Guides
Working with agents
How autonomous agents close routine loops — and how to supervise them.
Agents are how WarTable turns diagnosis into action. An agent watches for a signal, decides what should happen, and carries it through — either automatically or staged for your approval.
The agent loop
Every agent follows the same five steps — Plan → Act → Observe → Decide → Repeat:
- Plan — set the objective and the signal to watch for (a stalled deal, an aging invoice, a slipping SLA).
- Act — take one step, or queue it for one-click human approval.
- Observe — check the result against what was expected.
- Decide — continue, retry, escalate, or hand off to a person.
- Repeat — keep going until the goal is met, with clear exit conditions so nothing runs away unchecked.
Starting safely
When you enable an agent, start with approvals on. The agent does all the work up to the final step and then waits for you. As you see it make good calls, you can let it act autonomously for the cases you trust.
- Begin with one loop, not ten.
- Keep approvals on until the agent has a track record.
- Review the agent's activity log regularly.
Good first agents
- Overdue invoices → draft and queue the follow-up.
- Quiet high-intent leads → re-engage and book the meeting.
- Changed SOPs → update linked tasks and notify owners.
The goal isn't to remove humans — it's to move you from doing every step to supervising the exceptions. Agents handle the volume; you handle the judgment calls.
To pick where agents will help most, check your Business Health Index — the weakest engine usually has the most open loops leaking out of it.