AI coding agents will split software stocks into winners and losers

The release of Claude Opus 4.6 with agent teams, alongside GitHub Copilot and competing agentic coding tools, has triggered a structural repricing of the entire software sector — not just in what gets built, but in how companies are organized, how they sell, and who survives. AI-generated code now accounts for 41% of all code written, GitHub sees 230 new repositories per minute (+25% YoY), and 84% of developers use or plan to use AI tools in 2026. The result is a three-front disruption: (1) Engineering teams are being flattened from pyramids into 'centaur pods' of 3-5 humans orchestrating AI agent fleets — Microsoft has cut 15,000 roles, Amazon 14,000, and Indeed reports the top roles eliminated are software engineers, QA, PMs, and project managers. (2) Go-to-market is being rebuilt from the ground up as per-seat SaaS pricing collapses — when one person with AI agents does the work of five seats, customers cancel four licenses, forcing a painful transition to outcome-based pricing that EY estimates will take 2-3 years. (3) AI-native startups with 15-20 person teams are capturing 63% of AI application-layer revenue, operating 34% leaner than traditional startups at the same stage, and generating 300% more revenue per employee. Lovable hit $17M ARR in 3 months with ~15 people. BuiltWith does $14M/year with one employee. Meanwhile, large incumbents face an organizational paradox: 75% of enterprises using AI tools see no measurable team-wide performance gains because individual velocity does not translate through layers of security reviews, compliance processes, and coordination overhead. Series A/B startups without that bureaucracy can rebuild their entire product overnight. The winners will be companies that either enable the revolution (security, observability, infrastructure) or ruthlessly restructure around it. The losers are mid-market SaaS companies whose products, teams, and pricing models were built for a world that no longer exists.