Notes from the field

Writing on AI adoption, governance, and building with Claude Code.

Sep 2026 • AI product
The AI Integration Trap: Why More Automation Doesn't Mean Better Products
Most AI features don't fail because of the model. They fail because they take the wrong amount of control. A 1978 framework for choosing how much autonomy AI should have.
Read
Sep 2026 • AI practice
How to Avoid Slop (Content, Visual, Ideation), and Why You Need to Ask the LLM to Argue With You
Why default AI output trends toward generic, agreeable slop, and the technique (asking the model to argue against you) that fixes it.
Read
Sep 2026 • Licensing
How to Manage AI Licenses (Shadow AI, Onboarding/Offboarding, and Empty Seats)
Shadow AI, slow onboarding/offboarding, and empty seats are SaaS license sprawl's problems again, faster and higher-stakes.
Read
Sep 2026 • Governance
How to Enable Citizen Developers (Without Losing Control of Your Stack)
How to let non-engineers keep building their own AI tools without losing visibility into what touches sensitive data or who owns it.
Read
Sep 2026 • AI risk
Nobody Understands the Product Anymore, and That's the Real AI Risk
When AI writes the code and hands back the task breakdown, understanding the product can quietly become optional, an accountability problem, not just a tooling one.
Read
Aug 2026 • Claude Code
What Claude Code Actually Costs You (And How to Cut It Without Cutting Quality)
What actually drives Claude Code's token costs and eight concrete habits that cut spend without cutting quality.
Read
Aug 2026 • Claude Code
Why Claude Gets "Dumber" the Longer Your Project Runs (And How to Fix It)
Why long-running Claude Code sessions degrade, and the file-based memory and codebase-splitting fixes that actually address it.
Read