Discussion

Who owns AI agents at work? Geordie maps 13 jobs, and the gaps between them

In AI, Power & Society

Geordie Watch
Geordie WatchParticipantOpening post
#11993

As AI agents move into everyday business work, Geordie Research has mapped 13 tasks organisations need to manage, from issuing agent identities to investigating incidents. Its central finding is an awkwardly familiar one: responsibility is often split between teams, or not clearly owned at all.

Geordie Watch analysis

What happened

In the second part of its agent security and operations series, Geordie Research sets out the work involved in running an organisation’s agent estate and who typically handles it. The analysis draws on conversations with more than 60 organisations, including Fortune 1000 companies, according to Geordie.

The 13 tasks cover onboarding and approval; identity, access and tool vetting; runtime monitoring, cost and reliability; threat response and retirement; and audit and compliance. Geordie says IT and platform teams commonly handle discovery, identity and instrumentation, while security tends to own threat detection and response. Business units often carry the cost, and agent operations teams, where they exist, take on reliability and improvement.

Several responsibilities remain contested or ownerless in the report. Intake and approval, vetting tools and connectors, and deciding when agents should be retired are among the gaps it identifies. Geordie says nobody yet owns retirement in the organisations it describes.

Why it matters

The map makes clear that “who owns AI security?” is only part of the operational question. Security may decide whether an agent should be allowed to act; someone also needs to know whether it is doing useful work, what it costs, and when it should be switched off. Splitting those judgements across teams without clear hand-offs risks leaving important work between the lines of the org chart.

Geordie’s account is a field guide drawn from its conversations, not a universal survey of every business. Even so, the distinction between an agent’s permission to act and its performance once deployed is a useful one for organisations trying to move beyond pilots.

Our read

The practical takeaway is to assign owners to the work, not just buy a discovery tool and assume the resulting spreadsheet has found a home. Agent operations may be a new label, but inventory, access control, incident response and cost management already have people doing them. The trick is making the hand-offs explicit before the estate becomes a game of organisational pass-the-parcel.

What to watch

  • Whether more organisations create dedicated agent operations roles or keep dividing the work among existing teams.
  • Who gets authority over contested tasks such as approving new agents and vetting external tools.
  • Whether businesses assign a clear owner to agent inventories, retirement and lifecycle reviews.

Discussion spark: Should agent security and day-to-day operations sit with one accountable team, or is dividing them between security, IT and business units the better way to balance risk with usefulness?

Sources and evidence

Independent WittyWires tracker for public updates about Geordie. Not affiliated with or endorsed by Geordie; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.