How do employees use AI agents day to day?
You don't need a new app for AI agents. People approve drafts in Slack, Teams or email, ask their own AI tool, and review what agents left in their systems.
Employees don't need a separate app to use an AI agent. They meet it in three places: a message in the chat tool or inbox they already use, asking them to approve a draft or decide an exception; their own AI assistant, which reads company context through a connector; and their systems of record, where the agent leaves drafts and updates. Much of the work starts from an event, not a prompt.
Why there is usually no new app
Buyers ask it early: what will people actually see once the agents are live? The short answer is less than they expect. A new app is one more place to check, and people already move between tools all day. A study of 137 people in 20 teams at three large companies found they switched between apps and websites nearly 1,200 times a day, and spent just under four hours a week getting their bearings after each switch (Harvard Business Review, 2022). An agent that adds a screen adds to that. An agent that works where the work already sits does not.
Three places people meet an agent
| Where | What the person sees | What the person does | Example |
|---|---|---|---|
| Chat tool or inbox | A message with a draft, the sources behind it and a clear choice | Approves, edits or rejects, or answers a question the agent could not settle | "Order from a customer: two lines match the quote, one price differs from the agreed terms. Approve the draft or check the price?" |
| Their own AI assistant | Answers grounded in approved company context, with sources | Asks questions the way they already do | "What did we agree with this supplier on lead times?" in ChatGPT, Claude or Copilot |
| Systems of record | Drafted entries, updated records and notes, each with a record of what the agent did | Reviews and carries on working | A deal updated after a contact changed roles, with a draft follow up waiting to be sent |
Much of an agent's work never reaches a person at all. Reading, matching, checking and drafting happen in the background. People see the steps that need a decision.
Do people have to prompt the agent?
Mostly not. There are two ways an agent's work starts:
- From an event or a schedule. A new order arrives, a contract changes, a contact moves to another company, or it is Monday morning. The agent picks it up without anyone asking. In our view this is where agents save the most effort, because nobody has to remember to start the work.
- From a question. Someone asks the agent in chat, or asks their own AI assistant, and gets an answer built from the same company context.
A related question is whether people have to keep the agent up to date. They should not have to retype what the agent can read. An agent that works from a shared picture of the company can record what it learns, with the source attached, so the next job starts from it. People correct it when it is wrong, and that correction becomes part of the record too. Why this shared picture matters is covered in why AI agents give confident wrong answers.
What changes for the people doing the work
The job moves from doing every step to deciding the ones that matter. Someone who used to retype orders now reviews the few the agent could not match. Someone who prepared every meeting by hand now reads a prepared brief and fixes what is off. That shift only works if three things are true:
- The requests are worth reading. An approval message should carry the draft, the sources and what is unusual about this case. If every request looks the same, people start clicking approve without reading. How to design these requests is in approvals and control for AI agents.
- The answers come from approved context. If ChatGPT, Copilot and the agents each read different sources, the same question gets different answers. See the company context layer.
- Someone owns the rules. When an exception keeps coming back, one person decides what the agent does next time. See who owns the AI platform after launch.
How this works with Ortelian
Ortelian is a platform plus a forward deployed team, and the setup is done with your team inside an engagement. There is no self-serve product. Ortelian's agents run the workflows you set up, or you ask them in Slack, and they work inside the systems you already use, such as your CRM, Slack, Teams and email. Your team keeps working where it works today, and Ortelian's own app is where agents, workflows and limits are set up and monitored. Your own AI tools, such as Claude, ChatGPT and Copilot, connect over MCP. Both reach your business only through scoped tools, and agents are read only until a write is granted. Actions that leave the company, such as sending an email or an invoice, wait for a person, and every action is recorded with the sources behind it. The platform is hosted in the EU. For how an engagement runs, see how we work.
Questions people ask
Do employees need training to use AI agents?
Less than for a new tool, because they keep working in the tools they know. What they need is to know what the agent does, what it will ask them, and who to tell when it is wrong. Start from real examples of their own work rather than general AI training. With Ortelian, workshops during the build prepare your team to own the work.
Can people use agents from Microsoft Teams?
It depends on how the agent is set up. With Ortelian, agents work inside the systems you already use, such as your CRM, Slack, Teams and email, and Copilot connects over MCP. Approval points are set before a workflow goes live.
Will the agent act without anyone noticing?
It should not act on anything that matters without a record. With Ortelian, every action is recorded with the sources behind it, and anything that leaves the company waits for a person by default. Your team can allow a specific action to run without approval.
How do managers know the agents are doing good work?
By checking the work against a standard agreed before launch, not by reading every output. With Ortelian, evaluations are built from real examples of your business for quality, cost, time and human effort. They set the threshold before launch and keep running after it, and agents, workflows and limits are monitored in Ortelian's own app.
Sources
- Rohan Narayana Murty, Sandeep Dadlani and Rajath B. Das, "How Much Time and Energy Do We Waste Toggling Between Applications?", Harvard Business Review, 29 August 2022.
- Model Context Protocol, introduction.
- Microsoft, extend your agent with Model Context Protocol in Copilot Studio.
