Quick answer
To show users what your AI agent is doing, give every task five clear states, Working, Waiting for you, Done, Stuck and Stopped, and describe each one in a short, plain sentence about the task, not the technology. Say what the agent is doing (“Checking your last three invoices”), show progress for anything that takes more than a few seconds, make it obvious when it needs a decision from the user, and finish with a short summary of what changed. Users should understand the status at a glance, without reading logs.
Key takeaways
- The question users always have: “Is it actually doing what I asked?” Answer it before they ask.
- Use five states: Working, Waiting for you, Done, Stuck and Stopped.
- Talk about the task, not the system. “Comparing prices from 3 shops” beats “Executing tool call”.
- Make “waiting for you” impossible to miss, or the agent sits idle and users think it is broken.
- End with a receipt: what was done, what changed and anything left to check.
📘 Main article: This guide follows our article on the Meta Muse AI assistant and three lessons from Zuckerberg’s comeback, which explains why clear, friendly status updates are part of what makes Muse feel polished.
In this guide
- Why do users need to see what an AI agent is doing?
- What states should an AI agent show?
- How do you show users what your AI agent is doing? Step by step
- What do good and bad status messages look like?
- How does this work for Zapier and other automations?
- What mistakes should you avoid?
- Frequently asked questions
- The bottom line
- Related posts
When you hand a job to an AI agent, there is an awkward gap between asking and getting the result. Is it working? Has it stalled? Does it need you? If people have to dig around to find out, the magic disappears.
In our Meta Muse article, Nate B Jones describes how his assistant shows it is busy by pulling out a little laptop and typing, or scratching its head. You don’t need an animated character to get the same effect. This guide shows you how to show users what your AI agent is doing in plain words, whether you are building an agent, an AI feature or a simple automation.
Why do users need to see what an AI agent is doing?
Because when people can’t tell what is happening, they stop trusting the tool and start supervising it. Nate puts it neatly: every time a user has to stop and investigate, they are supervising software instead of getting on with their day.
AI agents make this worse than normal software. Tasks can take minutes, run several steps and pause for approval. Without clear status, users click again, cancel too early or assume it failed.
What states should an AI agent show?
| State | What it means | What to show |
|---|---|---|
| Working | The agent is busy on the task | What it is doing now, plus progress if possible |
| Waiting for you | It needs a decision, login or approval | A clear question and the button to answer it |
| Done | The task is finished | A short summary of what was done and any links |
| Stuck | It hit a problem it can’t solve alone | What went wrong in plain words and what the user can do |
| Stopped | The user or a limit ended the task | What was completed and what wasn’t |
💡 Tip: If you only fix one state, fix “Waiting for you”. An agent quietly waiting for an approval nobody has noticed is one of the most common reasons people think an automation is broken.
How do you show users what your AI agent is doing? Step by step
1. List the steps your agent takes
Write down each step in order, such as “read the email”, “find the order”, “check the refund policy”, “draft a reply”. These become your status messages.
2. Rewrite each step as a user-facing sentence
Use everyday words and mention the user’s own things: “Finding order #1042”, not “Querying database”. Keep each message under about ten words.
3. Show progress for anything over a few seconds
Use “Step 2 of 4” or a simple progress bar when you know the number of steps. When you don’t, a changing message (“Still checking prices, 2 of 5 shops done”) reassures people that it hasn’t frozen.
4. Make approvals stand out
When the agent needs the user, change the colour or icon, send a notification if they have left the page, and ask one clear question. Our guide on adding human approval steps to an AI agent workflow shows where those checkpoints belong.
5. Finish with a receipt
When the task ends, show a short summary: what was done, what changed, and anything the user should double-check. This also makes mistakes easier to spot.
6. Keep a detailed log, but hide it by default
Technical users and you, when fixing problems, will want the full record. Put it behind a “Show details” link so everyone else sees the simple version.
Want the story behind this? Read our Meta Muse AI assistant article to see how polish, a small team and a second attempt helped Muse reach a reported 10 million set-ups.
What do good and bad status messages look like?
| Situation | Unclear message | Clear message |
|---|---|---|
| Agent starts | Initialising agent… | Reading your request |
| Mid-task | Tool call 3 executing | Checking prices at 3 shops |
| Needs input | Awaiting input | Waiting for you: approve £24.99 order? |
| Problem | Error: 403 | Couldn’t open your calendar. Please reconnect it. |
| Finished | Task complete | Done: drafted 4 replies, ready for you to review |
How does this work for Zapier and other automations?
The same idea works for no-code automations in tools such as Zapier or Make. You can’t add a progress bar, but you can tell people what happened.
- Send a short confirmation by email, Slack or text when a long automation finishes, saying what it did.
- Add an “in progress” label or status column to the spreadsheet, CRM or task board the automation updates.
- Send a plain alert when a step fails, naming the step and what to check, rather than relying on the tool’s default error email.
- Use an approval step before anything that spends money, publishes or contacts customers, and make that request easy to spot.
⚠️ Warning: Don’t over-notify. A message for every tiny step trains people to ignore them. Notify on finished, failed and waiting for approval; keep the rest in a log.
What mistakes should you avoid?
- Technical jargon: “tokens”, “tool calls” and error codes mean nothing to most users.
- Fake progress: a bar that races to 90% and then sits there feels worse than an honest “still working”.
- Silent finishes: if nothing changes on screen when the task ends, people assume it is still running.
- Hidden approvals: a request buried in a log or a tab nobody opens.
- Overselling: saying “Done!” when parts failed. Be honest about what didn’t work.
If you are teaching an agent new routines, our guide on teaching an AI agent a task by showing it once pairs well with this one: clear steps make clear status messages much easier.
Frequently asked questions
What should an AI agent status message say?
What the agent is doing right now, in under about ten plain words, referring to the user’s own task, for example “Checking your last three invoices”.
Do I need an animated character like Muse?
No. Muse uses an animated character, but clear words, a progress indicator and an obvious “waiting for you” state do most of the job.
How often should an agent update the user?
Update the on-screen status whenever the step changes. Only send notifications when the task finishes, fails or needs the user.
Should I show the agent’s full reasoning?
Not by default. Most users want the simple status. Keep detailed logs or reasoning behind a “Show details” link for people who want them.
How do I show status in a Zapier automation?
Update a status field in the app the automation writes to, send a short message when it finishes or fails, and use an approval step before risky actions.
The bottom line
To show users what your AI agent is doing, use five clear states, describe each step in plain words about their task, make approval requests impossible to miss, and finish with a short receipt. Your next step: list the steps of one agent or automation you run today, and rewrite each one as a message a non-technical friend would understand at a glance.
Related posts
- How to Add Human Approval Steps to an AI Agent Workflow
- How to Teach an AI Agent a Task by Showing It Once
- AI Agents and Productivity Theater: 4 Principles to Make Them Do Real Work

[…] How to Show Users What Your AI Agent Is Doing, Without Technical Jargon […]
[…] Showing what the assistant is doing, with a simple icon, is a small touch that builds trust. We cover the same idea for software in How to Show Users What Your AI Agent Is Doing, Without Technical Jargon. […]