Agent-based AI coding costs more than it looks, and the tooling around the model matters

A developer found that chat-based AI coding hides costs, since every file change requires manual copy-paste and context re-sharing. Agent tools instead let the model run tests and fix its own errors, with a surrounding harness supplying files and a terminal. JetBrains' 2026 survey of over 15,000 developers reported that 90% use AI coding agents at work at least weekly. Artificial Analysis data showed one Claude Code run took 146 turns, about an hour, and roughly $11.42 at API prices. The author argues the harness shapes results as much as the model, though the comparisons cited mix both variables and do not isolate the harness.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.


Discussion (0)
Log in to join the discussion and vote.
Log in