Rachel Lee Nabors (RL Nabors) is a web engineer, developer educator, and former cartoonist who led the creation of React.dev and now designs more expressive, efficient ways for artificial intelligence to interact with the web. Their work spans browser standards, developer experience, agent interfaces, and right-sized AI: choosing models and deployment environments that fit actual product requirements.
Nabors learned web development to publish their webcomics, translating a background in illustration and storytelling into work on Firefox DevTools, web standards at the W3C, and the Web Animations API. They subsequently worked on Microsoft Edge, authored Animation at Work, and joined Meta’s React team, where they contributed to React Native documentation and led development of React.dev.
The React.dev launch announcement, which Nabors coauthored, credits them with leading the project, illustrating its documentation, and writing much of its educational material. Its interactive examples, diagrams, and exercises introduced modern React through Hooks, making an intricate framework more approachable without centering outdated practices.
Nabors also held a principal technical program management role at Amazon Web Services, worked through their consultancy Dressed for Space, and built Mima, a client for multiple social networks. At AI Engineer Europe and World’s Fair in 2026, they were affiliated with Arize.
The browser as an infinite canvas. Nabors challenges the assumption that agents belong inside text-only chat windows. Their interactive agent-interface work turns their comic archive into a demonstration of hosted Model Context Protocol tools and embedded readers: agents can discover storylines, retrieve transcripts, and navigate comics through ordinary HTML, CSS, and JavaScript. Practical constraints—including sandboxed iframes, content security policies, cross-origin resources, and host-controlled navigation—shape the design.
WebMCP and structured browser actions. Nabors advocates exposing callable website actions directly to browser agents, reducing dependence on screenshot interpretation and sprawling page markup. They distinguish WebMCP from full Model Context Protocol compliance and identify inconsistent support for MCP resources as an obstacle to giving agents useful context efficiently.
Prototype big, deploy small. Their model-selection framework starts by proving a feature with a capable model, defining measurable success, and testing smaller alternatives until the smallest acceptable option emerges. Their right-sized AI guidance treats privacy, latency, connectivity, energy consumption, and inference costs as product-design requirements.
Mima and evaluation-driven local inference. For Mima’s conversation summaries, Nabors evaluated Qwen, Llama, and Gemma models using Arize Phoenix, comparing factual consistency, structured output, reference accuracy, and latency. Llama 3.2 performed best for that workload; few-shot prompting and straightforward post-processing improved its results. Examining individual evaluation traces also revealed that some apparent failures reflected an overly strict automated judge, underscoring why production systems need human judgment and ongoing regression evaluations.
A local model can replace a frontier call when the task is narrow, the acceptance criteria are explicit, and evaluation covers the whole application workflow.
▶ Watch ↗