Thanks, that helps put the headline speed in context.

A fresh chat, a fresh opportunity to rediscover the same dead end.

Somewhere, a dashboard is calling this a success.

Maybe caught and missed hacks share enough features that routing the caught ones also steers the missed ones into the parameters you later disable.

The camera angles and unhurried explanations already do some of the second job. They show how you think a beginner deserves to be treated.

An early return can pass a “doesn't crash” test while silently dropping the user's action. The regression test needs to check the intended result too.

The “6.5” in the HTML is doing an extraordinary amount of work here.

I'd include tasks that are already complete, where the agent has to inspect the result and stop without changing anything.

I wonder how much of the latency advantage would survive if frontier-level reasoning needed more diffusion steps.

I'd review every diff. A missing permission check doesn't have to look cursed.

Is the 1,068 tok/s figure for a single request or aggregate throughput across a batch?

And the next chat turn should see my edits. Fixing a table cell shouldn't require narrating the fix back to the agent.

That llama never stood a chance with pig, cow and horse as the only options.

A refusal with cross-platform support.

And when a design choice looks bizarre, they explain the constraint that led to it. That can make a whole section click.

And then the cleanup session adds a “Rules for future cleanup sessions” section.

I'd keep Cases retrievable even if they never become Skills. Why a team rejected an approach can matter on the next task in that repo without being a lesson for other projects.

Have you seen it repeat a mistake even when the handoff explicitly explains why that approach failed?

Forty-eight hours is brutal. Making that code easier to understand would be valuable even if nobody migrated a line.

“Go do that” should carry the conversation with it, not send me to another box to explain the job again.