In this guide
Two kinds of strength
People and AI agents are good at different things.
People bring judgement, relationships, experience and accountability. They know when a rule does not fit the situation, when a customer needs a phone call rather than an email, and when a decision needs to go higher.
AI agents can take on defined support work: gathering information from several systems, checking it against a policy, preparing a response, or keeping a process moving between steps. They can help with repeated preparation work, but their outputs still need checks appropriate to the task.
In short
A digital workforce works best as support for a human team. Agents take on defined, checkable work. People keep the judgement, the relationships and the accountability.
Extend, do not replace
The useful question is not "which jobs can AI do?" It is "which parts of this work are taking time away from the judgement only our people can provide?"
In many teams, skilled staff spend much of their time on preparation: finding the file, checking the records, drafting the routine reply, chasing the next step. That work matters, but it is rarely where their expertise makes the difference.
When an agent takes on the preparation, the person starts from a prepared case instead of a blank screen. They spend their time reviewing, deciding and handling the exceptions.
The goal is not fewer people. It is more capacity for the work that needs them most.
What a well-designed agentic team looks like
An agentic team is people and AI agents with clear roles, working in the same system under the same rules. In practice that means:
- Clear roles. Each agent has a defined job, with set inputs and set limits.
- Shared rules. Agents follow the same permission checks and business rules as staff, with each agent's access limited to its assigned role, so nothing goes around your controls.
- People at the decisions. Work that matters, or is hard to undo, waits for a person to review and approve.
- Visible work. Your team can see what an agent did, pause it and correct it.
That is the difference between a team that is extended and one that is competing with its own software. What is agentic software? covers the model in more detail.
Where to start
Start with the work, not the technology. Look for tasks that are well defined, repeated often and easy to check, where a person currently spends time preparing rather than deciding. Those are usually the best first places for an agent to help.
If you would like to think it through for your team, start a conversation.
Questions
Frequently asked.
An AI agent with a defined role in your organisation's work, such as gathering information or preparing a draft. It works within set permissions, and its actions are recorded.
That is not how we design. The aim is to give the people you have more capacity for the work that needs them, not to remove them from it.
Work that is well defined, repeated often and easy to check, where a mistake can be caught and undone. Work with high stakes or real uncertainty should stay with a person, or wait for their approval.
Written by
Mare Pomana
Plumb Digital