A contractor marks a council repair complete. The resident’s case is still open. Staff have seen similar contractor updates close other cases automatically, and a ticket asks the team to make the two statuses agree. The operating procedure says the council can close a case once it has recorded the outcome.
A coding agent can make those statuses agree. It can find the fields, change the integration and add a test that checks they match. The harder question comes before any of that. Does the contractor’s update have the authority to close the resident’s case?
This is an illustrative council example. It shows how an apparently small engineering change can settle a service decision that the team has yet to make.
Find out what each record establishes
The contractor’s system can establish that someone marked the job complete. The council’s system can show that the case remains open. Both can be accurate. An engineer may have attended and closed the job, while a resident has reported that the problem has returned. A screen with one “Complete” field would hide that distinction.
An agent can help trace the source of each value and turn the discrepancy into a reproducible example. The ticket shows what someone asked to change. The source code shows what the integration is written to do. The deployed version, configuration and case history help establish what actually happened. The procedure sets out the intended process, whose application the team can check with the people who run it.
Put those findings in front of the service owner. Can a contractor’s completion message ever close the council case by itself? If so, for which repairs, under what conditions, and where is that decision recorded? Some decisions sit with the team. Others need policy or operations input. GOV.UK’s guidance on agile governance encourages teams to make decisions within their authority and escalate when needed. While the question is open, the team can reproduce the discrepancy and show both statuses in a prototype. Changes to the closure rule can wait for a decision.
Make the decision visible in the software
Suppose the council decides that a contractor update changes the job’s status and that the case closes when the council records a separate decision. The screen can then show “Contractor marked complete” alongside “Council case open”. An engineer can trace each to the event or decision it represents.
Give the implementation team examples that would expose a misunderstanding. A completion message arrives while the case is open. The service records the update and the case stays open. The message is delivered twice. It produces one recorded update. An earlier completion message arrives after staff have reopened a case. The reopened status is preserved.
Those examples give the independent tester something to check. A reviewer should be able to trace the checks to a named decision with a scope and effective date. If the council later approves automatic closure for a particular class of repair, the team can see what changed and which checks to revisit. That is the starting point for Who tests the tester?. The people writing acceptance tests need an agreed account of what ought to happen.
Decide what to do about the records already affected
The new behaviour governs future updates. It leaves a question about cases the old integration may already have closed. An agent can help identify records for review within an agreed scope and with authorised access. The council decides which cases to examine, whether to reopen any of them, and what to tell residents. A release that changes tomorrow’s behaviour and a correction to yesterday’s records are two pieces of work with different consequences.
If you’re commissioning AI-assisted delivery, ask your supplier to show you a change where the ticket, the live behaviour and the service owner’s understanding differed. Who recorded the disagreement? What evidence went to the person who could settle it? Can you trace that decision into the behaviour that shipped?
Further reading
GOV.UK Service Manual: Governance principles for agile service delivery
