---
description: 15 October 2026 · Thoughts
url: https://www.schmalzl.tirol/en/blog/2026-10-15-ein-mensch-viele-agenten
lang: en
date: 2026-10-15
deutsch: /blog/2026-10-15-ein-mensch-viele-agenten.md
---

# One person, many agents

_15 October 2026_

"I built an office for my AI agents." That is how an Instagram post begins that I made a note of this week. It shows an office with **8 teams and 31 AI agents** that share a common memory. One team searches GitHub for new things every morning. One writes texts. One edits videos, with sub-agents working in parallel. One plans the posts for social networks.

One sentence in it stuck with me: nothing bigger goes out without a human OK.

## From assistant to team

An **agent** is an AI that does not just answer, but carries out steps itself: reading files, searching the web, writing code, checking results. For a long time, that meant one agent for one task. Now they are becoming teams.

There are ready-made tools for this by now. [oh-my-claudecode](https://github.com/Yeachan-Heo/oh-my-claudecode) lets several agents in Claude Code work together as a team. [Multica](https://github.com/multica-ai/multica) wants to turn people and agents into one shared team, self-hosted. And in a video, the AI expert Allie K. Miller explains how a single person runs 34 agents.

## The work shifts

If you have many agents, you write less yourself. Instead, you have to define more:

- What should come out at the end?
- Who does which part?
- How do you tell that a result is good enough?
- When does a human have to decide?

The crucial question is: who decides the next step? If the goal is missing, the AI has to guess. With 31 agents, it guesses 31 times.

## My small example

I work like this myself, just on a small scale. For this blog there is an [editorial office with an AI editor](/blog/2026-10-11-mein-ki-redakteur). I send links to a collection inbox. Claude reads them, checks the sources and writes drafts. I see every draft exactly as it would appear in the blog, and I decide: approve, revise or discard.

This text was created the same way. It will only be published when I click "Approve".

## My conclusion

Agent teams take a lot of legwork off your hands. But they do not replace judgement. The more agents work on something, the more important the point becomes where a human says: this is right, this goes out.

Sources: [Instagram @bexruz.khamidullaev](https://www.instagram.com/reel/DeV1fyFiA8A/) · [oh-my-claudecode on GitHub](https://github.com/Yeachan-Heo/oh-my-claudecode) · [Multica on GitHub](https://github.com/multica-ai/multica)