Show HN: Uber burned its 2026 AI budget by April. Why? 73% redundant reads
A recent study revealed that 73% of the tokens used by AI coding agents were spent on redundant file re-reads. This inefficiency was particularly pronounced during longer coding sessions, where the agent would repeatedly read the same files without any changes. The findings suggest that the structural limitations of current AI architectures contribute to unnecessary costs for developers.
- ▪The study tracked AI coding agent usage over 30 days, revealing a significant amount of redundant token usage.
- ▪In longer sessions, the re-read percentage increased to 78%, with the agent reading the same files multiple times.
- ▪The projected annual cost of this redundancy for a team of 25 engineers could reach approximately $99,000.
Opening excerpt (first ~120 words) tap to expand
FIELD STUDY · AI CODING AGENTS I Spent 30 Days Tracking the Re-Read Tax in My Claude Code Bill. 73% Was Redundant. A field measurement of what AI coding agents actually spend tokens on — and a working framework for cutting it. By Aurelian Jibleanu · Founder, ArgosBrain · 9 min read A month ago I noticed something strange in my Claude Code bill. I was running long coding sessions on a 400-file Next.js project — real production work, multi-file refactors, the usual. The bills were creeping up faster than I expected. So I did what every engineer does when a number doesn't match their model: I added logging. What I found surprised me. 73% of the tokens I was paying for weren't reasoning. They weren't tool calls. They were the agent re-reading files it had already read minutes earlier.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at ArgosBrain.
Discussion
0 commentsMore from ArgosBrain
