Why it matters now
Agents take actions, not just produce text, and that changes the leadership question. Someone has to decide what an agent may do, check its work and answer for the result. It is close to how I run Herman Foundry: agents carry much of the routine work, and people carry the judgment. It works when the boundaries and the ownership are clear, and it fails when they are not.
- 62% of organisations are at least experimenting with AI agents.McKinsey, The State of AI in 2025 ↗
- 74% of organisations expect at least moderate use of AI agents by 2027, yet about 80% lack mature governance for them.Deloitte, State of AI in the Enterprise 2026 ↗
- Microsoft describes a new kind of manager, the “agent boss”, who delegates work to AI agents and manages them.Microsoft, Work Trend Index 2025 ↗
Common mistakes
- Agent projects driven by hype rather than a defined workflow.
- No clear accountability for what an agent decides or does.
- Governance that lags behind how agents are actually used.
- Giving agents broad access before they have earned trust.
What to do instead
- Start bounded. Choose a contained, end-to-end workflow with clear inputs and a clear result.
- Name a human owner. Every agent has a named person who answers for its work.
- Define autonomy. Write down what an agent may do alone, what needs approval and when it must hand over to a person.
- Redesign roles. Shift people towards setting goals, reviewing output, handling exceptions and improving the agents.
- Measure per workflow. Track quality, speed, cost and errors for each workflow, not overall activity.
Your first step this week
Pick one routine workflow and write down which steps an agent could do, which need a person’s approval and who would own the result. That is your first agent brief.
Related guides
- How do we govern AI risk and use AI responsibly?
- What happens to jobs and roles, and what do I tell people?
- How do we get from AI pilots to scale?
All AI leadership principles and guides · 100 AI use cases by function