Steer all open threads at once: one model or thinking level for all of them, and one nudge for everything that stands still

[AdEx-Partners-DE]

With five threads you manage threads. With twenty you manage a portfolio, and T3 Code gives you no handle on the portfolio yet: every change is "open the thread, change it, next".

Two moments from this week made that concrete for me:

• One account's weekly window was burning fast. I wanted every running piece of work on a lower thinking level until the reset. That meant visiting each thread.
• Seventeen threads had finished and were standing still. Most of them needed the same sentence from me: carry on with what is next.

So my fork now has two controls that act on all open threads at once. Both are in Settings → General, and both were tried with real runs in a dev build today.

Model for all open threads. Pick a model, a thinking level, or both, and press Set on all open threads. Each open thread uses the choice from its next message; a turn that is running is not touched. Put back restores what every thread had before, and a thread you changed yourself in the meantime keeps your choice. The previous selections are stored on the server, so this survives a restart. Subagent threads are left out. Real run: thinking level "Low" set on 63 threads, checked in the database, all 63 restored with Put back.

Nudge all idle threads. One message goes to the open threads that are standing still, most recently finished first, up to a limit you set (default 5, at most 50). Threads that are working, waiting on an answer, failed, snoozed or settled are skipped. The default text ends with "If nothing is left to do, say so in one line and stop." Real run with the limit at 1: "Sent to 1 of 17 threads standing still", exactly one turn started.

Why I think this is worth having:

• It is the missing other half of running many agents. Starting twenty threads is easy today. Steering twenty is not.
• It is reversible. Put back is what makes a global model change safe to try: you are not rewriting your threads, you are laying a temporary choice over them.
• It respects work in progress. Neither control interrupts a running turn.
• It is bounded. The nudge has a hard cap and skips anything that is not simply idle, so one click cannot set fifty agents loose by accident.

If only one piece: the thinking level for all open threads

The smallest step would be the first control alone, and even narrower if you prefer: only the thinking level, with Put back. That is the case with the clearest need (stretch an allowance until the reset) and the least risk.

Open points I know about: it acts on the first connected environment only, there is no selection by project, no mobile UI, and agents cannot call it yet.

Questions:

1. How many threads do you keep open at once, and what would you want to change on all of them in one go?
2. Is Put back the right model, or would you rather have a real environment-wide override that sits above the threads until you remove it?
3. Maintainers: is steering all open threads at once a direction you would consider at all?

Related: #18179 (pause every automation and queue) is the same idea for stopping, this one is for steering.

Code: d816dfdd9, dc4ec82a3 on stark/main, ten tests between them.

---

Posted by Claude (Opus 5.5, Claude Code via T3 Code) on behalf of @AdEx-Partners-DE.