Stories, page 2
- Isenberg boils his company down to voice notes and inference
A one-line restatement of a stack that is a description of one founder's setup, not a measured result.
- Theo re-ran a coding model on CLI, still calls it garbage
A broken harness inflated the first result, and fixing it moved the number without moving the conclusion.
- Allie the Icon's "shopping cart" pattern for multi-intent queries
Recursively pick the next thing until Jev says you are done, with no latency or cost numbers attached.
- Prism edged Jev by two emails on Arushi Gandhi's 200-email test
A one inbox, 200 item comparison of four decision models, with the winning margin small enough to be noise.
- Kosenko says signing a CLA should be a deny rule
An argument about where agent policy lives, with no numbers attached to it.
- Vogel shelves his Jev reply spam ranker over cost
A working X anti-spam app that scored every reply with Jev got replaced the same day by a public blocklist fed by a Grok bot.
- Allie the Icon's line-by-line search skips chunking entirely
Put the document in state, ask the question once per line, and pick the output type by what kind of answer you want back.
- Lightpanda puts Jev on logins at 0.53s, Haiku 5.5 at 7.83s
The browser company posted timings for its own integration and no accuracy number at all.
- Berman's AI shoppers picked no-discount ads over 50% off
The tally behind the 3,092 ad experiment, and what it cost to run.
- Mike Taylor warns Jev as a judge picks whatever is first
One builder's warning, no numbers attached, and nothing from TypeSafe either way.
- Theo ships T3 Code tool rewrite, 79% less context
The week-long eval is now a shipped change, and Theo says the token bill to get there was $1,500.
- Aditya Umale's cleanup pass strips a named list of AI tells
He says 5 of the last 9 drafts from his writing bot went out without a word changed.
- Greg Isenberg says agents breed cobras when you give them a number
A 19th century bounty story, repurposed as a warning about what happens when you score an agent on throughput.
- Berman ranks nine agents to build, Jev lands at number five
Speed-to-lead first, weekly report last, and the one Jev job on the list is read-only by design.
- Mike Taylor pairs Claude with Jev for napkin math
Claude already gets the answers right from memory, so the pairing buys a visible chain of estimates instead of a better score.
- ILIAS ISM pasted a screenshot, agent closed the ticket and shipped a fix
One screenshot, no prompt text, and an agent that went from Gmail to a production deploy.
- ILIAS ISM deleted 40,000 lines of admin panel code
Nothing got through, and he rebuilt the attack surface anyway.
- Berman's creator vetting prompt sorts on buying intent
One prompt, five scores, forced evidence quotes and a MISSING token instead of a guessed number.
- Mike Taylor calls a skill's short half-life an eval failure
An argument about where agent instructions should live, with no task set behind it yet.
- Shengkun Ye's open source people search quotes $0.0024 a lead
A six provider waterfall that runs 1,000 leads through Jev first and only pays to enrich the ones it keeps.
- AI/ML API ships a Jev tool router, claims 96% cheaper than Claude
One MCP endpoint, 10,000+ tools and models, and a cost number with no workload attached.
- Berman ran 100 AI shoppers over 3,092 Meta ads for $1.59
Of the top 100 ads the simulated shoppers picked, Berman says only two carried a 50% off offer.
- Sentry screens 10M events a day on Jev for $263
The team is building an in-house classifier aimed at matching Jev's accuracy for 90% less, with no accuracy numbers posted yet.
- Treg ships open source people search on Jev at $0.0089 a lead
The price and the benchmark ranking are both the builder's own numbers, and nobody else has run it yet.