Every agency has the same hidden problem: work arrives in chats. A client writes "please make the banners by Friday" between two other messages, someone reacts with a thumbs-up, and three days later nobody remembers who owns it. Chats feel like control, but they work like a sieve.
We decided to treat task control as what it is: a repeatable function that does not need empathy or creativity. It needs a perfect memory. So we connected an AI agent to our work chats.
What the agent does
- Finds tasks in conversations. It reads messages in client, team and contractor chats and decides whether a message is a task. Ordinary talk is ignored.
- Sets the deadline and the owner. "By Friday" becomes a dated task, assigned to the person responsible for that project.
- Keeps the history of every project. Which tasks were set, done and overdue, and by how much.
- Watches the tone. If a client's messages turn negative, the manager gets a notification to step in before the client decides to leave.
- Reminds people. When a deadline approaches and nothing has been said about the task, the agent writes to the assignee directly and keeps asking until the task is closed.
- Reports. Each morning the manager receives a summary: project health scores out of 100, tasks for the day, what the team did yesterday.
Why it works better than a person here
A project manager in this role spends hours re-reading chats and nudging colleagues. Before the agent, our manager spent roughly 1.5 to 2 hours a day on it, and a person dedicated to the job spent about 10 hours a week. Even then, tasks slipped, because the flow of messages is simply too large for one human.
Chasing people is also unpleasant work. It leads to fatigue and burnout, and it pulls managers away from strategy. The agent does not tire, forget or avoid the awkward reminder.
What changed for us
- The manager reads one summary a day and gets a complete picture of all projects in a minute. That frees about two hours a day for strategy and for training the team.
- Overdue tasks used to sit for five to seven days before someone noticed. Now an overdue task is usually closed within one or two days, because the agent follows up on its own.
- Clients get faster reactions and fewer forgotten requests, which is what keeps clients.
How to start in your own company
You do not need to replace anyone to benefit from this. Start with the most repetitive control function in your team, describe the rules (what counts as a task, who owns what, what a late task triggers), and let the agent handle only that. Keep a human for decisions and for conversations that need judgment.
If you want to explore what this would look like for your team, request a free consultation. We built this system for ourselves first and know where it helps and where it does not.
