A two-person company. Seven days to autonomy.
A husband-wife team renting worker accommodation in the Ruhr region spent 90% of their time on booking logistics. In seven days, an agent cluster replaced the manual work. Response time fell from hours to minutes. The founders got their time back.
The problem
The business was simple: rent furnished apartments to industrial workers in the Ruhr region. The product worked. Demand was steady. But booking was eating the founders alive.
Every inquiry meant a phone call. Then a calendar check. Then a manual invoice in Word. If the apartment was already booked, they had to call back and start over. Ninety percent of the founders’ time went into logistics, not growth. Every new booking meant less time for everything else.
This is the canonical “what to automate first” scenario: structured data, high volume, repetitive decisions, clear rules. The founders didn’t need AI to replace their judgment — they needed AI to handle the logistics so they could use their judgment.
What we built
An agent cluster in Direct Mode. No middleware. No platform. No dashboards. Five agents, each with one job:
Inbox Watcher
Monitors the booking inbox. Extracts inquiry data: dates, number of workers, special requirements. Triggers the router.
Router
Classifies the inquiry: new booking, change request, cancellation, general question. Routes to the correct handler.
Availability Engine
Checks the calendar for matching apartments. Returns available options with prices. If nothing matches, triggers the escalator.
Pricer
Calculates the price based on duration, number of workers, and season. Generates the invoice draft in the correct format.
Escalator
When the system cancan’tapos;t resolve something — no availability, ambiguous request, special pricing — it escalates to the founders with the full context.
The key design decision: humans above the loop, not in it. The founders review escalated cases — about 15% of inquiries — and otherwise stay out. Each escalation trains the system to handle that case type next time.
The result
Key insight
You don’t need a big company to benefit from AI. You need a clear workflow, structured data, and someone who can define the rules. This two-person company moved faster than most 200-person companies because there was no coordination overhead. Direct Mode: founders decide, agents execute.
The escalation design is the architecture that makes it safe. The system doesn’t need to handle everything — it needs to handle the 85% that are routine and escalate the 15% that aren’t. That’s the difference between building an autonomous system and building a useful one.
What workflow is eating your time?
Start with the MIURA-KO Quick Check. Ten minutes. No commitment.
Take the Quick Check