Tell people what the agent will do, what it will not, what changes in their own job, and who is still accountable when it gets something wrong. Brief managers first, then the affected team, then everyone else, before go-live. Log every question and answer it in writing, and say plainly what is not yet decided.
What is the communication for, and who owns it?
Staff hear about agents from the news, from rumours and from vendors' marketing. Without a clear account from the firm, they fill the gap with the worst version. Good communication gives people the facts they need to work alongside the agent, to trust its limits, and to raise problems without fear.
The process owner owns the message for their team, because they can answer questions about the work. HR and internal communications support. Senior leaders open the conversation. Staff representatives, where your firm has them or local rules require consultation, are involved before announcements, not after. A ready structure is in the staff communication plan template for an AI agent rollout.
What should every message cover?
- What the agent does, in terms of tasks, not technology: "it reads incoming direct debit failures and drafts the letter".
- What it does not do, and what stays with people.
- What changes for each role: new tasks, removed tasks, approvals people will now give.
- Who is accountable. People who approve held actions are accountable for those decisions; the process owner answers for the agent overall.
- How to report a problem with the agent, and that doing so is expected.
- What it means for jobs, as far as is known. If something is undecided, say so and say when it will be decided.
- Timeline and training.
What are the steps, and who acts at each one?
- Person, process owner, with HR: writes the core message and the questions-and-answers sheet.
- Agent, depth 2: drafts role-specific versions from the core message and the capability list, one per affected role. The owner edits and approves each.
- Person, HR: consults staff representatives where required, and records their input.
- Person, senior leader: briefs managers, who hear it first and can ask questions before their teams do.
- Person, process owner and team managers: brief the affected team in person, showing the agent working on real cases in shadow.
- Person, internal communications: informs the wider organisation.
- Agent, depth 1: collects questions from a shared channel and the briefings into one log, grouped by theme.
- Person, process owner: answers each question in writing, updating the questions-and-answers sheet.
- Person, process owner: runs follow-ups at go-live and after the first month, sharing results and what changed.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Core message | Nothing | none | Writing what is true |
| Role versions | Drafts per role | 2 | Editing and approving |
| Consultation | Nothing | none | Engaging representatives |
| Managers | Nothing | none | Leader's briefing |
| Team | Nothing | none | Owner's briefing and demonstration |
| Wider staff | Nothing | none | Communications |
| Collect questions | Logs and groups questions | 1 | Nothing |
| Answer | Nothing | none | Written answers |
| Follow-ups | Nothing | none | Sharing results |
What evidence does each step leave?
The approved core message and question sheet; each role version with its approval; consultation records; briefing dates and attendance; the question log with answers and dates; and follow-up notes. Keep them. When a dispute or audit later asks what staff were told, the record answers it. It also shows the next team to receive an agent what questions to expect, so their briefing starts better prepared.
Why does accountability matter more than jobs?
Most worry is about blame, not redundancy: "if I approve what the agent drafted and it was wrong, is that on me?" Answer it directly. Explain what the approver sees, what they are expected to check, and how mistakes in the agent's draft are treated. The argument is made in staff resistance is usually about accountability, not jobs, and it is the question to prepare for most carefully.
People who will approve held actions need to know how many items to expect and how their time is planned; how do you staff the queue of actions AI agents hold? helps managers answer that before the queue exists.
What should you measure?
- Attendance at team briefings.
- Questions raised, by theme, and time to answer.
- Problem reports about the agent after go-live. Few reports early on usually means people are unsure whether to report, not that nothing is wrong.
- Short pulse checks on understanding and confidence, before and after.
Where does it usually go wrong?
The announcement precedes the plan. Leaders announce agents before anyone knows what changes. Questions go unanswered and trust drains.
Promises that cannot be kept. "No one's job will change" is rarely true. Say what is known.
Technology talk. Staff hear about models and platforms when they want to know about Tuesday's queue.
Communication ends at go-live. The first weeks bring the real questions. Keep the channel open and report back.
No training follows. Telling people the agent exists does not teach them to work with it; see how to train staff to work alongside an AI agent.
Where Pryme fits
In Pryme, agents work with people in a shared Workspace, where staff see what the agent drafted, why an item was held and which bound decided. That makes it easier to show staff exactly what the agent can and cannot do. The rollout path is in deploying an AI agent: train, shadow, certify, widen.
Questions people ask
Should staff see the agent before it goes live?
Yes. Show it working on real cases in shadow, including a case it refuses or escalates. Seeing the limits in action does more than any description; the wider approach is in how do you get staff to accept AI agents?
Who should announce the rollout?
A senior leader opens it to show it matters, but the process owner and managers do the detailed briefings, because they can answer questions about the work itself.
What if staff representatives object?
Listen and record their concerns, change the plan where they are right, and explain in writing where you are going ahead and why. Consultation that changes nothing is noticed.