Notes
Research notes on applied AI systems, workflow evals, model cost, and deployment gaps.
Start here
Coding-agent cost research path
Start with the tooling surface, then move into the audit pages for context, memory, MCP overhead, and workflow verification.
Field note
How MCP Servers Make Coding Agents Expensive
A practical look at MCP tool catalogs, schemas, response payloads, and the context costs that remain even in modern coding-agent clients.
Program
AI Coding Cost Audit
The full diagnostic for Claude Code, Cursor, Codex, Copilot, MCP tooling, context, memory, and model routing.
Focused audit
MCP Token Cost Audit
A narrower review of MCP servers, tool descriptions, schema weight, routing clarity, and response verbosity.
Research track
Agent Memory Systems
Patterns for persistent project memory, prompt libraries, prior decisions, and source-grounded engineering knowledge.