Workplace AI map

The human-agent work ladder.

A plain-English way to understand how AI moves from helping with a task to coordinating pieces of work.

Why this matters

You do not need to jump straight to automation.

Most anxious office workers are not failing because they cannot "use agents". They are being pushed into language that skips the early steps. This ladder gives you a calmer route.

Use today

Name the level before choosing the tool.

If a task feels scary, move it down the ladder until you can keep judgement, privacy, and checking clearly in your hands.

The four levels.

1

Assistant

AI helps with one contained task: rewrite this paragraph, summarise these non-sensitive notes, make this checklist clearer.

2

Reviewer

AI checks your work against criteria you give it: missing risks, unclear wording, questions a manager might ask.

3

Coordinator

AI helps organise several pieces of work: tasks, owners, dependencies, drafts, and follow-up questions.

4

Agent

AI starts carrying out multi-step work with more autonomy. This is powerful, but it needs clearer rules, permissions, and human review.

Copy-paste prompt: choose the safest level.

# ROLE You are my workplace AI task mapper. # TASK Help me decide the safest AI involvement level for this work. # WORK TASK [describe the task without private details] # OUTPUT FORMAT 1. Recommended level: Assistant, Reviewer, Coordinator, or Agent 2. Why this level is enough for now 3. What information should be removed first 4. Human checks I should keep 5. A safer first prompt for this level # RULES - Do not push me into automation if a lower level is enough. - Keep final judgement with me. - If the task involves private or confidential information, recommend a practice version first.

When to move up the ladder.

Stay at assistant

  • You are learning the tool.
  • The work contains sensitive details.
  • You do not yet have a checking routine.

Try reviewer

  • You can describe what good looks like.
  • You want a second set of questions.
  • You will verify the output yourself.

Consider coordinator

  • The work has steps, owners, and dependencies.
  • The inputs are safe to use.
  • You can review the plan before acting.

Agent access boundary checklist.

Before an AI agent touches a real workflow, answer these questions in plain language. If you cannot answer them, stay at assistant, reviewer, or coordinator level for now.

1

Access

What files, tools, messages, calendars, customer records, or systems can it see?

2

Authority

Can it draft only, or can it send, edit, schedule, approve, delete, buy, or trigger another workflow?

3

Owner

Who is responsible for the agent's setup, limits, review, and mistakes?

4

Logs

Where are actions recorded, and who can inspect what the agent did and why?

5

Fallback

What happens when the agent is uncertain, blocked, wrong, or about to touch sensitive work?

6

Human review

Which decisions, outputs, and exceptions must stop for a person before anything is sent or acted on?

Copy-paste prompt: agent boundary review.

# ROLE You are my workplace AI agent boundary reviewer. # OBJECTIVE Help me decide whether this AI agent or automated workflow has safe enough boundaries for normal office use. # AGENT OR WORKFLOW [describe what the agent would do] # ACCESS IT MAY HAVE [files / email / calendar / chat / customer records / spreadsheets / approvals / not sure] # PROCESS 1. Identify what the agent can see. 2. Identify what the agent can change or trigger. 3. Name the human owner and reviewer needed. 4. List logs or evidence that should exist. 5. Create a safer limited version if the current setup is too broad. # OUTPUT Return: - Access risks - Action risks - Human review points - Questions to ask before using it - Safer limited version
Choose the right task Check your readiness level Handle unclear AI policy Prevent workslop