The Rough Edges of Upgrading a Self-Hosted AI Agent
A firsthand upgrade exposed how database bloat, live state, schema changes, and indexing can turn a package update into a full maintenance event.
Topic
intermediate
5 related items across notes, deep dives, spotlights, and projects.
How AI Agents Decide Which Tool to Use
Tool selection is not a fixed lookup. An agent weighs the task, available tool definitions, required arguments, and prior results—then revises its next move when new evidence arrives.
Why More Context Can Make an AI Agent Worse
A larger context window gives an agent more room. Filling that room with stale instructions, duplicated facts, and raw tool output can make the agent less reliable.
Why AI Agents Need Sandboxes
A sandbox limits where an agent can act when code, tools, or instructions go wrong. It is containment—not permission, judgment, or trust.
Judge AI by Results, Not Vibes
A practical way to tell the difference between AI that sounds impressive and AI that actually works.