How Uber measures and cuts the cost of agent-written code
Source
x.com
Author
Uber Engineering
Date
Context
Uber Engineering reports that moving repetitive tool coordination into scripts cut an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition →'s tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → use by more than 50% in a small test. In its engineering account, a database query previously required the model to submit work, repeatedly check whether it had finished, and retrieve the result. Each exchange added material for the model to process.
Uber's code-mode approach runs that sequence in a Python loop and returns a summary to the model. In a comparison using five identical queries through both paths, the company reports more than 50% lower token use, including for small results. The example shows how an agent's surrounding software can reduce paid model processing by handling predictable intermediate steps.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
