First, you no longer need a dedicated specialist for every technology in your stack. AI can now handle the "what does this obscure library do" questions faster than most humans care to Google. You just need a few core tech specialists to ensure architectural coherence.
Second, communication is still a very human bottleneck. AI can't attend your standups, and every minute in a meeting is a minute not building something useful. Large teams multiply this overhead quickly.
Third, merge conflicts: with AI-boosted velocity, a single developer can produce larger commits more quickly. Fewer developers means less overlapping work, and fewer opportunities for Git to ruin your day.
If you're working on an established product with a larger codebase, 2 to 5 person teams make sense. Keep responsibilities cleanly separated to minimize communication needs and conflicts. Maybe even split a larger team into smaller, non-overlapping sub-teams.
If you're starting fresh, go with two developers per team. This gives you some redundancy without turning task coordination into a full-time job. It's enough to ride the early AI productivity wave without drowning in overhead. As complexity grows, you can add more people gradually, just not so many that Slack becomes your main dev tool.
