AI agents and business automation
An operating model for AI agents in a growing business
Give agents owners, controls, support and improvement routes before they become invisible infrastructure.
Create a visible portfolio
Maintain a simple register of live, testing and retired agents with purpose, owner, users, permissions, supplier and review date. This matters because AI agent operating model decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. If the business cannot see the portfolio, it cannot govern cost, risk or duplicated effort. Record the choice, the evidence still required and the person who will return with it. Keep the mechanism proportionate: the purpose is better judgement and follow-through, not additional ceremony.
Separate business and technical ownership
The business owner remains accountable for the workflow and outcome while technical ownership covers reliability, integration and change. This matters because AI agent operating model decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Neither role can safely assume the other is monitoring the whole service. Record the choice, the evidence still required and the person who will return with it. Bring specialist judgement into the decision where required while retaining business ownership of the outcome.
Use proportionate design standards
Define minimum expectations for testing, access, logging, human oversight, documentation and fallback by risk tier. This matters because AI agent operating model decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Standards should make common work faster while forcing consequential work to earn stronger assurance. Record the choice, the evidence still required and the person who will return with it. Keep the mechanism proportionate: the purpose is better judgement and follow-through, not additional ceremony.
Plan service support
Decide who responds when the agent fails, data changes, a supplier model updates or users lose confidence. This matters because AI agent operating model decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. A successful prototype without support becomes a fragile production dependency. Record the choice, the evidence still required and the person who will return with it. Bring specialist judgement into the decision where required while retaining business ownership of the outcome.
Run incidents and learning together
Provide clear routes for reporting harmful output, security concern, unexpected behaviour and repeated user correction. This matters because AI agent operating model decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Treat incidents as operational evidence that can improve design and policy. Record the choice, the evidence still required and the person who will return with it. Keep the mechanism proportionate: the purpose is better judgement and follow-through, not additional ceremony.
Review value and retirement
At agreed intervals assess usage, outcomes, cost, risk, user experience and whether a simpler alternative now exists. This matters because AI agent operating model decisions rarely fail through a lack of possible technology. They fail when the business problem, operating context and responsibility for the outcome remain implicit. Bring evidence from the people doing the work, the systems supporting it and the leaders accountable for the result. Test assumptions about time, behaviour, data quality and implementation effort before treating them as facts. Retiring a weak agent is good portfolio management, not a failure of the AI strategy. Record the choice, the evidence still required and the person who will return with it. Bring specialist judgement into the decision where required while retaining business ownership of the outcome.
What to carry into the work
- If the business cannot see the portfolio, it cannot govern cost, risk or duplicated effort.
- Neither role can safely assume the other is monitoring the whole service.
- Standards should make common work faster while forcing consequential work to earn stronger assurance.
- A successful prototype without support becomes a fragile production dependency.