We replaced five roles with two people and 350 tools
A working account of putting AI agents into a real business, including what they took over, what we refused to let them touch, and the one rule that keeps a human in every outbound decision.
For Founders and operators running service businesses of 10 to 100 people
- The read versus write rule that decides what an agent may do unsupervised
- Which parts of a pipeline are assembly work and which are judgment
- What actually broke, and what it cost to fix
Services don't scale, and other things I believed for ten years
The rule that services cannot scale was never about services. It was about founders. A decade of running a tech company, and which constraints turned out to be self-inflicted.
For Founders, particularly bootstrapped and service-based
- Why founder dependency is a stage rather than a property of the model
- How to make expertise structural so people leaving is a setback, not a hole
- The growth ceiling word of mouth builds, and the three ways out of it
Building a tech company outside Silicon Valley
What it takes to build durable technology businesses from Albania, without venture capital, where cash flow is the measure rather than a funding announcement.
For Founders and ecosystem builders in emerging tech markets
- Why the no-capital constraint produces better unit economics
- Hiring and keeping senior engineers outside the established hubs
- What regional ecosystems need more of, and what they have too much of
What breaks when you put AI agents into a real business
The unglamorous version. Approval queues, permission scoping, the failure modes nobody demos, and why the leverage comes from removing assembly work rather than removing people.
For Technical leaders and operators evaluating agentic systems
- Designing the approval surface before designing the agent
- Why scheduled playbooks beat chat interfaces for operational work
- Honest failure modes from a system running in production
