The Present and Future of React, and AI: A React Summit US 2025 Report

React Summit US 2025 was held as a hybrid online/offline event at the Liberty Science Center from November 17 to 21, 2025, drawing about 800 in-person attendees and over 10,000 online participants. The event was held in conjunction with JSNation US, and featured Summit/Base Camp tracks, a Q&A zone, and Discussion Rooms.

The most strongly highlighted trend was practical AI engineering in the frontend space. Addy Osmani distinguished between simple vibe coding and agentic engineering, noting that while AI excels at generating individual components, it exhibits a complexity cliff where performance drops sharply when it comes to integrating with existing codebases, managing complex state, or handling changes to business logic. He emphasized that AI is skilled at logical implementation but lacks taste and global context, so the engineer's role in planning first (spec first, plan first) and reviewing outputs becomes more important.

In the panel discussion, the launch of the React Foundation and changes in governance were key topics. Decision-making, previously centered around Meta, is expected to become more transparent, with better support for core contributors and lower barriers to entry for open source. React Compiler was interpreted not merely as automatic memoization but as a change to the rendering model itself, and there was discussion that in the AI era, framework authors need to clearly provide instructions and evals.

As tools become more powerful, the importance of fundamentals also grows. There is no single correct answer among SPA, SSR, and streaming, and the panel raised the point that documentation guiding which architecture to choose in real production settings needs to be strengthened further. Ultimately, the discussion converged on the conclusion that the ability to filter out AI hallucinations and manage complexity depends on engineers' design skills and debugging capabilities.

Ken Wheeler redefined vibe coding as an extension of research- and intuition-based development that has existed since the past. However, he stressed that domain knowledge and the responsibility of verifying generated code remain essential, and that determinism in outcomes must be secured through tools such as function calling. Following this, Alexander Lichter introduced Rust-based unified tooling, explaining that Rolldown and Oxc reduce dev/prod mismatches and decision fatigue, with Oxc in particular improving DX through linting speeds 50 to 100 times faster than ESLint.

In addition, Amy Dutton pushed back against the misconception that RSC sacrifices DX to gain UX, framing it instead as a return to web standards and a reasonable evolution. Chaitanya Deorukhkar presented a workflow using MCP to connect Figma with custom design systems, enabling AI to understand design context and convert it into production code. Overall, the core message of the conference was that while tools are becoming more magical, engineers need to understand deeper principles, and that teams who use AI cleverly are seeing leaps in productivity.