AI Coding Agents Push Troubleshooting Costs to $400 Billion

AI-powered coding agents have dramatically increased software development speed, but they've also created new problems in production. According to Eric Schwartz, product manager at AI site reliability startup Traversal, this has pushed the crisis deeper into the stack.

Enterprises now spend upwards of $400 billion a year on troubleshooting, with 40% of executives calling it a team-level issue. Engineers lose seven-plus hours a week to on-call troubleshooting. Schwartz argues that root cause analysis is not an observability problem but a causal one, and current tools are watching the wrong thing.

At American Express, Traversal now acts as the first responder to every incident, posting a root cause analysis to Slack within three minutes. In one case, 50-plus engineers were spared being paged. At Pepsi, engineers previously faced backlogs of 700 alerts each; Traversal now pre-investigates and prioritizes them.

Schwartz outlines a five-level autonomy ladder toward closed-loop 'self-driving production,' where AI finds issues, diagnoses causes, applies fixes, and verifies them without paging anyone. He claims 80%-plus root cause accuracy across trillions of logs.