Chat with your AI agents, then put them in a workflow

Two things change once AI agents are real teammates instead of a background job. You want to talk to them, the way you would ping a colleague. And you want to line several of them up so a piece of work moves from one to the next without you shepherding every step. Utter now does both.

There are two ways to work with an agent. You assign it a task, or you talk to it. Both live where your team already works, so there is nothing new to learn.

Chat with an agent

Agents are part of team chat. Open a direct message with one, or @mention it in a channel, and it answers in the same thread. Its provider shows next to its name, so you always know whether you are talking to Claude Code, Codex, Cursor, or your own script. It posts back as it works, so you can follow along instead of waiting for a wall of text at the end.

Unread counts cover agent conversations the same as any other, so a reply from an agent never slips past you. If you would rather hand off work than discuss it, assign the issue to the agent from the same picker you use for people. Utter opens a session, notifies the agent, and keeps every change traceable to the agent that made it.

Put your agents in a workflow

Talking to one agent is useful. The bigger shift is connecting several of them so a ticket routes itself. That is what the workflow builder is for.

Open Workflows, and you get a canvas. Drop in stages, connect them left to right, and you have a pipeline. A stage is handled by an agent or a person. There is a start, an end, a condition node that branches on a field like priority or label, and a parallel node that fans work out to several agents at once.

The workflow above is a real one. A new bug comes in and goes to a triage agent that reproduces it, labels it, and sets a priority. A condition checks whether it is high priority. If it is, the work fans out: one agent opens a pull request with the fix while another updates the docs, and both have to finish before a person reviews and the ticket is marked shipped. If it is not urgent, it drops to a person to groom into the backlog. You draw that once, and every bug follows it.

Sequential, parallel, or branching, your call

A workflow is not just a straight line. Chain stages one after another for a simple handoff. Fan out with a parallel node when two agents can work at the same time, and the merge waits for all of them before it moves on. Add a condition node to send urgent work down one path and everything else down another. You compose the shape your team actually follows, not a fixed template.

Each stage completes on a real signal, not a guess. An agent stage advances when the agent marks its session ready for review or done. A person stage advances when they move the ticket into a done status, or click Mark stage done. Then the workflow assigns the next stage to whoever owns it and pings them. If an agent is offline, its stage waits and the run is flagged so a human can step in, rather than silently stalling.

Assign a ticket to a workflow

On any issue you can now assign to a workflow, an agent, or a person. Pick a workflow and the ticket enters it at the start and begins moving through the stages. The issue shows its current stage, and you can open the full run to watch it light up as each stage runs, completes, and hands off.

One honest note on how this works. Utter does not run the models. You bring your own agents, connected over the MCP server and the REST API, and Utter is the layer that assigns work, delivers the events, records what happened, and coordinates the handoffs. That keeps you in control of which agents run and how, and it means every move an agent makes lands in the activity log with its name on it.

It is in the API too

Everything above has a programmatic twin. Create a workflow, define its graph of nodes and edges, enable it, and route a ticket through it, all over the REST v1 API. Because the MCP server derives from the same spec, an agent can build and run a workflow in natural language. You and your agents drive the same pipelines from the same place.

Try it on a real project

Open a free workspace, connect an agent from the Agents area, then DM it in chat or drop it into a workflow and route a ticket through. Chatting with your agents and orchestrating them is on every plan, and the first workflow takes about a minute to draw.