Blog: Governance
Your AI agent needs a spending limit. Not just a job description.
Published by Forever Foundry
An AI agent can be given a clear goal and still make a decision your business would never approve.
Resolve the complaint. Recover the sale. Keep production moving. Each sounds sensible, until achieving it involves an excessive refund, an unprofitable discount or a purchase nobody authorised.
More capability means more decisions to govern
On 11 September, Salesforce announced a new portfolio of job-ready AI agents and technology designed to let agents pursue goals over days and weeks, including controls for business rules, permissions and when seller approval is required. Some capabilities are available now; others remain in pilot or are scheduled for later release. Salesforce's announcement
Our view: as agents take on longer-running work, businesses need to define decision boundaries as carefully as they define objectives.
Agree the limits before switching it on
Consider an agent helping a manufacturer source replacement materials. Finding available stock is one task. Committing the business to a different supplier, price or specification is another.
Before deployment, operations, procurement, quality and finance should agree what the agent may recommend, what it may execute and what requires human approval. A cheaper material is not a saving if it creates a quality failure.
These are business decisions, not settings to leave entirely to the implementation team.
Put control into the business case
A credible AI proposal should identify the problem, expected benefit, accountable owner and limits of authority. It should also explain how people can stop or reverse an action.
Measure more than speed. Track the cost per successfully completed task, exceptions, errors and time spent reviewing the agent's work. Faster execution has little value if checking and correcting it consumes the saving.
This is where the Forever Foundry Process™ starts: listening across the organisation through AI Compass™, then helping business and technical teams prioritise opportunities through Slid3rs™. AI helps gather the evidence. Your people decide what matters and what should move forward.
Human control should not begin when something goes wrong. It should be part of the investment decision.
This follows on from where AI agents tend to get deployed first. Read about finding AI opportunities between teams.
Know what to prioritise before you automate it. Explore Slid3rs™.
Next