AI coding hype overblown, Bain shrugs (www.theregister.com)

Bain & Company’s Technology Report 2025 finds that generative AI in software development has largely under-delivered versus the hype: roughly two‑thirds of firms have deployed GenAI tools but developer adoption inside those firms is low and measured productivity gains are modest (around 10–15%). Independent studies deepen the cautionary tale — METR found AI coding tools can slow developers because they must spend time checking and correcting AI outputs, and benchmarking work (e.g., Cognition’s Devin, Gartner, Carnegie Mellon) shows agentic, multi‑step AI systems often fail or get stuck, with high cancellation or failure rates. The report’s key technical takeaway is that accelerating code authoring alone won’t move the needle much—writing and testing make up only about 25–35% of the development lifecycle—so real payoff requires reengineering the entire lifecycle (discovery, requirements, design, CI/CD, testing, deployment, maintenance) around AI. Bain warns companies need executive direction, change management, training in prompt engineering and review, and clear KPIs to prove value; firms that combine end‑to‑end process transformation with GenAI sometimes report 25–30% gains, but widespread positive ROI will demand bold, measurable, AI‑native workflow redesign rather than piecemeal pilots.