Lattice was right about the problem

On 9 July 2024 Lattice, a performance-management software company, announced that it would treat AI agents as employees: onboard them, add them to the org chart, review their performance. On 12 July, after hundreds of objections from HR and technology professionals, it withdrew the plan. The episode is usually told as a cautionary tale about hype. It deserves a more careful reading, because Lattice saw something true.

What it saw was that agents have a lifecycle and nobody is running it. They arrive, they are given access and tasks, they perform well or badly, they change when the vendor updates them, and eventually they are switched off. Every one of those moments is a decision someone should own, and in most organisations nobody does. Lattice's instinct was to reach for the only machinery organisations have for managing a lifecycle of working entities, which is the machinery of employment. The instinct was correct. The machinery was wrong.

It was wrong because the employee is a category built around personhood. An employee has rights because an employee can be harmed. An employee has a career because an employee has a future that matters to them. An employee is reviewed in a conversation because an employee can hear, reflect and change. None of this is true of an agent. An agent can be copied a thousand times at the cost of computing. It has no stake in its own continuation. It cannot be held responsible, because responsibility requires someone to whom the consequences matter.

The people who objected to Lattice understood this instantly and expressed it as offence: the plan "disrespects the humanity of your real employees," one wrote. The offence was accurate. Putting an agent on the org chart does not elevate the agent. It dilutes the category on which every protection a worker has depends. If an agent can be an employee, then being an employee no longer implies being a person, and everything that follows from personhood, from dignity to due process, becomes negotiable.

Since then the evidence has caught up with the instinct of those who objected. In May 2026 a team from Boston Consulting Group and Boston University reported in Harvard Business Review on a large experiment in which agents were framed as employees: individual accountability fell, unnecessary escalation rose, review quality dropped, and employees became less sure of their own roles, with no gain in adoption. The anthropomorphism is not a matter of taste. It measurably weakens the thing an organisation most needs from the people around an agent.

So the lesson is not "do not manage agents." It is "manage agents with the rigour of human resource management and none of its anthropomorphism." Agents are selected, not hired: qualified like a supplier against the task they will do, the cost of checking their output and the named human who will answer for it. They are deployed, not onboarded: placed inside a written boundary of what they may do alone, what only with review, and what never. They are evaluated, not reviewed: measured on volume, quality, incidents and the verification hours they consume, with no conversation about motivation. They are updated, not developed: a vendor's new version is a redeployment, re-tested and re-signed. And they are retired, not terminated, after one question employment never had to ask: can anyone here still do this work by hand?

Who runs that lifecycle? Not the vendor, whose interest is in the agent staying deployed. Not IT, whose job is the system working rather than what it does. Not the business unit alone, which bought the agent to solve its backlog and is not placed to see the organisation-wide pattern. The lifecycle needs an owner with authority over both the human and the artificial parts of the workforce, which is the role I have called the Chief Workforce Officer.

Lattice reached for the wrong tool because the right one did not exist. Two years on, it still does not exist in most organisations. The company's three-day retreat settled only that agents are not employees. It left open the question that actually mattered, which is who is responsible for them, and that question is still waiting.