Theo says GPT-6.2 Astra stalls on questions, burns the task
An analogy about a waiter asking how to cook the fries, and nothing you can check against a log.
Theo posted a complaint about GPT-6.2 Astra in the form of a restaurant story. You order a steak medium rare with fries. The waiter asks whether you want the fries medium rare too. You tell him to cook them normally. Fifteen minutes later he comes back in a panic because the whole kitchen is held up until you decide whether the fries go at 413 degrees or 471 degrees. You tell him to do whatever the chef wants. Thirty minutes after that the food arrives.
"The waiter comes out with mostly passable fries, and an entirely burnt steak," Theo writes. The chef, the waiter explains, got distracted waiting for the answer about the fries. The offer is to put the steak in the fridge to cool it off. Theo closes with one line, that this is how it feels to use GPT-6.2 Astra.
The waiter comes out with mostly passable fries, and an entirely burnt steak
What is actually in the post
Nothing you can check. The post contains no transcript, no task, no model settings, no timings and no comparison run. The two numbers in it are oven temperatures in a joke about fries. It is a description of a feeling about an agent harness, written by someone who clearly uses it, and that is the whole of the evidence.
The failure mode he is pointing at is specific enough to be worth naming. An agent that interrupts a well specified task to ask a clarifying question about a detail the user does not care about, blocks on the answer, and then delivers the part you actually asked for in worse shape than if it had never asked. The clarifying question is usually sold as a safety feature. Here it is the thing that eats the attention budget.
For people building on it, treat this as a symptom to go looking for in your own logs rather than a result. If your harness lets the model stop and ask, check what the waiting costs the primary task, and check it against a run where the model is told to pick a default and keep going. One person's vibe on one harness is not a benchmark, and until someone posts a transcript there is nothing here to reproduce.
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.
