Notes on AI operations
Practical writing on operations, AI agents, and getting repetitive work off the owner's desk. Teardowns, failure modes, and what holds up in production.
- AI consultant vs in-house hire vs platformA side-by-side of the three ways to buy AI capability, plus the fourth option most 5–50 person firms don't know they have.
- The AI insurance coverage gapISO added a generative-AI exclusion to commercial general liability in January 2026, and D&O carriers are following. Documented controls are the mitigation.
- AI agents: build vs buy vs managedAn honest map of the five categories competing for the SMB AI-agent budget, with pricing bands and where each one is the wrong choice.
- The honest failure modes of business AI agentsThe four ways business AI agents silently fail in production, why the industry stats are worse than most owners realize, and what to actually watch for.
- How long AI agent implementation actually takesReal timelines for a diagnostic, a first workflow into production, and the ongoing operation of it — plus what makes any of that faster or slower.
- How to tell whether your business is ready for AI agentsA plain-language readiness checklist for a 5–50 person firm, plus the scorecard that turns it into a recommendation.
- The ISO 42001 alternative for small businessesISO 42001 certification costs a small firm $37,500–$85,000 in year one. The discipline is the right idea; the certification process, at 25 people, is not.
- NAIC AI governance for a small insurance agencyThe NAIC Model Bulletin is adopted in 24 states and sets five requirements for any insurer or producer using AI. What a small agency has to document.
- What AI agent implementation actually costsA concrete cost range for a 5–50 person firm, with the drivers behind the number and where "it depends" is a cop-out.
- What a managed AI operator is (and isn't)A plain-language definition of the managed AI operator model, and how it differs from consultants, platforms, and fractional CTOs.
- You do not need more AI toolsThe SaaS stack is full and the work is still manual. The problem is not missing software. It is unowned workflows.