AI Costs Are Cheap Compared with the Cost of Poor Engineering Operations

Summary

Aviv Ben-Yosef argues that engineering organizations are focusing too narrowly on AI-agent, token, and compute costs while overlooking the much larger cost of inefficient operations. He estimates that startups with more than 30 engineers can waste over $1 million a year because of weak processes, and cites Atlassian and other surveys reporting that more than 15–20% of engineering time is lost to process and quality problems. The article recommends collecting evidence for two weeks before deciding what to optimize. Leaders should measure unplanned work caused by firefighting, trace quality issues to code, review, testing, or scaling problems, and stop treating every urgent request as a top priority. They should also reduce misalignment by giving teams better context and descoping low-impact work, while using smaller iterations to catch misunderstandings in product requirements. The author warns that prompt-generated specifications and large AI-assisted pull requests can remove useful review friction without removing the underlying need for verification. Other sources of waste include excessive meetings, communication overhead from poorly sized teams, and processes that have become outdated. In his view, AI is usually the cheapest budget line but the most visible invoice; improving how people work can reclaim far more time than modest token savings.