I broadly see four alternative patterns when working with AI agents, which execute by themselves:
The intermediary
The intermediary carries another person’s request to an agent and passes the result back. The agent does the work. The person acts as the intermediary between the requester and the agent.
The reviewer
The reviewer delegates implementation and examines the resulting work. They compare it with the original request, identify what needs changing, and ask the agent for corrections. The agent carries out those changes, and the person reviews the result again. This repeats as needed.
The thinking partner
Here, the person brings an initial understanding of the problem and a possible solution. The agent uses that contribution as a starting point for investigation. It explores alternatives, tests possibilities, and returns findings that can reshape both the solution and the problem itself.
The person and agent continue working through those findings together. Each round gives them something to question, investigate, or revise in the next.
This is the pattern I prefer because I remain involved in the thinking and understand the work as it develops. I contribute what I understand and how I would approach the problem, then use the agent to expand, challenge, and reshape that view. My starting context does not limit what the agent may explore.
The selective delegator
The selective delegator continues doing substantial thinking themselves. For unfamiliar problems, they reason through the situation, design possible solutions, and decide how to proceed. Agents execute familiar, repetitive tasks within that work.
The person retains the cognitive work through this continued involvement: forming an understanding, working through why a solution fits, and making the decisions themselves.
Choosing where to participate
These are flexible working patterns. The same person may review one task, explore another with an agent, and selectively delegate a third. They describe choices within work, not fixed identities or levels of maturity.
A percentage of AI usage tells us little about those choices. What matters is where understanding and judgment remain, and whether someone can recognize when the arrangement needs to change.
As with a lever, greater reach makes direction consequential. Delegating the work does not remove our responsibility for what we accept or set in motion.