Stories
- David K Piano says tell agents what you want, not what to do
A one-line workflow claim that the poster himself flagged as possibly painfully obvious, amplified by Kent C. Dodds.
- Levels used Claude Code to design furniture and order parts
A non-designer, non-coder worked with the agent in the terminal and got back a parts list.
- Theo pins the Opus 5.5 nerf reports on the reasoning summarizer
A per-account change to how reasoning traces are rendered, not a quieter model, is Theo's read on the November 3 shift.
- Opus 5.5 fails twice as many T3 turns on max as on high
Telemetry from real T3 Code sessions, not a benchmark, and the extra failures cluster in the first 30 seconds.
- dax put every subagent on DeepSeek Flash, Opus stays the boss
The model split is cheap. The scoping and proof-of-work rules are the part worth copying.
- Ball says remote coding is solved because you are not coding
Responding to a podcast debate with Armin Ronacher and Ben Vinegar, Thorsten Ball argues agents removed the editor-shaped reasons to stay local.
- Ball says agents cause fewer regressions than humans
Two engineers who already doubt unit tests tell each other the regressions never showed up.
- Wes Bos moved a 307-post WordPress site to Emdash
One reported migration, 307 posts, 15 custom fields, and a Bluehost bill he had been paying for 15 years.
- Cognition ships SWE-2 Flash, swyx says 3x faster at same evals
The claim is distillation with no quality cost, and so far nobody outside Cognition has tested it on a real codebase.
- Levels gets decompiled BF1942 running multiplayer on phones
The agent also upgraded the game to v1.61 and explained, step by step, how it unstretched the HUD in portrait mode.
- Pocock's prompt of the day is "how do we make this change fully verifiable?"
One question, no numbers, and an agent that goes off and clicks through the frontend by itself.
- Ball says watching an agent use GitLab beats a unit test
The verification argument moved from code review to tests, with Thariq arguing smarter models will not need to check their own work at all.
- Anthropic loosens Claude Code's game-copying rule in 2.1.317
The fix landed, the complaint about server-side prompt edits did not go away.
- Willison reproduced the weights trick on a laptop with Qwen 3.8 27B
The buzzy result showed up in a small local model, which is the point he is making.
- GPT-6 Astra edited 0.016% of Qwen3.5-0.8B's weights to fix its math
No gradients, no training, just a frontier model reading and rewriting another network's parameters.
- levelsio decompiled Battlefield 1942 into a web build
A port he abandoned six months ago, redone with Opus 5.5 and shipped with multiplayer.
- Claude Code gets a server-sent rule against cloning games
chasm found the flag, Addy Osmani confirmed Anthropic added two sentences, Peter Steinberger used it to argue for your own harness.
- dax routes architecture questions to a Gemini subagent
One task type goes to a different model inside the session, and the reaction says more about Gemini's reputation than the workflow.
- Karpathy filtered 43M HN comments into a YouTube app in 2 hours
The only work he did by hand was the three steps that live behind a browser login.
- Theo drops Opus 5.5 to high, swyx says go lower
Two shipping engineers now say the maxed reasoning dial is the wrong default, and neither posted an eval.
- Ball pushes back after devs balk at no-branches workflow
The replies surprised him, so he answered the two objections that kept coming back.
- Ronacher says agents can write their own test suite
A disagreement about whether an existing conformance suite is required, or just helpful.
- Steve Ruiz's prompt hunts implicit state machines
One line, no code written, and one engineer came back with a number he did not like.
- Theo says Argon 4 is 10x worse, contradicting ThePrimeagen
Two engineers with their own harnesses reached opposite verdicts on the same model in a single day, and neither posted the evidence.