Blog
The economics of AI coding agents
Real numbers, runnable snippets, honest limitations. Subscribe via RSS.
What a coding-agent session actually costs, request by request
A single prompt to Claude Code or Codex becomes many billed API calls, each re-sending the whole conversation. Here is the anatomy of that bill: the four token classes, their price multipliers, and the six things that make one session cost ten times another.
The number your developers see is not the number you pay
Claude Code computes its cost figure locally, at list price, from one machine's history. Four documented ways that diverges from your invoice — and what a savings claim needs to survive a finance review.
Build or buy an LLM gateway for coding agents: the five questions that decide it
Most teams routing Claude Code and Codex through a gateway are choosing between LiteLLM, Bifrost, a cloud vendor's proxy and their own. The decision turns on five properties, and one of them — byte fidelity — quietly decides your cache bill.
How Uber cut AI coding costs 52% — the playbook, applied
Uber grew AI coding usage 7× while unit costs fell by half. Their levers are public, mechanical and mostly configuration. Here is each one, what it is worth, what it costs to adopt, and the case against adopting it blindly.