Answer-engine optimization — getting your product to be the one models reach for — is quietly becoming a growth channel.

  • Codex housekeeping trick (LLMJunky): ask Codex to clean up your stale threads — traditional automations reuse the same title every run, so they're easy for the model to identify and sweep.

  • GPT-Live API design partners wanted: Greg Brockman is soliciting design partners for OpenAI's GPT-Live API — creative new apps or integrations into existing products.

  • Cognition stakes out its lane (Jared Zoneraich, RT'd by LLMJunky): "Cognition only cares about software engineering. We're not trying to build an AI God... just a really great AI Software Engineer" — a pointed contrast on the same day OpenAI pivoted its dev tool toward general work.

  • Compression claim to watch (post): a purported breakthrough running models "losslessly" on 13% of the usual hardware, with calls for someone to validate it with a full terminal-bench-2.1 trace. Extraordinary claims, no trace yet.

  • The Theo–Charlie Grahn saga continues (post): Theo reposted the year-old video that's still drawing takedown attempts, then reported another copyright strike with identical content and timestamps — "Despite being informed that these are invalid and malicious. Willful ignorance. Probably illegal."