What Claude Code chooses (amplifying.ai)

Article summary

A study analyzed the tool choices of Claude Code, a coding agent, by pointing it at real repositories and observing its decisions. The results showed that Claude Code tends to build custom solutions rather than recommending tools, and when it does pick a tool, it often chooses established ones. The study also found that newer models tend to pick newer tools, and that deployment is fully stack-determined, with Vercel being the top choice for JavaScript and Railway for Python. The study's findings have implications for devtool companies and the future of coding.

Main themes

• Coding agents
• Tool choices
• Custom solutions
• Devtool companies
• AI-led development
• Monetization strategies

What commenters say

• The study's methodology of not naming tools in prompts is questioned, with some arguing that it is unrealistic and others seeing it as a way to test the agent's true preferences.
• Some commenters are concerned that coding agents like Claude Code may be influenced by companies paying for promotion, leading to biased tool recommendations.
• Others argue that the rise of AI-led development will lead to a new form of SEO, where companies optimize their tools to be recommended by coding agents.
• There is disagreement over the value of software patents, with some seeing them as overly broad and others arguing that they are necessary for innovation.
• Some commenters believe that the future of coding will involve more custom solutions and less reliance on established tools, while others think that established tools will continue to dominate.
• The potential for invisible advertising and conflict of interest in AI-led development is seen as a concern by some, while others think that it is an inevitable consequence of the technology.