- Management was always about behaviour, not output.
No organisation can write down everything a worker should do. Where instructions run out, managers shape behaviour through incentives, norms, culture and judgement. AI agents bring that problem back.
- An agent is anyone, or anything, given a goal and discretion.
An employee and an AI agent both receive goals they must interpret and act on in ways no one fully specified. That makes both of them workers to be managed, not tools to be installed.
- Incentives move people. What moves agents?
People change what they do because of what they expect to gain or lose. An agent’s dispositions are set by reward during training; at work it responds only to instructions, context and constraints. Managing agents starts with knowing which lever works, and when.
- A harness is an organisation.
The roles, permissions, memory, checkpoints and evaluations built around a model are organisational design written in code. They can be studied with the theory and evidence of organisational behaviour, and they deserve the same rigour as the models themselves.
- Norms can be written down, and they can emerge.
Policies and constitutions put norms into an agent’s instructions. Agents that interact repeatedly may also settle into conventions of their own. Both kinds need to be understood and managed.
- People stay at the centre.
Steering a probabilistic system through language is a principal–agent problem with no incentives to align the two sides. Human judgement remains the anchor, and the question worth asking is what people add on top of the best an AI can do.
- When agents manage agents, a person is still accountable.
Delegation chains between agents must always trace back to someone who answers for the outcome. Accountability does not emerge on its own; it has to be designed in.
- Test it where work actually happens.
Every claim in this field should meet evidence from experiments, trials and fieldwork, across formal and informal work, office and field, in India and beyond.
The questions we are raising
- Question 1Can incentives steer an agent that wants nothing, or only the people who train it?
- Question 2If an agent’s dispositions come from its training, who is its real employer: the lab that trained it or the organisation that deploys it?
- Question 3Can social norms be instilled in agents by instruction, and do they emerge among agents on their own?
- Question 4What does it mean to appraise a worker that can be copied, paused and rolled back?
- Question 5Is a model capital an organisation owns or a resource it rents, and what does that change about how it is managed?
- Question 6When an agent delegates to another agent, where does responsibility sit?
- Question 7What happens to human expertise when the routine work that once built it goes to agents?
- Question 8Which management practices transfer from people to agents, which fail, and what has to be invented?
The research agenda takes up each of these questions. See the order