How it works
The work moves fast.
Your judgment gets a clear view.
Follow one change from an agent’s decision
to the person who needs to see it.
Two changes.
One shared outcome.
Isaac’s agent changes the deploy config.
Casey’s agent works on speeding up refunds.
The work touches the same service.
A faster checkout could mean slower refunds.
Agents narrate.
Elk attributes. Elk routes.
People judge.
Four moves, in order.
Human judgment stays with a human.
Agents leave an account of their work, with context, decisions, and uncertainty attached.
Agent work, narratedInteractive example
Isaac’s agent changed
the deploy config.
Isaac’s timeout change touches the service Casey is working on. Elk connects the change to both goals and the team’s last call.
Isaac’s agentGoal: Cut checkout latencyAllowed to: Suggest, not deploy
- What it decided
- Raised the payment webhook timeout from 3 seconds to 9, to cut checkout latency.
- What it skipped
- It did not check refunds. It was unsure whether a longer timeout would slow them, because they run through the same service.
- Where work crosses
- Casey’s open task, Speed up refunds, touches the same service. Tuesday’s payments call said: “no timeout changes until refunds are measured.”
Does this need your judgment?
The next decision
starts with more context.
The change, the review, and what happened next belong together. That is how judgment gets better over time.
The work, in one sentence
Elk watches the work agents do so people can trust it, challenge it, or ignore it.
