Posts
Your AI Bill Is Mostly a Context Bill
Many agent loops pay for the same instructions, history, and tool definitions on every model call. Here is how I think about paying once instead: cache the stable prefix, keep it stable, trim what does not need to be there, and measure all of it.
- AI
- Developer Tools
- Agents
- Cost
Consumer AI Apps Need Browser-style Profiles
Consumer AI apps treat you as one person with one grant per connector. I want built-in profiles, like a browser: each profile owns its connectors, chats, memories, and instructions, and a small global layer holds the things that are actually me.
- Agents
- Developer Tools
- Product Engineering
- MCP
A Release Pipeline With No Long-lived Tokens and No Silent Tampering
How I secured the Hookie CLI release pipeline with OIDC trusted publishing, staged npm releases behind 2FA, and immutable GitHub release assets.
- Supply Chain Security
- npm
- GitHub Actions
- CI/CD
A CLI Should Do More Than Wrap Your API
Notes from building Hookie: API-shaped commands are useful, but AI-native dev workflows need CLI commands that own the whole task.
- Developer Tools
- CLI
- Product Engineering
Shipping Code Faster Is Not the Bottleneck
Agent-loop discourse sells parallel code output. For solo builders the binding constraint is usually demand, and a useful receipt needs cost, artifact, and a user-visible before/after.
- AI
- Agentic Programming
- Productivity
- Thinking
The Interview Gap
The industry says AI agents will write most of the code. Senior engineering interviews still gate on whether you can produce a graph traversal cold in 45 minutes. A checkpoint piece, four months into the job search, on the gap between the two.
- Product Engineering
- Engineering Interviews
- Thinking
- AI
