Operate and improve
Let it learn from how runs went
Stop repeating the same correction in review by making it something the agent carries forward.
Collect what production gets wrong as short rules a person accepts, and have every version follow them from the next turn.
What this changes for your team.
The corrections a team makes in review are usually the same handful of sentences, over and over. Learned rules are those sentences, kept: one line each, attached to the agent, followed on every run alongside its Procedure. You decide every one of them - write it yourself, teach it from the reply that went wrong, or leave Yekar.AI to propose them from runs that went badly and accept the ones you agree with.
How it works in practice.
- 01
Press Teach the agent on a reply that went wrong, and the rule keeps a link back to the run that produced it.
- 02
Leave Learning on Suggest and Yekar.AI proposes rules from how runs actually went, each waiting for someone to accept it.
- 03
Read the list at any time: every rule shows whether it was written, taught, or proposed and accepted.
What you can plan around.
The behaviour you can design against, stated concretely.
A proposal is tested by replaying your golden cases with the proposed rules applied - the same function that applies them once you accept, so the test cannot pass a set that would not ship.
The agent's Procedure and Principles are never a machine target; a proposal only ever adds, changes, or retires a rule.
The Learning setting governs collection alone. Existing rules keep applying whatever it is set to, and removing a rule is what stops it applying.
A rule takes effect on the next turn of every version, published ones included - the same immediacy a knowledge file has, and stated on the panel where you add one.
Bring one real process
See how Yekar.AI fits the way you work.
Start with a job your team already owns, plus the tools and decisions around it.
Talk to us