Why Managers Outperform Programmers Using Claude Code: Surprising Success Rates

The article argues that future essential skills lie in management and economics, treating AI agents like Claude Code as extensions of agent management rather than pure engineering tools.

Anthropic recently released a report displaying profession‑wise success rates for tasks completed with Claude Code; managers top the list at 95%, slightly ahead of software and mathematics professionals at 94%, with law, finance, and medicine close behind.

Wharton professor Ethan Mollick shared the chart, calling it early evidence that management may become an AI super‑power, while cautioning that unseen factors could affect relative success rates.

Mollick recounts a university experiment where MBA executives—most without coding experience—used Claude Code, Google Antigravity, ChatGPT, and Gemini to prototype a startup in four days, producing results far beyond previous cohorts that spent an entire semester on similar projects.

He explains that these participants leveraged soft skills such as problem framing, deliverable definition, and rapid error detection, which are more critical in AI collaboration than technical expertise; the bottleneck is articulating intent clearly to the model.

The piece notes that for managers, writing code is rarely the difficulty; the real skill is precisely describing desired outcomes, turning development into a management activity.

Critics warn that if managing AI agents becomes the most valuable skill, traditional apprenticeship periods may disappear, leading to a generation adept at directing agents but lacking foundational intuition, and that detailed specifications cannot compensate for poor aesthetic judgment.

Agent engineer "zakk" proposes a practical "coloring book" framework: first outline the task (the sketch), then let the model fill in details, and finally verify and refine the output. This three‑step loop—outline, fill, polish—remains constant even as models improve.

In summary, the model supplies the crayons, but humans decide what to draw.