Update
Plan-patch telemetry, a sixth critic rule, and better dashboard delivery
TL;DR
The agent can patch its own plan while a task runs. Until now, whether that patch actually helped was invisible: a rewrite that failed looked the same as one that worked. Outcomes are now recorded, a sixth critic rule was added off the back of what that telemetry showed, and dashboard delivery quality improved alongside it.
What shipped
- Plan-patch outcome telemetry — when the agent rewrites its plan mid-run, whether the patch applied and what happened next is recorded rather than lost.
- A sixth critic rule, added because the telemetry showed a specific failure the existing five didn't catch.
- Dashboard delivery quality improvements on the reporting lane.
- Account routing by goal — the clarifier now ranks accounts named in the goal itself, so mentioning an account by name in your request actually reaches it.
- Campaign status decoding for the Google Ads UI's own Status column, which does not map one-to-one to the API's raw status field.
Why silent rewrites were a problem
Self-correction is only trustworthy if you can tell whether it worked. A plan patch that quietly failed left the run continuing on the original plan with no signal that the correction hadn't landed — which looks identical to a run that never needed correcting. Recording the outcome is what turns self-repair from a hopeful behaviour into a measurable one.
Frequently asked questions
- Does the agent change its own plan mid-run?
- It can patch the plan when something it learns mid-run invalidates a later step. What's new is that the outcome of that patch is recorded, so a rewrite that failed to apply is visible rather than silently leaving the run on the original plan.
- Why does naming an account in my goal matter?
- The clarifier now ranks accounts mentioned by name in the goal itself, so writing the account name in your request routes to it rather than leaving the selection ambiguous.