Stories, page 2
- 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.
- Anthropic rewrote its Messages API in Rust with no Rust devs
Somi's summary of the Latent Space episode says humans owned the plan and the verification, and the model owned the code.
- Claude drops the 5-hour limit for verified Ultra accounts
Max 20x becomes Ultra, Max 5x becomes Max, and the prices stay where they were.
- Theo's computer use bench puts macOS last at 27%
An early, unpublished benchmark measures the desktop the agent drives rather than the model driving it.
- Zechner batches commits, 5 to 439 agent runs per second
A durability write queue, not the model, was capping 128 parallel conversations in pi-durable.
- Vercel lets your agent file its own bug reports
A new submit_feedback tool in Vercel MCP turns agent frustration into a support ticket, with structured context attached.
- Thariq says coders feel useless as non-coders envy them
An assertion about mood on both sides of the agent divide, with no measurement behind it.
- Theo needles open weight fans over a benchmark re-ranking
A leaderboard reshuffle becomes a punchline, and nobody involved ran any code.
- Utah ran a Claude Code pilot with 34 state developers
Anthropic's Boris Cherny posted the pilot numbers as the company expanded its GSA OneGov agreement.
- Malte Ubl ran 24 DeepSeek sub agents under Opus 5.5 for $9
A frontier model doing the orchestration, cheap open weights doing what he says is 99 percent of the work.
- Theo says a $700 Linux PC beats his $12,000 Mac
A T3 Code visualization comparing the two machines, with no published timings.
- Theo drives 8 T3 Code threads across machines from one MCP server
The feature is behind a labs toggle on the nightly build, and Theo says it is still very early.
- Pocock splits one-way-door PRs into a revertible half
His prompt of the day turns an irreversible change into an expand/contract pair you can test in production first.
- Theo says Grok 4.7 should not be used
A verdict with a link attached, no test, no repro, and a swipe at day-one reviewers.
- Theo's tool eval cuts T3 Code context 79%
A week of eval building and $1,500 in tokens to fix tool schemas nobody was measuring.
- Mario Zechner's agent workflow is two sentences long
The reply was a joke about prompt scaffolding, not a benchmark, but it landed in a week full of elaborate agent machinery.
- Theo ends the week calling Opus 5.5 the greatest model ever
The nerf argument closes out with one side on feel and the other on a benchmark Theo himself called the worst of all time.
- Steinberger says Astra running codex cli in worktrees beat subagents
The two week migration loop finished, and the setup detail is the part worth copying.
- Theo says Gemini hallucinated in over 80% of his review runs
One engineer's harness, one set of receipts, and a failure mode that looks exactly like a real code review.
- Theo pledges one meaningful T3 Code improvement every day
The search rebuild that ran 18 hours straight is now day 21 of a shipping streak he says he will keep up as long as he can.
- Anthropic tells companies to delete subagents
One agent with skills beat the multi-agent setup, and the advice now is to fold subagents back into the main loop.
- Theo says Anthropic's Agents SDK burns 5-hour limits faster
One user emptied a full 5-hour window in about 30 minutes, and Theo thinks a recent limit boost never reached the SDK.
- dax has opencode post progress to a Slack thread
One instruction turns the agent's status into something the team can check instead of something one person relays.
- Sumner runs Claude Code on a Bun AOT build at 267 MB
The Bun author is dogfooding ahead-of-time compiled JavaScript inside his own agent harness, and reporting the memory numbers from Activity Monitor.