What to notice in the first response

The content of the first response matters less, for getting a feel for tCode, than the order things happen in: a read before an edit, a search before a broader change, a test run after a fix rather than before it.

This ordering is not a fixed script applied to every request; it comes directly from the model reasoning about what it actually needs to know before acting responsibly on your specific request. A one-line fix and a broad refactor both follow the same underlying instinct, they just apply it at very different scales.

It is the kind of detail that is easy to consciously notice once, in a first session, and then stop noticing entirely once it becomes the expected baseline for how tCode behaves in every session afterward.

This same ordering instinct is also the fastest way to spot when something has genuinely gone wrong versus when tCode is simply being thorough. A turn that reads several files before proposing anything is usually being careful, not stuck; one that proposes an edit with no investigation at all on a request that clearly needed some is worth a second look.

See The chat loop for the full mechanics behind why this ordering happens the way it does, driven by the model's own reasoning rather than a fixed script.