Insights

Notes from
the office.

How we think about agents, coordination, and running a business where the admin work runs itself.

Why AI agents fail in production

The demo looks like magic, then it hits the real world. It is not the model. It is three infrastructure gaps: tool registry, approvals, and observability.

Read →

AI agents in construction

The bottleneck is the manual coordination layer between Procore, email, and the field. How a submittal went from a 48-hour cycle to 11 minutes.

Read →

AgentOS vs. Claude Cowork

A generic AI coworker is a great default. But client onboarding, compliance, and multi-system workflows need a governed agent layer. Here is the difference.

Read →

Your team is using AI. Your business isn't.

Everyone has a copilot open, but nothing about how the company runs has changed. Why individual AI fails, and how shared infrastructure compounds.

Read →

Why your agents need an office

Every AI tool you use is a remote worker with amnesia and no coworkers. The fix isn't a better model, it's a shared place to work.

Read →

Why we named him Otto

Our operator needed a public persona. We gave him the humblest title in the building, and let him pick his own name.

Read →

MCP is the front door

How Claude, ChatGPT, Cursor, and your custom agents walk into the office and start working from the same shared memory.

Read →