From Web Apps to Agent-First UI Framework Development
The AI agent era describes a shift from user-driven applications toward systems where autonomous or semi-autonomous agents coordinate tasks, tools, and interfaces on behalf of people, which is now reshaping expectations for UI framework development and the surrounding infrastructure that powers these intelligent workflows. ArrowJS and Vercel’s Eve both position themselves as AI agent framework options, but they approach the problem from opposite directions. ArrowJS starts with the browser and keeps everything as close to the DOM as possible, while Eve starts with directories, workflows, and observability for production agents. Together they show what agentic era tools look like when UI is no longer only for humans, but also for large language models that need to read, generate, and safely execute interface logic. For developers, the question becomes which assumptions they want baked into their agent stack from day one.
ArrowJS: Minimalist Reactive UI for Agents and Humans
ArrowJS began as a tiny reactive UI library and has now been recast as a framework for the agentic era, reaching a stable 1.0 after about three years of experimentation. It is built on core web primitives—JavaScript modules, tagged template literals, and the DOM—with no JSX, compiler, or required build step, and its runtime ships at under 5kb. The public API centers on three functions—reactive, html, and component—so the entire documentation fits in a small fraction of a large language model context window, which is key to its claim of being agent-friendly. According to InfoQ, ArrowJS benchmarks on par with Vue 3 while staying dependency-free and close to vanilla JavaScript. The 1.0 release adds a QuickJS-based WebAssembly sandbox that can run untrusted agent-generated UI code against the real DOM, plus optional layers for async components, server-side rendering, and hydration.
Eve: An Open Source Agent Framework Where Directories Define Behavior
Vercel’s Eve takes a different path by defining an AI agent as a directory and then compiling that directory into a running agent. Each folder describes the agent’s behavior: one file selects the model, another holds the system prompt in Markdown, and each tool is a TypeScript file whose filename becomes the tool’s name. Skill.md files and MCP servers plug agents into external tools, while the Workflow SDK turns every conversation into a durable workflow that can checkpoint, pause, and resume after crashes. Vercel compares Eve to “Next.js for agents,” and it currently runs in public preview as an open source agent framework on GitHub under Apache 2.0. Agents can call subagents, connect through OpenAPI documents, and reach users on Slack, Discord, Microsoft Teams, Telegram, Twilio, GitHub, and Linear through built-in channels, making Eve feel closer to an application runtime than a UI toolkit.
Developer Experience: Minimalism vs. Directory-Driven Agent Composition
Viewed side by side, ArrowJS and Eve show two extremes in AI agent framework design. ArrowJS gives developers three core functions and expects them to assemble patterns on top, trading rich conventions for minimalism and the ability for both humans and agents to reason directly about DOM code. The optional WASM sandbox tightens the loop between agent-generated UI and safe execution. Eve, in contrast, bakes in directory-based composition, durable workflows, observability via OpenTelemetry traces, and deployment through the familiar vercel deploy command. Vercel reports that agents already trigger around 29 percent of deployments on its platform, up from less than 3 percent a year earlier, which explains why it leans so hard into production features. The trade-off is that Eve assumes Vercel’s stack and more structure, while ArrowJS stays platform-agnostic and closer to the browser. Teams must choose between light abstractions and a heavier, batteries-included agent stack.
What Agentic Era Tools Mean for Future UI Frameworks
These two projects hint at how future UI framework development will adapt to agent-first work. ArrowJS focuses on making UI code compact, explicit, and discoverable so coding agents can read, modify, and generate components without facing proprietary syntax or large configuration surfaces. Its npx-installable skill further positions it as a small, composable layer in broader agent systems. Eve, on the other hand, treats UI as one piece of a wider runtime that must handle workflows, sandboxes, and production observability. In practice, teams might pair a minimalist UI library like ArrowJS with a separate orchestration layer, or adopt an all-in-one open source agent framework such as Eve where directories and workflows define agents from the start. Either way, the agentic era is pushing frameworks to think of UIs not only as screens for users, but as interfaces that agents themselves can manage and evolve.
