Jev Daily

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.

The framing that spread this week started with Allie the Icon, who argued that treating Jev's questions as decisions puts the model at the wrong layer of abstraction and that your reliability will suffer for it. That post is earlier coverage here. What is new is that the argument picked up two more voices from the same side of the table.

Diogo Almeida, replying to the post, said one of the biggest anti-patterns he sees with Jev is plugging it in to make Big Decisions instead of having code do what code is best at and using Jev only for the parts that require judgment. The TypeSafe AI account then restated the whole thing under the name System One, with the position that reliability requires code and that reliable decisions come from composing reliable judgments in code.

What the argument actually asks you to change

The prescription is narrow. TypeSafe's post says to be clear about what you need to know, and be clear that a plain reading of your data is enough to know it. The implied failure mode is the opposite habit, which the account describes as smuggling semantic logic into string caveats, embedding vectors and whatever other ML you have lying around, on the theory that data distributions will always be more ridiculous than you can be clever.

For people building on it, note what this is and is not. Nobody in this exchange posted a number. There is no eval, no error rate on a workload, no before and after on a pipeline that was refactored from one big model call into judgments plus branching code. It is the vendor describing the intended shape of a Jev integration, plus one person agreeing from experience, and the reliability claim at the center of it is asserted rather than measured.

If you are currently handing Jev a whole decision and reading the answer as final, the argument is worth a read before your next eval run, because the cost of being wrong here is a rewrite of your control flow, not a prompt tweak.

TypeSafe AI
@typesafeai
X
reliability requires code, and so reliable decisions are made by composing reliable judgments in code
Nov 6, 2026 · View on X
Diogo Almeida
@CompleteSkeptic
X
one of the biggest anti-patterns we see with jev is just plugging it in to make Big Decisions instead of having code do what it's best at
Nov 6, 2026 · View on X

Know what Jev is good at before you bet on it

One email a day. Under three minutes. What builders found out about Jev yesterday.

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.