Can AI agents actually
use your software?

Most SaaS products weren't designed for agents. The agent opens your app, can't find the button, can't parse your pricing, doesn't know if its last action worked. We built the framework — and the evidence — to measure this.

39.7
Average score
across 100 products
0
Products scoring
above 60/100
8
Principles in the
AFE framework

New terminology for the agent era — definitions intended to become standard

agentability
/ˌeɪdʒəntəˈbɪlɪti/
noun

The degree to which a software system can be successfully operated, navigated, and completed by an AI agent without human intervention. A property of the software, not of the agent — agentability describes what the product affords, not what the agent can do.

Usage: "The onboarding flow has low agentability — the CTA is inside an SVG with no accessible label." · Analogous to usability (for humans) and accessibility (for assistive technology), agentability extends the design quality space into a third dimension.

Agent Factors Engineering
/ˈeɪdʒənt ˈfæktərz ˌɛndʒɪˈnɪərɪŋ/ · AFE
noun phrase, discipline

The discipline of designing, evaluating, and optimising software systems to be operable by AI agents. Agent Factors Engineering is the agent-era successor to Human Factors Engineering (HFE) — where HFE asks "can a human complete this task?", AFE asks "can an AI agent complete this task reliably, without ambiguity, without requiring human intervention?"

Context: HFE emerged in the 1940s as aviation and manufacturing demanded systematic design for human operators. As AI agents begin acting on behalf of humans across software interfaces, AFE addresses the same need at the agent layer. The two disciplines are complementary — a high AFE score does not substitute for good UX.

Agentability Score
0 – 100 composite
noun phrase, metric

A composite metric (0–100) measuring a software product's agentability across 8 principles of the AFE framework. Computed as the mean of 8 equally-weighted principle scores. Scores above 45 indicate the product can support basic agent workflows without intervention. Scores below 20 indicate the product is effectively opaque to agents.

Baseline (2026): 100 top SaaS products audited. Mean score: 39.7. Highest: 51.8. No product exceeded 55. The agent-readable web does not yet exist.

The AFE Framework

8 Principles of Agentability

Each principle reflects a question an AI agent asks when operating your software. A product that answers all eight clearly scores well. Most products fail at least four.

Machine Readability

Can an agent parse your content reliably? Semantic HTML, JSON-LD structured data, stable element IDs, and text that doesn't live inside SVGs or canvas elements.

machine_readability

Content Chunking

Is information broken into navigable, reference-able units? Agents can't scroll and skim — they need clear headings, sections, and scoped content blocks.

chunking

Agent Control

Can an agent operate your UI programmatically? Labeled buttons, keyboard-accessible controls, and form elements with predictable, stable attributes.

control

Status Communication

Does the product clearly tell agents what happened? Loading states, success/error feedback, and progress indicators agents can read without visual inference.

status

Smart Defaults

Does the product require minimal configuration to reach a working state? Agents struggle with open-ended choice — sensible defaults reduce decision burden.

defaults

Clean Handoffs

Are next actions, links, and escalation paths unambiguous? Agents need clear signals about where a workflow ends, where it continues, and how to hand off to humans.

handoffs

No Shadow UI

Is the interface free of hidden patterns agents can't see? Cookie walls, interstitials, invisible overlays, and dynamically injected elements that only appear on interaction.

shadow_ui

Transparency

Is critical information — pricing, rate limits, capabilities, constraints — explicitly published and machine-readable? Does the product offer agent-specific documentation?

transparency
Classification

Agentability Tiers

Score ranges correspond to practical agent capability. A product in the Lagging tier can be partially automated but will require human intervention at key decision points.

45 – 100

Agent-Ready

Core workflows can be completed by an agent without intervention. Basic tasks succeed reliably. Minor friction remains.

35 – 44

Developing

Agents can complete some tasks but will hit friction points. Partial automation is feasible with workarounds.

20 – 34

Lagging

Most agent workflows fail or require significant human intervention. The product was designed exclusively for human operators.

0 – 19

Agent-Blind

The product is effectively opaque to AI agents. No meaningful automation is possible without major architectural change.

The Agentability Index 2026

100 products.
One uncomfortable truth.

We audited 300 pages across the 100 most-used SaaS products in the world. The results show a category that has not begun to design for the agent era.

39.7
Average Agentability Score
51.8
Highest score (n8n)
0
Products scoring above 60
29
Checks per audit · 8 principles

"The agent-readable web does not yet exist. Not because it is hard to build — but because no one has defined what it means."

View the Full Index → Audit Your Product
Free Audit · 3 per email

Audit Your Product

Paste any public URL and get an Agentability Score in under 60 seconds. The audit checks all 8 principles and returns a principle-level breakdown.

Run an audit

Enter any publicly accessible URL — homepage, pricing page, or docs. Results are stored to track changes over time. Privacy policy.

Audit a URL — enter the page address and your email to receive your Agentability Score
Enter the full URL of any public page including https://. Example: https://yourcompany.com/pricing
Your email is used to enforce the limit of 3 free audits per person and is stored securely. See privacy policy for details.

Your Agentability Score will appear here.

3 free audits per email · Results stored in Supabase · By auditing you agree to our privacy policy
Transparency & Agent Documentation

Built to practice what it preaches

A site that scores products on transparency should itself be transparent. Here is everything an AI agent needs to know about this site.

Audit API

Endpoint
POST https://n8n.icuboid.in/webhook/public-audit
Request body
{"url": "string", "email": "string"}
Response
{"success": bool, "audit_id": "uuid", "score": number, "principle_scores": object, "report_url": "string"}
Rate limit
3 requests per email address (lifetime). Returns HTTP 429 when exceeded.

Pricing & Access

Cost
Free — no credit card, no account required
Limit
3 audits per email address, no expiry
Login required
No — email address only
Methodology version
AFE Rubric v0 — 8 principles, 29 checks, equal weights
Data stored
Email, audited URL, score — Supabase EU region
Cookies & tracking: None. No analytics, no tracking pixels, no cookies are set on this site. The only data stored is what you explicitly submit via the audit form. See privacy policy.