Agentable
Make software legible to people and AI. Turns websites and apps into a machine-readable UI map, guides humans in-flow, and lets AI agents complete the same tasks.
SectorAISecurity
A confusing website or app is really an undocumented operating system. Agentable makes that operating system readable in real time. A person gets immediate help inside the flow. An AI assistant gets explicit tools instead of brittle guesses. Product teams get a live feed of where the experience is actually breaking.
Average support ticket cost is about $15.56. Deflecting 10% of 500,000 tickets saves about $0.78 million per year.
Meanwhile, 18% of U.S. online shoppers abandon because checkout is too long or complicated. Clarity converts. Confusion bleeds.
The Problem
Interfaces change faster than documentation.
Humans get stuck and bounce. AI assistants guess and fail. Both miss the long tail of edge cases, mislabeled buttons, split tests, and locale quirks.
Product teams learn about issues late, via vague tickets. We need a simple, standard way for websites to publish current “how to use me” instructions that serve both people and agents, plus a feedback loop that converts real attempts into fixes.
BottleneckTrustCoordination
Solution Hypothesis
Publish a live, machine-readable task map grounded in the actual interface.
Continuously validate it against what the site really renders. Then use the same spec to power human guidance and agent automation.
One map, two runtimes, continuous truth.
Enabling TechnologyLarge Language ModelsAutonomous AgentsVision AIThe Agentable Loop
One map, two runtimes, continuous truth.
Live App UI
Ground Truth
Agent Sitemap
Machine-Readable
Human Guidance
In-flow visual overlay
Agent Automation
MCP Tool execution
QA / Validation
Release confidence
Dev-published updates, preferred path
- Agent Sitemap: smaller JSON file listing core tasks. Semantic locators prefer ARIA roles over brittle CSS. Instructions stay stable.
- UI Changefeed: signed JSON "what changed" feed pushing updated labels immediately.
- MCP Export: exposes tasks as MCP tools automatically. Any assistant, same map.
Continuous validation for everyone else
Risk-weighted crawls open key flows, read the Accessibility Tree, verify steps. Critical flows validate more often. Catches silent breakage on sites without published updates.
Two runtimes, one spec
- Human Guidance Overlay: native overlay highlights right controls, offers fallbacks on demand.
- Agent Tools: same taskscallable by assistants like Gemini Computer Use. Your site is agent-friendly by design.
Secure agentic logins
1Password Secure Agentic Autofill: enterprise-safe credentials for hosted agent browsers.
Convert navigation into insight
Telemetry: logs quantitative and qualitative signals routing straight to Dev issue trackers.
Ideal Customer Profile
A specific example.
The Problem
A subscription software company maps three painful account tasks: Reset Password, Update Billing, and Change Plan.
The Resolution
A confused admin asks for help, the overlay points to the exact control and explains the next step, and the same underlying map exposes a tool that the company’s support copilot can call directly.
Support volume drops, retries fall, and the product team gets screenshots plus transcript snippets showing exactly where the flow broke.
Universal Accessibility
Making complex software accessible to everyone, from enterprise admins to elderly users.
Autonomous Navigation
Exposing structural architecture so autonomous agents can navigate deep software reliably at scale.
Neglectedness
Market
This starts as a support and conversion product. It does not end there.
Every high-volume website pays a hidden tax for interface confusion. In an agentic web, that same tax expands. Now every important site also needs to be legible to software actors that browse, click, fill forms, and complete workflows on behalf of users.
Layer 1: Support & Conversion
Enterprise teams buy because confusion is already expensive.
Layer 2: QA & Confidence
The live UI map becomes an always-on validation layer for critical flows.
Layer 3: Agent Routing
Reliably machine-readable sites get preferred treatment from assistants and marketplaces.
Layer 4: Web Infrastructure
The winning company becomes the neutral trust layer between websites, humans, and AI agents.
Why Now
A unique confluence of technological shifts makes this achievable today.
Models can now operate browsers. Computer-use systems have made live UI interaction economically real.
A neutral standard now exists. MCP makes cross-model tooling portable enough for one map to serve many assistants.
The budget owner already exists. Support and self-serve teams already measure the cost of confusion and can justify spend fast.
"This is the moment UI metadata flips from nice-to-have documentation into revenue, support, and automation infrastructure."
ReadinessBuild Now
Business Model
B2C For People
- Pro: paid access to the Human Guidance Overlay on any site.
- Free: no-cost plan sharing anonymous transcripts to improve maps.
- BYO Assistant: UI maps plug into user's AI of choice. Native overlay answers.
B2B For Companies
- White-label overlay: drop-in navigation support for every visitor.
- Verified-Current badge: pay to be continuously validated and marked current. Priority routing in AI.
- Developer workflow: CI checks for Agent Sitemap, MCP generation, issue tracker integrations. Pricing aligns with deflection.
Moat & Defensibility
The moat is not the schema alone. The moat is the schema plus validation plus telemetry plus workflow embed.
Difficulty to Market
This is buildable now, but hard in the ways that matter. The product has to be useful across messy real-world interfaces before the market fully believes the category exists.
Go-to-Market
Launch the Agentability Index
A public leaderboard ranking major websites on task clarity, validation freshness, and agent-readiness. This gives the company a built-in media engine. The best companies buy to improve their ranking and earn the Verified-Current badge.
Founder FitTechnical FounderVenture-ScaleAGI Future Edge
Humans and agents learn from the same map.
Human traces harden instructions. Agents stress-test edge cases at scale. The combined telemetry discovers the long tail of bugs and odd behaviors that QA misses.
Reliability compounds. As computer-use models improve, more guided steps promote to automation without changing the spec.
- De facto standard: USB-C for website intent.
- Aligned monetization: Verified quality over volume.
- Compounding advantage: Better models make the validator stronger.
Civilizational Impact
If AI agents become a major way humans interact with digital systems, then legibility becomes a safety primitive.
Agentable pushes the internet toward a future where software does less blind guessing and more explicit, inspectable action. A web that can explain itself to both humans and machines is easier to navigate, easier to audit, and easier to trust.
Overall Impact Score
"In the biggest version of the story, Agentable becomes the trust and clarity layer for the agentic web. That is a large company. More importantly, it is useful civilization-scale plumbing."
Open Source Priority
KPIs
Transferable Insight
The non-obvious lesson is that the same structured layer that makes software usable by AI often makes it better for humans too. Build machine legibility at the point of action, and you usually get clearer UX, better analytics, and lower support burden as a side effect.
First Experiment
Quick falsifiable hypothesis: for one high-friction account flow, a live UI map plus guidance overlay will reduce failed completions by at least 20% versus control within two weeks.
Smallest real test: instrument one flow, recruit 200 sessions, turn on the overlay for half, and compare completion rate, retries, and support contacts. If the effect is weak, the problem is not yet painful enough or the map is not precise enough.
Validation Experiment
Run a 30-day pilot on three high-volume flows: Reset Password, Update Billing, Change Plan.
- Publish your Agent Sitemap and UI Changefeed.
- Enable the overlay for 20% of sessions.
- Expose MCP tools to one assistant integration.
Win criteria: at least 10% ticket deflection on those tasks, at least 25% fewer step retries, and a ranked list of long-tail UX issues with screenshots and optional transcript snippets. Then expand.
Acronyms & References
[1] HDI, "Metric of the Month: Desktop Support Cost per Ticket."www.thinkhdi.com/library/supportworld/2017
[2] Baymard Institute, "49 Cart Abandonment Rate Statistics."baymard.com/lists/cart-abandonment-rate
[3] Model Context Protocol, official specification.modelcontextprotocol.io
[4] Google, "Introducing the Gemini 2.5 Computer Use model."blog.google/technology/google-deepmind
[5] Business Wire, "1Password and Browserbase Partner..."businesswire.com
[6] IDC, "Worldwide Spending on Artificial Intelligence Forecast..."my.idc.com
[7] Grand View Research, "AI Agents Market Size..."grandviewresearch.com
[8] UC Today, "Gartner Predicts 40% of Enterprise Apps..."uctoday.com
Valuation Forecast
Probability that the category leader in this space reaches at least each valuation threshold.
AI Rationale
The AGI Futures forecaster model expects the commoditization of foundational models to rapidly shift value toward specialized context-aware orchestration. Agentable possesses a high probability of yielding a $1B+ category leader quickly (by 2030), given the massive TAM of white-collar task automation and relatively low capital requirements for software deployment.
Implied Valuation Distribution (2030)
While the chart below displays cumulative probability, these boxes break down the exact probability of landing specifically within each valuation band.
Builder Proof-of-Work
Community submitted artifacts, notes, and implementations for this idea.