Something fundamental is changing within enterprises. AI is no longer limited to helping people draft, search, analyse, or recommend. Increasingly, agents can act on their own: pulling information, making decisions, completing tasks, and interacting with systems that once required a person at every step.
That makes the next phase of AI adoption less about deploying another technology and more about managing a new kind of workforce. A one that has humans and agents working together. Most companies aren’t prepared for that yet.
The question for business leaders is therefore changing. Instead of asking, “How do we deploy more AI?” they need to ask, “How do we manage a workforce that includes AI?” That is simultaneously an operating-model question and a people question. Technology is only one part of the answer. The other is deciding what work should remain human, how to organise teams that include both, and what changes when a manager’s span of responsibility includes machines that can act autonomously.
Building the foundation
The starting point is identity and access. Every agent operating within an enterprise should have a verifiable identity, a defined purpose, and only the permissions it needs to do its job. Giving an agent broad access simply because it is easier is no more sensible than doing the same for an employee.
Governance needs the same attention. There should be clear ownership for an agent’s actions, explicit guardrails, and a defined point at which a human must take over. When an agent takes an important action, the organisation should be able to establish what happened and who approved that access.
But governance is not just an infrastructure problem. It is an organisational design problem. Every agent needs a human owner: not just someone who approved its deployment, but someone accountable for its work and its impact. Agents that operate without clear human ownership exist in a grey zone that no governance framework can fully close.
The same applies to lifecycle management. Agents are easy to create, but that should not mean they remain indefinitely. Their performance needs monitoring, their behaviour reassessing, and their permissions updating. An obsolete agent with active permissions is not harmless. It is an unmanaged risk.
Running AI like a workforce with accountability
Once the basic controls are in place, the harder challenge begins: running these systems day to day.
Whether the work was completed properly, whether the result met the required standard, and whether the organisation can rely on the agent to perform consistently—these are the questions that matter. Enterprises will need to monitor agents at scale, enforce policies consistently, and respond quickly when something goes wrong.
The shift also affects the humans in the organisation. When agents handle routine work, the remaining human roles become more demanding, not less. The tasks that stay with people require judgment, context, relationship management, and the ability to intervene when things go wrong. Organisations that automate work without investing in the people who remain will find they have efficient machines and an increasingly disengaged workforce.
Managers will also have to decide, routinely, how work is divided between people and machines. Should a person handle a particular task? Should an agent? Should the agent do the first pass and a person make the final call? These choices will become part of everyday management rather than being left solely to technical teams.
Preparing and empowering people, not just systems
This is where many organizations must invest heavily. Deploying agents is a technology decision. Preparing people to work alongside them is a cultural one, and culture does not change through mandates alone.
A top-down directive to “use AI” rarely produces lasting adoption. People adopt new ways of working when they see colleagues solving real problems with them; for example, when a peer on their team shows them a better way to handle a task, not when leadership issues a policy memo. The most effective AI adoption is often peer-to-peer: organic, practical, and grounded in the actual work.
That means organisations need to create the conditions for this to happen: time, permission, and visible examples of people using AI well. It also means investing in skills that go beyond tool training. When agents take on transactional work, people need to develop the capabilities that remain distinctly human: critical thinking, oversight, creative problem-solving, and the judgment to know when a machine's output should be trusted and when it should not.
This is not a one-time training programme. It is a continuous investment in the workforce that will work alongside AI for the foreseeable future.
The leadership imperative
No single leader can make this transition alone. A company head can set the direction, but a blended workforce will only work if the leadership team across various departments is genuinely aligned on what it means and how it will be managed.
That requires a level of collaboration many leadership teams have not yet built. HR leaders need to be in the room when agent deployment decisions are made, because those decisions directly reshape roles, skills, and culture. Technology leaders need to understand that governance is not just a security concern, but a people concern. And the leadership team needs to create the environment where these conversations happen honestly, where everyone can raise concerns about readiness, capability gaps, and the pace of change without being seen as obstacles.
The companies that will manage this well are not those with the most advanced technology. They are those where leadership has the confidence to redesign how work gets done, the humility to involve the right people in those decisions, and the discipline to invest in their workforce even as the composition of that workforce changes.
Every company may eventually become an AI employer. The ones that thrive will be those whose leaders understood, early enough, that managing this new workforce was never just a technology project. It was also a people leadership one.
(The author is managing director-India, Snowflake. Views are personal.)