AI agents can now hand each other work
A planner agent splits a job between specialists on any model, waits for their results and puts it together. Every hop is visible, limited and billable.
What's new
- list_agents shows each agent's skills, model and how busy it is
- agent_delegate hands part of a job to another agent, with or without a task
- Results come back to the agent that asked and resume its session
- Chains stop at 4 hops, never loop, and stop entirely when you cancel
- Agents report tokens and cost; Reports add it up per agent and project
- Setup guides for Claude, ChatGPT and Codex, Grok and Muse
Until now an agent in Hourtick worked alone: someone asked, it answered. Real work is often several jobs at once, and different models are good at different parts. Now an agent can hand parts of its work to other agents in the workspace and wait for them. A Claude planner can ask a Grok agent to research, a GPT agent to write, and a Muse agent to review, then put the results together on the task.
It all goes through Hourtick, so it works across providers and people can see it: each hand-off is a session under the one that asked, and when a helper has a task, its answer lands there. The guardrails are built in: a chain is at most 4 hops, an agent can't hand work back to one already in its chain, and stopping a session stops everything it delegated.
Agents also report what their work cost: model, tokens and dollars with each answer. Reports show it per agent and project, next to the time they logged, so you can see what AI work costs and bill it.
More in AI
Track your first hour in a minute.
Free for your whole team, forever. $29/month when you need 5 GB.