The Work You Choose to Keep

What can AI do? It is a useful question, and it can occupy an entire conversation. A second question is more personal: what am I choosing to keep responsibility for, and why?

The answer depends on the work. A tool may help draft a document without being in a position to judge its consequences. A workflow may be suitable for automation while a conversation inside it still requires close human attention.

Delegation becomes more thoughtful when it follows this distinction, rather than a general ambition to automate as much as possible.

Generation and verification are different tasks

Producing a competent-looking first draft is only one part of knowledge work. Someone must check the claims, notice what is missing, interpret the context and decide whether the result is fit for use.

Verification is less visible than generation. It may end in a shorter document, a rejected proposal or a decision to do nothing. That does not make it a smaller contribution.

If you delegate the draft, make the review explicit. What could be wrong in a way that a casual reading would miss? What evidence is required? Who understands the consequences well enough to approve it?

Expertise can help with those questions, but seniority alone is not a guarantee. Review requires attention and a willingness to look beyond a fluent answer.

Four useful ways to organise AI work

Think of these as practical patterns, not a ranking of professional maturity.

First, AI can assist with individual tasks: a draft, a summary or a list of options. The task remains clearly bounded, and you check the result before using it.

Second, it can become part of an iterative thinking process. You propose an idea, ask for objections and revise. The value depends partly on the questions you bring and your ability to assess the responses.

Third, you may delegate a sequence of steps. The review then needs to cover both the final output and the choices made along the way. Permissions, exceptions and stopping points matter more as the scope expands.

Fourth, AI can become part of an ongoing operating system. In that case, someone must own maintenance, monitor failures and decide when human intervention is required. Automation does not remove organisational responsibility.

Keep attention where consequences live

The work you keep may include a sensitive interview, a judgment about competing priorities, or a decision that affects other people. Tools can support each of these. That support does not settle who is accountable.

Ask what you need to understand firsthand. What would you lose by receiving only a summary? Which uncertainty requires a conversation rather than another generated answer? Where would an error be expensive to repair?

These questions help identify the points at which human attention should be protected. They also clarify why a task is retained, so that keeping it does not become resistance to change by habit.

Design the handoff

Choose one recurring piece of work. Break it into preparation, generation, checking, deciding and communicating. Mark what a tool can help with and where a person needs to stay involved.

Then name the owner of the whole result. A series of efficient steps can still create a poor outcome if nobody is responsible for the transitions between them.

The aim is to make more room for the contribution you want to bring, while being precise about the responsibilities that remain. What would you delegate willingly, and what would you keep because understanding it directly is part of doing the work well?