ANSWER
When should a legacy system not be retired by an agent?
Reviewed .
A legacy system should not be retired by an agent when it is the official record of critical state without an adequate replacement, when no reliable read path exists, when inputs and outputs cannot be compared, when essential rules remain unknown, or when the organization cannot approve, stop and reverse cutover. In those cases, an agent may support work around the core but should not pretend to replace it.
Step by step
- Confirm whether the legacy system stores critical state or only executes a rule.
- Verify reliable data access and coverage of essential rules.
- Test whether outputs can be compared through objective criteria.
- Require a risk owner, human approval and executable rollback.
- If a requirement fails, reduce scope to surrounding work or decline the case.
How an agent does this
The AgenticosCore Diagnostic qualifies these conditions before a proposal. Without access, equivalence, approval or rollback, the size is reduced or the case is declined. Control exists to make 'do not cut over' a verifiable outcome.
Frequently asked questions
Can an ERP or core banking platform never enter Legacy Exit?
The core may require another product or a specialized migration. An agent can still retire reconciliation, reporting, triage and manual approvals around it.
Does a failed dual-run mean the project failed?
No. Discovering before cutover that equivalence was not proven is a control outcome. Failure occurs only if that evidence is ignored.
Primary sources
- NIST AI Risk Management Framework — NIST
- Retiring service components securely — UK Government Security