158 M C K I N S EY Q UA RT E R LY next ten, as though you had a super-fast pair programmer sitting beside you. AI provides a meaningful productivity boost. Level 3: Automating entire steps in the work flow. A developer describes a new feature to the AI agent in plain English. The AI generates the first version of the code, the tests, and the documentation automatically. The productivity boost is very substantial. - Level 4: Delivering entire applications. A small team guides a coordinated system of AI agents that can deliver an entire application end to end—from design to code to testing to integra tion—raising only the decisions that truly require human judgment. The result is 20 times lever age: a few practitioners delivering what once required a large department. - - Most companies are at Level 2 of this progression. Level 3 is increasingly being adopted as large lan guage models (LLMs) have evolved from simple inline completion tools to autonomously executing long-running multifile refactoring and moderniza tion tasks. Level 4 is largely experimental as of the writing of this book, though with promising developments already emerging. - - BEST PRACTICES FOR ADOPTING AI IN SOFTWARE DEVELOPMENT McKinsey analyzed nearly 300 publicly traded com panies to understand how AI is reshaping software development. We found that a small group of top performers—roughly the top quintile—are achieving 16 to 30 percent improvements in productivity, time to market, and customer experience, along with 31 to 45 percent gains in software quality. - The key insight here is that simply giving devel opers AI tools does not meaningfully move the needle. The companies that unlock real value are those that rearchitect how they build software and deeply embed AI across the entire development life cycle—not just for coding. - They deploy multiple AI development use cases spanning ideation, requirements, design, coding, testing, deployment, and operations, enabling con tinuous acceleration and compounding benefits. - These organizations also make their develop ment model AI-native, evolving roles, practices, and workflows so that humans act as orchestrators of AI agents. Developers shift from writing every line of code to supervising generation, validating architec ture, and managing quality; product managers and designers take on more system-level thinking and integration of AI into features and experiences. It’s a fundamental change in how teams work. - - Behind these shifts are three critical enablers that determine success. Top performers: 1. Invest in serious upskilling, using hands-on workshops, real sprint simulations, and coaching rather than passive training. 2. Institutionalize tracking outcomes— release frequency, defect rates, customer experience— not just simple adoption metrics. 3. Reinforce change through aligned incentives and performance management. In fact, about 80 percent of top performers link gen AI goals to the evaluations of product managers and developers. These enablers create accountability, accelerate learning, and help teams internalize new ways of working. Without them, organizations fall back into old habits, and AI’s potential dissipates. The implications are clear: AI can transform software development, but only for companies willing to rethink their operating model. Those that redesign workflows, roles, and governance around AI have the chance to create a true per formance advantage. AI can transform software development for com - A FACTORY OF AI AGENTS: HOW DOES THAT WORK? AI agents make it possible to run software devel opment like a two-shift digital factory. Humans take the day shift, setting direction and enforcing - p