Imagine your agents had a channel to your people
Most of the work in an agent team is done by agents. Some of it is not theirs to decide: whether to send the message, whether to spend the money, whether this is the right fix. Today that part is awkward. Here is a way to think about it, an idea and not a feature, and then what harnsy actually does now.
Key points
- 1
- person who answers your agents’ questions in v0.9.1: you
- 4
- things the idea asks of a channel between agents and people
- 1
- real question in your project to name an answerer for
In this article
Your team is agents and people
We call it “an agent team”, but it never was only agents. Agents write, check, build and plan. People decide the things that are theirs: what the company says in public, what money is spent on, what the law allows, what is good enough. If you leave the people out of the picture, the picture is wrong.
A real team has both, and the work moves between them all day.
Imagine a channel
Imagine your agents had a channel to your people. Four things would be true.
- The agent starts the talk. It does not wait for you to open a window and ask how it is going. When it is stuck on a decision that is not its own, it says so.
- It holds a dialogue. One question, one answer, and then another question if the first answer opens one. A conversation, not a form.
- It reaches the right person. A question about a contract goes to the person who answers for contracts. A question about the release goes to whoever owns the release. Not to whoever happens to be online.
- It knows who holds which job. The team has a plain list: this person answers for this. The agent reads it instead of guessing.
None of this needs a clever agent. It needs the team to be written down: who answers what.
Why a name on the question is not enough
A channel that only delivers is not the whole idea. Builders of scheduled jobs describe the trouble well: a yes tapped from bed at night, a job that runs at six in the morning on data that has changed since. The right person said yes. The yes was about something that was no longer true.
So the idea has a second half. The answer has to belong to a question, and the work has to wait for it. An agent that asks, then goes on, has asked nothing.
A question for you, now
Take one real question from your project. Not a grand one: the last time an agent stopped and waited for a person. Now write down who should have answered it, by name. If you cannot name one person, that is the finding. The channel would not fix that. Deciding who holds the job would.
What works today, in v0.9.1
This part is not an idea.
- Your agents ask you in Telegram, with buttons, and wait for your answer.
- Teammates you link can message your project’s lead.
That is the whole of it. The person who answers an agent’s question is you, the owner. A linked teammate can write to the lead, but cannot answer the agent’s question for you. There is no list of who answers what yet, and a question does not go to a particular person.
So the picture above is where we think a team should be able to go. It is not where harnsy is.
Run your agents as a team
harnsy connects your Claude Code, Codex and OpenCode agents into one team with roles, and asks you in Telegram when it needs an answer.
Install harnsyharnsy is not affiliated with Anthropic, OpenAI or OpenCode; their names are theirs.