Jev Daily

Rivet split one Jev question into six

Nathan Flurry says Rivet asked Jev for a ticket's severity in one question and got mediocre results, then got a working classifier by asking three yes/no questions instead. The rule he landed on: Jev does not reason, so your code does the ranking.

Built with it

Rivet says its own support stack resolves hard tickets in 5h 1m

Nathan Flurry says Rivet built its own support system on Rivet plus Jev, and that time to resolution for hard engineering tickets dropped to 5h 1m. He says support is still staffed entirely by human engineers, with the models assisting. Rivet had earlier been running support on Plain, which Flurry called the best they had tried but not there yet.

Nathan Flurry 🔩
@NathanFlurry
X
100% human engineers doing support, with LLM assistance
Nov 18, 2026 · View on X

Read the full story

Writing criteria

Rivet replaced one severity question with three yes/no ones

Nathan Flurry says Rivet first asked Jev for a ticket's severity in a single question with SEV1 to SEV4 as outputs, and got mediocre results. Splitting it into three yes/no questions, is_down, is_blocked and is_broken, with the ranking done in plain if-statements, worked.

Nathan Flurry 🔩
@NathanFlurry
X
jev does not reason
Nov 18, 2026 · View on X

Read the full story

The case against

Max Blade says Gemini 4 Neon is completely broken in his agent CLI

Max Blade says Gemini 4 Neon is completely broken in agy cli and acp, posting what it produced for a 3d printable fidget spinner and a lawn mow simulator after multiple follow up prompts. Greg Isenberg, who tested it early, says it scored about the same as Opus 5.5 on coding for around half the cost per task.

GREG ISENBERG
@gregisenberg
X
It scored about the same as Opus 5.5 on coding for around half the cost per task!
Nov 19, 2026 · View on X

Read the full story

Get the next one by email

Jev, read daily so you do not have to. The builds, the benchmarks, the criteria that worked and the cases where it lost, from the people shipping on TypeSafe AI's System One model.