People Don’t Agree On What AI Can Do, But They Don’t Even Use The Same Product

There's a significant disconnect in perceptions of AI's capabilities because people are evaluating different products. While many consumers interact with AI through tools like ChatGPT, primarily for tasks like information lookup and writing—accounting for over 70% of its usage—a separate world of powerful, agentic coding tools is rapidly evolving. These tools, such as Anthropic’s Claude Code, GitHub Copilot, and Anysphere, are dramatically increasing developer productivity and driving substantial investment, with companies like Anthropic raising billions at high valuations. The rapid improvement in coding AI is due to the verifiable nature of its results—code either works or it doesn’t—allowing for effective reinforcement learning. This divergence is impacting the stock market, with software companies losing market capitalization as investors adjust to the potential for reduced engineering headcount. The key takeaway is that AI is not a single category; it’s evolving into a consumer utility focused on reach and an enterprise agent focused on verifiable output, with the latter already demonstrating clear leadership and creating a widening moat for early players. Future opportunities lie in developing domain-specific coding agents in areas like legal tech and biotech.

(Source：Forbes)