AI Coding Shifts Development's Bottleneck from People to Thinking

Anthropic's AI coding tool "Claude Code" is reportedly raising engineer productivity threefold in practice. In response to this change, Anthropic instructed its growth team to increase hiring of product managers—professionals who determine product direction and priorities. With code-writing speed no longer a bottleneck, the shortage of talent to decide "what to build" has become apparent.

To understand this change, it helps to look back at how engineers' work methods have evolved over the past few years. Before 2022, the standard solution when developers hit a roadblock was to search for answers on the Q&A site "Stack Overflow." However, since ChatGPT's launch in November 2022, new posts on Stack Overflow have declined by approximately 77%. This indicates that engineers have begun choosing generative AI as their problem-solving tool.

Subsequently, the way AI tools are used has evolved in stages. In the initial stage, engineers repeatedly asked AI questions in their browsers and manually pasted the answers into their code editors. In the next stage, tools like Cursor and Claude Code embedded AI directly into editors and enabled direct access to repositories (the storage for entire codebases). In place of "Bash," the command-line tool that veteran engineers long called "the most continuously used tool," as of 2026, many developers' first action when starting new work is to launch the "claude" command.

Furthermore, from 2025 to 2026, as AI's ability to handle context (context window) expanded, a "specification-driven" development style became widespread—one where work that previously would have required multiple design documents and weeks of sprints (development cycles) can now be completed in a short time from just one specification sheet. The Amazon code editor "Kiro" development team reported using this method to shorten feature development time from two weeks to two days. Additionally, an AWS engineering team completed a system redesign originally planned for 30 people over 18 months in just 76 days with 6 people.

In April 2026, Anthropic released "Claude Code Routines." This is an AI agent feature that autonomously operates on a set schedule, further automating development work. This feature allows AI to continuously handle certain repetitive tasks without people constantly issuing instructions.

What these changes demonstrate is that the "rate-limiting step" (the point that determines overall speed) in software development has fundamentally shifted. In the past, code-writing speed determined development speed, but now whether one can clearly define "what to build and how" determines the pace and quality of the entire project. It can be said that engineers are increasingly required to have not just technical skills but also the ability to think about product direction.

The fact that Anthropic increased product manager hiring in its recruitment policy is also instructive for the entire industry. The more AI automates and accelerates code-writing tasks, the more human roles become concentrated in "giving instructions" and "determining direction." This structural change is positioned to continue affecting engineer career development and how companies organize their development teams going forward.