AI agents are moving beyond chat, calling tools and APIs and taking actions inside business processes. WSO2 CTO Asanka Abeysinghe argues that governance must follow those actions, with clear authority, limits and evidence of what happened.
Watch Desk analysis
What happened
In a Forbes Technology Council article, Abeysinghe says businesses cannot rely on checking an agent’s final answer alone. They also need to govern the tools it used, the data it accessed, whose authority it acted under and the decision to continue, stop or ask for help.
His practical prescription starts with discovery: organisations need to know which agents, models, tools and delegated authorities they have before they can govern them. He recommends giving agents distinct identities rather than hiding their actions behind shared accounts, and applying controls where agents reach models, APIs, data and other systems.
Abeysinghe distinguishes controls inside an application, which can reflect the local workflow, from controls enforced at shared infrastructure boundaries, which can provide consistent coverage across teams. He argues businesses need both, and should keep evidence of actions so they can inspect and audit what happened.
Why it matters
Permission to use a tool is not the same as permission to use it in every way. An agent may be allowed to send email, for example, while still needing checks on the recipient, content or authority to make a commitment. The boundary is not just which button the agent can press; it is what the button does in context.
That makes this useful advice for organisations deploying agents, not merely another call to write a more impressive AI policy. Discovery, individual agent identities, controls at action points and records of what happened are concrete things teams can examine before handing software more responsibility.
Our read
Abeysinghe’s strongest point is that governance should attach to the action, not stop at the model or its answer. Start by finding the agents and shared accounts already in use, then map what each can access and do. It is less glamorous than unveiling an AI policy, and rather more likely to catch a real gap.
What to watch
- Whether organisations inventory agents, tools and delegated authority before expanding deployment.
- Whether agents get distinct identities instead of acting through shared accounts.
- Whether companies combine workflow-level checks with consistent controls at shared system boundaries.
- Whether action logs make it possible to inspect and improve decisions after the fact.
Discussion spark: Should an AI agent be allowed to take an action whenever it has permission to use the tool, or should each consequential action need a separate check?
Sources and evidence
- Trusted AI Governance Starts Where Intent Becomes Action – Forbes (4 November 2026, 17:30 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.