What problem does BiModal Design solve?

A large share of AI agents — HTTP retrievers and many LLM-powered browsers — fetch only the initial HTML response and never execute JavaScript. Client-rendered single-page applications are invisible to them: the initial HTTP response contains only an empty container. BiModal Design ensures the content agents need is present in the initial payload, then layers progressive enhancement on top for human users and for higher-capability agents that do run JS.

How is BiModal Design different from web accessibility (WCAG)?

WCAG ensures human users with disabilities can use a site; BiModal Design extends that foundation to ensure non-human agents (LLM-based assistants, crawlers, tool-use agents, protocol-native systems) can also access and act on the content. Most BiModal Design work also improves WCAG compliance, but the reverse is not always true — an SPA with great ARIA still fails BiModal Design FR-1.

Do I need to rewrite my site to use BiModal Design?

Not necessarily. If you are already using SSR or SSG (Next.js, Nuxt, Astro, SvelteKit, Remix, or traditional server-rendered apps), the foundational requirement is met and the work is incremental: add semantic landmarks, JSON-LD, ARIA, and optional agent protocol endpoints. If you are running a pure client-rendered SPA, you will need to introduce server rendering for the initial payload before the other layers can help.

Who created BiModal Design?

BiModal Design was created by Joel Goldfoot. Joel designed the framework, wrote the documentation, and authors every example on this site. The framework is released under the Apache 2.0 license.

ByJoel GoldfootLast reviewed May 30, 2026

Introduction

What is BiModal Design?

BiModal Design (formerly AgentUX) is a standards-based design framework for creating dual-mode interfaces that serve both human users and AI agents with equal effectiveness.

The framework addresses a gap most sites have today: a large share of AI agents — HTTP retrievers and many LLM-powered browsers — fetch only the initial HTML response and never execute JavaScript. This means beautifully designed, semantically perfect interfaces can be completely invisible to those agents if they rely on client-side rendering. Higher-capability agents (browser automation and vision agents) do execute JavaScript but still perform more reliably when the markup is stable and semantic.

The Critical Discovery

Through real-world implementation and research, we identified that conventional UX optimization often has a critical blind spot: it extensively covers WHAT to put in the DOM (semantic structure, ARIA roles, structured data) but never addresses HOW to ensure that DOM exists for agents.

⚠️ Key Insight

When an AI agent accesses a web page, it makes a simple HTTP GET request (like curl). The agent parses HTML WITHOUT running JavaScript. It sees only what's in that initial HTML.

The curl Test

Every BiModal interface must pass this foundational test:

$ curl -s https://yoursite.com | grep "main content"

✓ Should return: actual content
✗ Should NOT return: <div id="root"></div>

Try it on your site right now. Open your terminal and run this test on your production website. What do you see?

The BiModal Concept

The "BiModal" name reflects two fundamental modes of interaction:

Human Mode: Rich, interactive experiences with JavaScript enhancement, animations, and real-time updates
Agent Mode: Accessible, semantic content present in the initial HTTP payload that agents can parse and understand

The key insight: these aren't separate versions—they're the same interface, experienced through different interaction models.

What agent benchmarks show

Public agent benchmarks report a wide gap between human and agent success on the general web, and the gap narrows sharply as agent capability rises:

Agent capability level	Task-completion range on the general web
HTTP retrievers (L0-1)	~12-20%
Browser automation (L2)	~35-50%
Vision agents (L3)	~40-55%
Human success (reference)	~72-89%

Read this table as directional, not measured. These figures come from public agent benchmarks that evaluate agents on the current web — not from a controlled A/B test of BiModal Design against a conventional interface. BiModal Design maps the design patterns those benchmarks reward (semantic HTML, structured data, ARIA, agent protocols) to the capability levels above, but the framework itself has not been independently benchmarked head-to-head.

Sources: WebArena (Zhou et al., 2023, arXiv:2307.13854), VisualWebArena (Koh et al., 2024, arXiv:2401.13649), ST-WebAgentBench (Levy et al., IBM Research, 2024, arXiv:2410.06703).

Why It Matters Now

The web has fundamentally transformed from a human-only medium to a collaborative space where AI agents perform critical business functions:

Autonomous Web Agents: Navigate and interact with websites independently
Agentic Systems: Multi-agent workflows coordinating complex tasks
Web Automation Agents: Execute repetitive tasks like form completion
Conversational Interface Agents: Bridge natural language to web actions

📊 Market Evidence

Microsoft Build 2025 introduced the "agentic web" with the NLWeb protocol. 230,000+ organizations use platforms like Copilot Studio for agent automation. The agent-web interaction revolution isn't coming—it's here.