Jev Daily

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.

Allie the Icon published a Jev pattern she calls speculative questions, which is batching questions into a single call even when you do not yet know whether you will need the answers. The usual instinct is the opposite, ask only what the current branch requires and ask again later if the logic takes you somewhere else. She argues that instinct costs you both time and money.

Allie the Icon
@allietheicon
X
Questions never affect the output of other questions in the same call, so accuracy will not suffer.
Nov 13, 2026 · View on X
Allie the Icon
@allietheicon
X
When your state is large, batching questions often saves cost compared to sequential calls that duplicate state.
Nov 13, 2026 · View on X

The speed argument is the easy one. If the questions are all in one call, you often finish the task in that call rather than running multiple sequential calls down different logic paths, each waiting on the one before it.

The cost argument is about state

The cost argument is less obvious. By her account you pay for tokens in state once per call, so batching avoids paying for that same state again on the next call. Sequential calls in different logic paths duplicate it. Her framing is explicit that this is a function of size, and that batching often saves cost compared with sequential calls when the state is large.

That is the line worth reading twice before you refactor. The saving comes from amortising state across questions, so the smaller your state, the less there is to amortise. A handful of questions over a thin state split across two branches is not obviously cheaper batched, and she does not claim it is. She also does not post numbers here, no token counts, no latency figures, no before and after on a workload. This is a pattern recommendation from someone who writes about Jev patterns, not a measured result.

Why the extra questions do not cost you accuracy

The thing that makes the pattern safe, in her telling, is isolation between questions inside a call. Questions never affect each other's output within the same call, so adding ones you may discard does not degrade the ones you keep. That is the claim that turns speculative asking from a gamble into bookkeeping. Without it, stuffing unrelated questions into one call would be the sort of prompt crowding that quietly moves answers around.

TypeSafe AI reposted the pattern with the line "Jev first, figure it out later", which is a vendor amplifying a user, not independent confirmation.

For people building on it, the decision rule is your state size, not your question count. If you are carrying a large context into every call, batching the branches together is the version worth trying. If your state is small and your branches are cheap, the duplication you are avoiding may not be worth the questions you throw away. Allie the Icon has argued before that Jev works best when judgments live in your code, and this pattern sits on the same side of that line.

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.