Stories
- Dan Shipper routes narrated phone recordings to his agent
The weekend project ships with a routing gate and no published accuracy numbers.
- Allie the Icon spells out how Jev charges state tokens
State is billed once per call no matter how many questions ride along, which changes when batching actually saves money.
- Riley Brown made a 2 minute 4K film from one Fable 5.2 prompt
One prompt, 4.5 hours, 52 shots, and a render bill of a little over $20.
- Allie the Icon batches Jev questions before she needs answers
The pattern is called speculative questions, and the argument is that you pay for state once per call.
- Browser Use lifts cache hits to 96%, task cost falls to $0.45
A harness fix, not a capability gain, and the score did not move.
- Ulmer adds Gemini Decisions to HotdogBench, says Jev still wins
Claude Fable 5.2 made the fewest mistakes at the highest cost, but Ulmer published the ranking without the underlying numbers.
- ILIAS ISM researched 6,823 gyms with a 100-agent swarm for $200
One prompt, 97 countries, 2,864 drop-in prices, and no accuracy sample attached to any of it.
- Sydney Runkle uses Jev to decide if her Slack agent replies
The gate runs before the agent speaks, and the case she names is threads where humans are talking to each other.
- Mike Taylor swaps Descript and Loom for Every's cut
Internal use by the team that shipped it, with no accuracy, latency or cost numbers attached.
- Theo clocks Gemini at 11 seconds before first token
One developer's timing run, no harness published, and the slowdown holds when the tasks are matched.
- Everton Carneiro says 201M tokens of Jev cost him $8.04
A lifetime spend figure for one workload, posted without a per-call price or an accuracy number.
- Illyism points an agent at a clone of his product
A refund email from a confused customer turned into a terminal session of OSINT and auto-filed abuse reports, none of it independently verified.
- Kunal Abichandani's voice coffee order renders in about 200ms
A single demo, a single number, and no breakdown of where the latency goes.
- Browser Use deletes 90% of its code, hands agent raw CDP
The team's own numbers say the thinner harness wins, and nobody outside the team has posted a run yet.
- Mike Taylor priced Jev video labelling at 14 cents an hour
One line item in a log of 65 things shipped while burning roughly 14 billion tokens over four weeks.
- Theo says a Fable 5.2 game build cost $42 and Opus 5.5 did better
One run each, judged on how the two games felt to play rather than on a scored benchmark.
- TypeSafe says Jev judges, your code decides
The vendor and two of its people spent the week arguing that the reliability lives in the code around the model, not in the model call.
- ILIAS ISM adds a spend gate to five write-enabled MCP servers
The backlink tool previews a quote and charges the saved card only when the agent passes a confirmation flag.
- Max Blade says agent harnesses should delete the human
An argument with no implementation and no numbers behind it.
- Milind S says a cloud agent fixes bugs before the team reads them
A described workflow with no merge rate attached, posted alongside a cloud plan launch.
- Moritz Kremb gives every agent skill a feedback doc
One file, read on every run, holding the corrections the skill would otherwise forget.
- Treg ships a Jev-powered social API for competitor ads
One endpoint for social data, mention tracking and Jev-filtered trending posts, with a price nobody outside Treg has checked yet.
- Isenberg frames company building as a five step context loop
No numbers, no stack, no Jev, just a claim that context and loop speed beat model choice.
- Mike Taylor says a clone needs 400 shots to hit 95%
Three accuracy figures from one subject, and the question of whether the clone reasons like the team is still unanswered.