Agentic work: how your job changes when AI does tasks

What it is

"Agentic work" means working alongside AI agents that can carry out tasks with several steps. Instead of doing every step yourself, you describe the goal, hand part of the work to the AI, and then review what comes back.

This is a shift from doing to delegating and reviewing, much like first managing a team. You spend less time typing and more time explaining what good looks like, checking, and deciding.

Think of a head chef. They still cook, but much of their value is in setting the menu, tasting before plates leave the kitchen, and taking responsibility for what reaches the customer.

The change is real, but it is gradual and uneven. Some tasks suit AI well, many do not, and the person remains accountable for the result.

Why it matters

• The quality of your brief now shapes the quality of the work, so clear thinking pays off more.
• Reviewing becomes a core skill: catching errors before they reach colleagues or customers.
• Judgement, relationships and accountability stay with people, and become a larger share of the job.
• Being honest about what AI cannot do protects you, your team and your customers.

How to do it

1. Pick one task you do regularly that has clear steps and a clear result, such as preparing a weekly report.
2. Write a brief: the goal, the background, the sources to use, the format and the limits. For bigger tasks, write a fuller spec (see guide 15).
3. Define "done": what a finished, acceptable result looks like, in checkable terms.
4. Set checkpoints: points where the AI stops and shows you its progress, such as after the plan and before the final version.
5. Review the output against your definition of done, checking facts, numbers and tone.
6. Decide and own the result: approve it, send it back with notes, or do that part yourself.
7. Note what you changed, and improve the brief for next time.

Example

Task: a monthly summary of customer feedback for your manager.

Before: you read every comment, sort them by hand and write the summary over half a day.

After: you brief an approved AI tool with anonymised feedback, then review.

Goal: A one-page summary of this month's customer feedback for my manager.
Background: We are a [type of team]. My manager cares most about [delivery delays and product quality].
Input: the anonymised comments attached. Use nothing else.
Done means:
- Top 5 themes, each with how many comments mention it.
- 2 short example quotes per theme, copied exactly.
- One line on anything new compared with last month: [paste last month's themes].
Checkpoint: show me the themes and counts first, before writing the summary.
Limits: do not guess causes. Mark anything uncertain with [check].

You check the counts on two themes by hand, confirm the quotes are real, adjust the wording to fit what you know about current priorities, and send it under your own name. The AI did the sorting; you did the judgement.

Common mistakes

• Vague briefs. State the goal, audience, format and limits.
• No definition of done. Write two or three checkable criteria.
• Reviewing only the last step. Use checkpoints so problems are caught early, not after an hour of work.
• Passing on work you have not checked. Your name is on it. Read it as if you wrote it.
• Assuming the job disappears. For many roles, the work changes shape rather than vanishing; focus on the parts that need you.

Try it this week

1. Choose one task from last week that took more than an hour (2 minutes).
2. Write a brief with goal, background, inputs, definition of done, one checkpoint and limits (8 minutes).
3. Ask a colleague to read it and tell you what is unclear, as if they had to do the task (5 minutes).

Going further

• Keep a small library of briefs that worked, and turn the best ones into saved skills or commands for your team.
• Agree as a team which tasks AI may draft, which need a named reviewer, and which stay fully human.
• Track time spent on briefing and reviewing, not just time saved, so you see the true picture.
• Build the new skills deliberately: clear writing, breaking work into steps, spotting errors, and explaining your decisions.
• For your career, be the person who can show a reliable, checked AI-assisted process, not just fast output.

Key words

• Agentic work: working alongside AI agents that carry out multi-step tasks while you brief and review.
• Brief: written instructions describing the goal, background, inputs, format and limits of a task.
• Definition of done: checkable criteria that say when a result is good enough.
• Checkpoint: a planned pause where the AI shows progress and you approve or correct it.
• Accountability: being responsible for a result, which stays with a person, not the AI.

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (245 KB), or the complete AI Field Guide (1.7 MB).