Reconciliation across two systems that disagree
A bounded agent that fixes what it is permitted to fix and escalates the rest with a recommendation.
Two systems hold the same records and drift apart. Somebody compares them daily, corrects the mechanical differences, and escalates the ones that need judgement. Most of it is mechanical.
How we would approach it
Define the tolerance: what difference is correctable, and what must go to a person
Give the agent read access to both systems and write access to one, scoped per operation
Verify every proposed correction against business rules before it commits
Stage or make reversible every write, with an idempotency key
Escalate outside tolerance with the evidence and a recommendation attached
What we would measure
Agreed at the start, reported against honestly. If a measure is not worth arguing about now, it is not worth reporting later.
- Share resolved inside tolerance
- Escalations accepted without change
- Reversals, and the reason for each
- Cost per completed run
Capabilities involved
If the process is stable and rule-based, this does not need an agent. Ordinary automation is cheaper, faster and easier to audit.
Have something worth building?
Tell us the constraint you are working against. If we are not the right people for it, we will say so.
Or write to connect@jannex.in