The last run
What the agents did overnight: the steps they actually took, what they found, and what they wrote, with each part labelled by how solid it is.
The platform works while you are not looking. The last run is the readable record of that work: what ran, what it did, and what it left behind.
What you'll learn
- What a run is and when one produces a brief
- Why the three parts of a brief have different levels of certainty
- How to act on it in five minutes
What a run is
An agent run is a scheduled episode: a plan, a set of steps, and whatever those steps produced. If a run had no executable plan and took no actions, it is recorded as failed and never appears as a brief. "It ran" and "it did work" are different claims, and only the second produces something to read.
A brief also requires that at least one step which leaves something behind actually succeeded. A brief has to be able to quote something real.
The three parts, and their different bases
| Part | What it is | How solid |
|---|---|---|
| What it did | The run's own steps | Counted. These rows belong to this run and nothing else |
| What it found | Insights created during the run's window | Attributed by time, not by link |
| What it wrote | Recommendations created during the run's window | Attributed by time, not by link |
The second and third are a proxy and the brief says so on the rows themselves. Nothing in the data links an insight or a recommendation back to the run that produced it, so they are matched by creation time inside the run's own start and finish window. If another job wrote something during those same minutes, it lands here too.
That note is not a disclaimer for the sake of it. If you are counting agent output for a client report, count "what it did". The other two are indicative.
The run's own start and finish times travel with it, so the word "overnight" is never taken on trust. The card reads the clock off the record.
Reading one in five minutes
- Check what it did. The counted steps. This is what actually happened.
- Read the top of what it found. Anything that changed a rival's position or a surface's rate is worth more than everything else on the list.
- Move anything worth doing into the plan. A recommendation written overnight is a proposal, nothing more, until a person approves it. See This week.
Try this in Outercite
Ask the assistant "What did you find overnight?" If no qualifying run has completed, you are told that plainly instead of being handed a summary of nothing.
Related
This week
Where an overnight recommendation goes to be ranked and approved.
Changes
What happens to it after approval.
Set up alerts
Getting the significant findings pushed to you instead of waiting for the brief.
Last updated on
This week
The ranked plan: which open change to do first, what it is projected to be worth, and how much of that projection is calibrated against shipped outcomes.
Intelligence
One surface with six lenses: Intelligence, Ground, Where you lose, Engines, Gaps and Explore, plus the forecast record. Scale and Enterprise.