March 2026

• What It Actually Costs to Run AI on Your Own Hardware — March 30, 2026
  Everyone keeps saying, “just run AI locally.” Let’s put some numbers against that. If you want to run a decent local model – not toy models, but something in the 14B to 70B range – you are stepping into real infrastructure territory. Here’s what that actually looks like today. Option 1 – NVIDIA GPU (performance-first)

• Safer Experimentation With Coding Agents: Why I Built SafeAgent.ca — March 23, 2026
  Coding agents are getting more capable, and more people want to try them on real repositories. The problem is simple: curiosity moves faster than caution. A lot of early experimentation happens in the worst possible way. Someone points an agent at a working copy, gives it network access, and hopes for the best. That may

• When Institutions Drift, Systems Emerge — March 4, 2026
  Viewed as a system, the story is familiar. An institution encounters a change at the edges. People closest to the activity notice it first. They suggest adjustments. Leadership listens politely but remains anchored to the existing model. Nothing changes. Eventually, the people pushing for change stop pushing. They build something else instead. That pattern played