-
The Five I Actually Run
Not a trending roundup. Five tools that turned the principles this blog has argued for into software I open every day, each one tied to the post that made the case.
-
Know When to Stop
The skill that separates a senior from an eager junior is knowing when the model is the wrong tool. Availability is not fitness. Four times to put the prompt down.
-
Fix the System, Not the Output
Correcting what the AI produced fixes today. Correcting how it works fixes every tomorrow. Re-explaining context is not just slow, it lets the goal drift. Capture the decision once.
-
Don't Just Use the Model
The series says the developer owns the decisions. But a decision about a machine you treat as magic is a guess. Three real calls that came from knowing how the model works, and three readable books to get you there.
-
Did It Actually Work?
Working with AI assumes output you can trust. This post measures it. Velocity is a vanity metric now. Trust is the number that survives, and three health metrics tell you whether the loop is working.
-
Build Your Own Tools
A tool that used to cost a sprint now costs an afternoon. The decision of what to build is the bottleneck. A direct challenge: pick one task you did twice this week and have a custom tool for it by tomorrow night.
-
Many Agents, One Chat
Once you run more than one agent locally, you become the bottleneck. A chat interface plus a vault is the upgrade. Why agentchattr and Obsidian pair so well for the personal scaling problem.
