He Manages AI Like Employees; Commenters Tell Him to Let the Bot Clock Off

Dep posted a template for a multi-bot collaboration workflow on X, titled “the only bot i talk to.”

The structure isn't complicated: one lead bot, with four function-specific bots underneath. Scout handles information gathering, coder writes code, metrics looks at data, and reviewer does review. Dep only talks to the lead bot; the lead assigns the rest of the tasks.

Every morning at 07:30, after the reviewer finishes running, the lead bot has to deliver a briefing with four sections:

1. Decisions I need to make today, max 3, with drafts already prepared
2. What has changed in my core metrics, limited to what the reviewer has approved
3. New developments in my space, also limited to what has been approved
4. Tasks that failed overnight, and why they failed

Hard requirements: it has to fit on one screen, and every item must include a link.

During the day, tasks are assigned to the corresponding bot based on ownership. Dep emphasized one point: when handing off a task, you need to give the goal, context, and “why it needs to be done,” not just throw over a file. If no one owns a task, tell him directly.

After this thing was shared, the comments went in a different direction.

Someone replied: I took another path. My bots clock off at 23:00. What they find at 3 a.m. won't die just because it waits until morning. I'd rather read something fresh in the morning than wake up to a backlog from five bots.

Another reply said the 07:30 briefing schedule sounds like a nightmare.

There's a mismatch here. Dep's design is not essentially a technical solution; it's a management solution. Its assumption is that AI agents should work around the clock like human employees: if something goes wrong, someone has to backstop it; if something produces results, it has to go through a reviewer.

The netizen who told his bots to clock off was poking at another point: much of what overnight automation produces is actually noise. What bots find at 3 a.m. will most likely still be the same things at 9 a.m.; all that's added is the illusion of “processed.”

But one point in Dep's design is worth calling out separately: when handing a task to a bot, pass along the goal, context, and reason, not just a file. This is hedging against an old problem: agents only execute, they don't understand. No matter how finely you break down a task, if you don't give the why, the bot can only spin its wheels on the literal wording.

There's another detail: every item in the briefing must include a link. This is a clumsy way to prevent AI fabrication, but it works.

As for the 07:30 time, a reasonable guess is that he's personally in the habit of getting up early. For other people, clocking off at 23:00 may be the more realistic version.