Stories, page 5
- Shumer says local models are useless, Theo writes off most labs
Two blunt calls on model quality, and not a benchmark or a dollar figure between them.
- Steinberger's test cleanup prompt names a 20% target
The instruction that did the work was a number, not a request to tidy up.
- Opus 5.5 built a 277k-gate computer in one 30-hour run
One prompt, no supervision, and a machine you can actually run in the browser.
- Codex user says his $200 plan now lasts two days
calle bought a $200 Claude plan out of frustration and says Opus 5.5 feels like five times the usage.
- Rauch says AI slop risks people discounting reading itself
The Vercel CEO says the damage from slop is not just bad code, it is that nobody trusts text anymore.
- Steinberger says skip the Mac mini, rent a VPS instead
A four line reply turns an always-on agent setup from a hardware purchase into a hosting decision.
- Isenberg says humans now err more than frontier agents
One line, no benchmark, and a whole lot of founders quietly staffing as if it were already true.
- Marc Lou builds revenue-gated chats that unlock as you earn
Entry is verified revenue, and the groups open up as your last 30 days of income grows.
- Rauch says Anthropic fell from 69% to 40% of Vercel spend
Two months of gateway spend data shows developer dollars moving fast, and open models taking about half of the difference.
- Tibo says a product flat at $2k MRR for six months hit $15k
Six months of nothing, then one month of what he calls insane growth, from a founder who says he does $1m a month across his products.
- Opus 5.5 cut Arvid Kahl's test suite to 6m27s
He says the model found an inactive cache setting in his PHP runtime and refactored several thousand serial tests to run in parallel.
- Grok 4.7 hits 94% on Next.js evals at 2x to 7x cheaper
Three models tie at the top, and the interesting number is the one underneath them.
- Opus 5.5 ported HAProxy to Rust in 9.5 hours, Anthropic says
Anthropic's own head to head has the new model finishing a real port faster and cheaper, which is exactly why it needs an outside run.
- Josh Pigford lists where Jev runs inside two live apps
The pattern is cheap verification calls wrapped around expensive models, not a model swap.
- Engineer says his team ships Claude output nobody reads
A first-person account of what happens when shipping speed becomes the only metric a team is measured on.
- 37signals hired Rob Zolkos, who never applied
Jason Fried says the work happened on 37signals' own product, with no job posting and no application.
- Anthropic merges Claude Cowork and chat into one Claude
Hand over a task, close the laptop, and Claude keeps going, with Docs, Slides and Design now inside every conversation.
- Jev is Vercel's fastest adopted model ever
Vercel put a number on the rush, and Guillermo Rauch thinks it says more about pricing than about the model.
- Marc Lou says one sponsor deal took TrustMRR to $26,031 MRR
He says the client wanted all 20 slots and was willing to spend $30k a month.
- Theo tears down Jev instant compaction in six points
A cheap-model trick for shrinking agent context runs into how caching and reasoning traces actually work.
- Lemkin says use the cheap model additively, never as a filter
The cheap model sweeps work nobody was running, and the expensive model still makes the call.
- Lemkin says Jev judged 600 candidates for $0.064 and still failed
Roughly 90x cheaper on identical inputs, and still not usable for SaaStr Connect.
- Rauch says the best part of the AI wave is everyone wants to build
One line, no numbers, and a claim about what people want from software right now.
- Marc Lou says 912 people finished his sponsor puzzle game
A sponsor handed him $30,000 in API credits, so he built a 15 minute game that ends in an email address.