Faros Finds AI Triples Software Incidents as 22,000 Developers Generate More Code
Updated
Updated · Financial Times · May 15
Faros Finds AI Triples Software Incidents as 22,000 Developers Generate More Code
2 articles · Updated · Financial Times · May 15
Telemetry from 22,000 developers on Faros found AI is producing more code and starting more projects, but the software reaching production is lower quality and incidents have tripled from the low-adoption baseline.
Review, testing and completion stages are all taking substantially longer, suggesting AI speeds the front end of coding while creating downstream bottlenecks that still require human judgment and capacity.
Faros said even high-performing engineering organizations are seeing the same deterioration, challenging earlier DORA research that suggested stronger teams were managing AI-related quality problems better.
The findings add to broader doubts about AI productivity claims: gains measured at the individual task level may not translate into higher organizational value once rework, defects and operational strain are counted.
If AI makes workers faster, why are companies slowing down and experiencing more failures?
Is your company's AI investment creating a hidden 'productivity tax' of rework and system errors?
The Acceleration Whiplash Paradox: How AI Tools Boost Output but Cause a 242.7% Increase in Software Incidents
Overview
AI-powered development is facing 'Acceleration Whiplash,' where the use of AI coding tools is speeding up code writing but causing a sharp decline in software quality and a dramatic rise in incidents. Developers are merging more pull requests without proper review, leading to a 242.7% increase in incidents per pull request. This rush for productivity sacrifices thoroughness and quality assurance, as rapid code generation is often not robust or reliable. With reduced oversight and low trust in AI-generated output, more errors are slipping into production, highlighting the urgent need for better review processes and responsible AI integration.