As AI agents become capable of operating software on behalf of users, a fundamental question emerges: how well does existing SaaS infrastructure support autonomous interaction? The Agentability Project's audit of 100 leading SaaS products provides the first systematic answer, and the results suggest most products—including established platforms like PagerDuty—face significant architectural challenges in the agent era.
The research introduces a measurement framework called agentability: a 0-100 score assessing how effectively AI agents can operate software products. The score aggregates performance across eight Agent Factors Engineering (AFE) principles, each addressing a specific dimension of machine operability.
Across 100 audited products, the average agentability score stands at just 38.3 out of 100, indicating that most SaaS products remain optimized exclusively for human operation.
The Agentability Distribution
The audit evaluated homepage implementations—the first touchpoint for both human users and autonomous agents—using a standardized rubric. Products fell into four distinct tiers based on their scores:
- 22 products achieved Agent-Ready status (45+ points)
- 54 products landed in the Developing tier (35-44 points)
- 17 products fell into the Lagging category (20-34 points)
- 7 products scored as Agent-Blind (under 20 points)
This distribution reveals that while roughly one-fifth of products have begun building agent-compatible interfaces, the majority remain in early stages of adaptation. No product achieved a score above 70, suggesting even the most advanced implementations have substantial room for improvement.
Where Products Struggle Most
Breaking down performance by individual AFE principles exposes systematic weak points across the SaaS ecosystem. Machine readability scored highest at 71 out of 100, reflecting decades of investment in semantic HTML and accessibility standards that incidentally benefit machine parsing.
The remaining principles tell a different story:
- Transparency: 5/100 average
- Shadow-UI avoidance: 17/100 average
- Defaults: 35/100 average
- Clean handoffs: 40/100 average
- Chunking: 45/100 average
- Status: 47/100 average
- Control: 48/100 average
Perhaps most striking: 83 of 100 products scored zero on transparency, indicating almost universal absence of machine-readable capability declarations.
Understanding the Weak Principles
Transparency in the AFE framework refers to a product's ability to programmatically declare what actions an agent can perform, what permissions are required, and what constraints apply. A transparency score of 5 means products typically provide no structured metadata about their capabilities—agents must infer functionality through trial and error or rely on external documentation written for humans.
Shadow-UI avoidance addresses whether products implement invisible or dynamic interface elements that appear only in response to user behavior. These patterns—hover menus, right-click contexts, gesture-triggered panels—work for humans but create blind spots for agents operating through accessibility APIs or DOM inspection.
With 80 of 100 products scoring under 20 on shadow-UI avoidance, these interaction patterns remain deeply embedded in modern web design.
Defaults, scoring 35 on average, measure whether products ship with sensible starting values that allow agents to accomplish tasks without exhaustive configuration. Poor default design forces agents into complex negotiation dialogs before simple operations can proceed.
PagerDuty in Context
PagerDuty's position within this landscape represents a microcosm of the broader challenge facing incident management and on-call platforms. Like many SaaS products built during the mobile-first era, PagerDuty optimized for rapid human response to time-sensitive events—a design priority that emphasized notification systems, mobile interfaces, and real-time dashboards.
The shift toward agent operation introduces different requirements. An AI agent managing incident response needs programmatic access to escalation policies, structured status information about service dependencies, and clear signaling about what actions require human approval versus autonomous execution. These capabilities demand explicit design work beyond traditional API development.
The full agentability assessment for PagerDuty is available at https://agentability.io/index/pagerduty.html, providing principle-by-principle scoring and specific implementation observations. Product teams can review detailed breakdowns for all 100 audited products in the Agentability Index.
What Product Teams Should Do
The systematic nature of these gaps suggests opportunities for structured improvement. Product teams don't need to rebuild their applications—they need to augment existing interfaces with agent-compatible affordances.
Start with transparency
Implement machine-readable capability declarations. A simple JSON-LD schema describing available actions, required scopes, and rate limits gives agents the contextual information they currently lack. This requires documentation work more than engineering effort.
Audit for shadow-UI patterns
Inventory interface elements that require hover, right-click, or gesture interaction. For each, ensure a visible, permanently accessible alternative exists. This often means adding toolbar buttons or menu items that duplicate hidden functionality.
Improve defaults intelligently
Review configuration-heavy workflows and ask: what percentage of users choose non-default options? For settings where 80%+ of users make the same choice, that choice should be the default. Agents benefit from the same sensible starting points that reduce cognitive load for humans.
Test with accessibility tools
Run your product through screen readers and keyboard-only navigation. These tools expose many of the same barriers agents encounter. If a screen reader can't discover a feature, an agent probably can't either.
Measure your baseline
Product teams can run a free audit to establish their current agentability score and identify highest-impact improvement areas. The rubric provides specific, actionable criteria rather than subjective assessments.
The Path Forward
An average score of 38.3 across 100 leading SaaS products indicates the industry stands at the beginning of a significant infrastructure transition. Products like PagerDuty that enhance their agentability early will be better positioned as enterprises deploy AI agents for operational workflows.
The good news: most improvements require additions rather than replacements. Products can layer agent-compatible interfaces onto existing human-optimized designs. The challenge lies in recognizing that optimization for autonomous operation requires explicit design attention—it rarely emerges as an accidental byproduct of building for humans.
As the data shows, nearly every product has room for improvement. The question isn't whether agentability matters, but how quickly product teams will prioritize it relative to competing demands. For platforms managing critical infrastructure and incident response, the answer may determine which tools agents—and by extension, their human principals—choose to rely on.
How agent-ready is your product?
Run a free agentability audit and get a scored, prioritized fix list in minutes.
Run a free audit