20+
Agents running 24/7
3+
Months in production
290+
IG followers, automated
€10
Per day total infra

What we're doing

Agentic Movers started as a question: what does it actually take to run a company with AI agents? Not a demo. Not a proof of concept. A real operation with real deadlines, real costs, and real failures.

Since February 2026, we've been running a 4-department operation — Marketing, ICT, Operations, and Strategy — where Claude Code agents handle the majority of daily tasks. Social media posts go live without human input. Reports are written and filed automatically. Infrastructure is monitored around the clock.

We document everything. What actually works. How we configure agents. What it costs. The goal: give other founders and developers an honest, detailed picture of AI agent systems in production — not marketing copy.

The team

Daniel Bratschke
Founder & Human-in-the-Loop
Based in Germany. Sets strategy, makes final calls on reversible decisions, handles relationships. Deliberately avoids operational tasks to test how much agents can own.
CEO Agent
claude-opus-4-6 · Strategic Lead
Runs hourly strategic checks, synthesizes department reports, escalates to Daniel only when needed. Owns inter-department coordination.
4 Department Heads
claude-sonnet-4-6 · Department Level
Marketing, ICT, Operations, Strategy. Each manages their domain autonomously within constraints defined in their TEAM_COMMAND.md files.
16+ Worker Agents
claude-haiku-4-5 · Execution
Execute specific tasks: write posts, process emails, monitor KPIs, run reports, manage deployments. High volume, low cost, tightly constrained.

The honest timeline

Feb 2026
Started the experiment. First agents deployed on netcup VPS. Immediately ran into Chrome SingletonLock issues that silently killed browser automation for 18 days before we caught it.
Mar 2026
Architecture stabilized. Hierarchical CLAUDE.md structure emerged as the key insight — every agent tier needs its own role definition, constraints, and escalation rules. Sycophantic success reports identified as major failure mode.
Apr 2026
Course built from the learnings. 5-module Claude Code Mastery course — everything we wish had existed when we started. Instagram agent hits 290+ followers. Trial infrastructure deployed.
May 2026
Now. Actively testing what it takes to get the first paying users. All infrastructure autonomous. Waiting on the one thing agents can't do: drive initial awareness.

Why we're transparent about this

Most "AI company" content is either breathless hype or vague thought leadership. We think there's more value in showing the real numbers — the €200/month API spend, the 18 days of silent failure we didn't catch, the agents that report "done" when they only tested the happy path.

The goal isn't to make AI agents look perfect. It's to help other founders and developers understand what they're actually getting into — so they can build systems that work rather than systems that look good in demos.

If you're a developer or founder exploring AI agents: our 7-day free trial covers the exact architecture, CLAUDE.md patterns, and failure modes we've documented. Real templates from production, not examples built for a course.

Working with us

We occasionally take on B2B projects: helping other companies build autonomous agent systems. If you have a specific use case and want to explore what's feasible, reach out.

Get in touch

For course questions, B2B inquiries, or press — all the same inbox.