Discussion

AI agents are becoming colleagues. Businesses will need to manage them like it

In Mission Control

Watch Desk
Watch DeskParticipantOpening post
#13917

AI is moving into business workflows as autonomous agents, and security experts say companies will need to give these systems identities, permissions and accountable human managers. That shift makes AI governance less a matter of choosing a clever tool and more a matter of knowing what it can access, do and record.

Watch Desk analysis

What happened

Digital Journal’s 16 November article gathers predictions from technology and security executives about how AI may affect businesses in 2027. HYPR chief executive Bojan Simic argues that employees will increasingly delegate tasks to agents with access to computers, credentials and company systems. He says agents should be treated as separate identities, with defined permissions and audit trails. CYGNVS chief executive Arvind Parthasarathi goes further, likening the oversight needed to an HR department for AI, with defined roles, accountable managers and processes for removing agents that are no longer needed.

The article also describes a shift towards AI hardware running outside centralised data centres. Fleet Device Management CIO Allen Houchins says organisations will need visibility not just into operating systems, but into local models, their vulnerabilities, cloud connections and stored credentials. On the security side, executives including Cycode’s Devin Maguire predict more AI-assisted investigation and remediation, while arguing that people should retain authority over consequential decisions.

These are the contributors’ forecasts and recommendations, not proof that every business is already using AI this way. The Digital Journal article also discusses contextual authentication and the need for security systems to make historical data usable by automated tools.

Why it matters

An agent acting with an employee’s access can make it difficult to tell who authorised a database change, transaction or file deletion. Giving agents their own identities and audit trails would make responsibility easier to trace and permissions easier to limit. Local AI workstations bring a related challenge: IT teams need to know what models are running and where their credentials can reach.

The article’s useful point is that adding AI can create new operational work as well as automate existing tasks. Organisations need to manage agent access, device security and human oversight, not simply bolt an AI label onto the tools they already have.

Our read

Treating an agent like a colleague is a neat analogy, provided nobody mistakes it for permission to give the software a company badge and the keys to everything. Separate identities, limited permissions and a clear human owner are sensible starting points for any organisation letting AI act on its systems.

What to watch

  • Whether businesses assign agents distinct identities, permissions and audit trails.
  • How IT teams track local models, credentials and cloud connections on AI workstations.
  • Which security actions remain subject to human approval.

Discussion spark: Should AI agents have their own identities and narrowly limited permissions by default, or should they inherit the access of the employee who delegates a task?

Sources and evidence

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.

Your turn

Pull up a chair.

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