Category
AI
- ·3 min read
MCP vs CLI: The Token Cost You're Not Tracking
Every MCP tool call ships your entire tool schema to the model. On a long session that's thousands of tokens before you've done anything. CLI tools don't have that problem.
- ·3 min read
AI Agents That Work: Give Them Structure, Not Just Prompts
Agents don't fail because they're dumb. They fail because they're given ambiguous context. This is how the four tools in this series eliminate that ambiguity.
- ·3 min read
Beads: A Local-First Task Graph for Developers and AI Agents
GitHub Issues is great for tracking work across a team. Beads is a Dolt-backed task graph for the local development loop — and for the AI agents running inside it.
- ·3 min read
OpenSpec: Write the Contract Before the Code
Freeform prompts to an agent produce freeform code. Three commands and a human review step change that. Misunderstandings surface at proposal review, not after an hour of wrong code.
- ·3 min read
Git Worktrees: Branches as Places, Not States
One repo, three running apps, zero context switches. Git worktrees have been in git since 2015. Almost nobody uses them. Here's why that's a mistake.
- ·3 min read
Why Agentic Coding? It's Not About the AI
Agentic coding isn't a smarter autocomplete. It's a workflow shift — and the structure you build around the agent is 90% of the value, even without the AI.
- ·3 min read
Not on Claude? The Cross-Tool Configuration Guide
The concepts in this series aren't Claude-specific. Every major AI coding tool has a configuration layer. Here's how they map — and what Windows and Linux users need to know.
- ·3 min read
Project Settings, Permissions, and Team Sharing
What your team shares, what stays personal, and how to stop the 'Claude asked me to approve running npm install' prompts that break every flow.
- ·3 min read
Hooks That Pay for Themselves
A session-start hook runs once, costs nothing, and gives your agent context it would otherwise ask for or guess wrong. The highest-leverage 10 lines you'll write this week.
- ·3 min read
Writing CLAUDE.md That Agents Actually Follow
A 1,500-line CLAUDE.md doesn't make agents more consistent — it makes them less. The agents that work best have short rules backed by tooling, not long rules backed by hope.
