AI agents working in finance need more than permission to use a tool: they need clear limits on what they may decide and do. In a Forbes Council essay, LinkedIn finance technology director Pankaj Prasoon argues that formalising those limits could let companies scale automation without confusing a confident answer for authority to act.
Watch Desk analysis
What happened
Prasoon says companies should give each financial agent a written delegation of authority that software and controls can interpret. For every agent, he recommends recording its identifier, named business and control owners, the transactions it may handle, its autonomy level for each, monetary limits, reversibility, escalation triggers and the evidence it must leave behind.
He suggests calibrating autonomy to the materiality and reversibility of an action, how well it is understood and whether anyone outside the company will see it. A routine, reversible internal reclassification might warrant more independence than an irreversible payment or an action affecting a customer or regulator. He also argues that controls should act before an agent takes a consequential step, and that records should show what the agent knew, which rule it applied and why it had authority.
Why it matters
That is a useful distinction for finance teams: an agent can be right about an action and still lack permission to take it. Prasoon’s proposed record gives organisations a concrete way to set that boundary, name the people responsible and review what happened afterwards.
It also shifts the question from whether a model is capable to whether a particular action has been properly delegated. As agents move from drafting recommendations to moving money or posting entries, that distinction becomes rather less academic.
Our read
Prasoon’s strongest point is that “a human remains accountable” is not much of a control unless someone can say which human owns the outcome and which owns the authority framework. His one-page register is not a complete governance programme, but it gives finance teams a practical place to start: define the agent’s powers before a successful demo quietly turns into an operating process.
What to watch
- Whether finance teams document named business and control owners for each agent.
- How organisations set monetary limits and escalation rules for different transaction types.
- Whether audit records capture the agent’s authority and reasoning, not merely who clicked approve.
Discussion spark: Should finance agents be allowed to execute routine, reversible transactions on their own, or should every action that moves money or changes a ledger require human approval?
Sources and evidence
- AI Agents And Delegations Of Authority: Best Practices For CFOs – Forbes (19 October 2026, 11:45 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.