Which speeds up development more: AI Coding Agents or Pair Programming?

AI tools are touted to increase coding speed, but their true impact on overall development efficiency is questionable. A simulation was created to compare different development workflows, including traditional pull requests, AI-enhanced pull requests, pair programming with trunk-based development, and AI-enhanced pair programming. The simulation assumed an infinite backlog of tickets, each taking about a day to complete. AI assistance was modeled with a 30% coding speedup. Results showed that AI alone didn't significantly reduce lead times due to bottlenecks in the pull request review process. Pair programming with trunk-based development drastically reduced lead times and rework, while AI amplified these benefits further. Defect rates matter, but workflow optimizations like removing queues offer the biggest gains. Optimize processes before heavily investing in new tools like AI. Combining AI with pair programming appears to be the most effective approach. Data-driven decisions, observability, and good technical practices are crucial when adopting AI.