DocuSign and the Agentability Gap: Why Most SaaS Products Struggle With AI Agent Access

As AI agents move from prototype to production, a critical question emerges: can they actually operate the software tools that power modern business? DocuSign, the electronic signature platform used by over a million customers worldwide, offers a revealing case study in what researchers call agentability—the measurability of how well AI agents can interact with software products.

According to recent data from the Agentability Index, which evaluated the homepages of 100 leading SaaS products using Agent Factors Engineering (AFE) principles, the software industry faces a significant readiness gap. The average agentability score across all audited products stands at just 38.3 out of 100, with DocuSign's own performance detailed in its dedicated assessment.

Understanding Agentability

Agentability measures how effectively AI agents can discover, understand, and operate software interfaces. Unlike traditional API integration, agentability focuses on whether agents can interact with products the same way human users do—through web interfaces, documentation, and standard workflows.

The measurement framework relies on eight Agent Factors Engineering principles, each scored from 0 to 100:

• Machine readability: Whether interface elements are semantically structured for programmatic interpretation
• Transparency: How clearly products document their capabilities and requirements
• Shadow-UI avoidance: Minimizing hidden interface layers that agents cannot perceive
• Defaults: Whether sensible default values reduce decision complexity
• Control: Explicit mechanisms for confirming or reversing actions
• Chunking: Breaking complex workflows into discrete, manageable steps
• Status: Clear communication of system state and process progress
• Clean handoffs: Smooth transitions between automated and human intervention

The Transparency Crisis

This transparency deficit has profound implications. When an AI agent encounters a product for the first time, it must understand what the product does, what inputs it requires, and what outcomes it produces. Without explicit documentation in a machine-readable format, agents resort to inference and assumption—the same behaviors that lead to hallucination and errors in language models.

The average transparency score across all audited products measured just 5 out of 100, making it the weakest principle by a considerable margin. This suggests that most SaaS companies have not yet considered how automated systems might need to discover and understand their products.

The Shadow-UI Problem

Shadow UI refers to interface elements that appear or behave differently based on runtime conditions invisible to agents—dynamic overlays, context-dependent menus, hover-triggered controls, and client-side state that doesn't persist in the DOM structure. While these patterns create polished experiences for human users, they're often imperceptible to AI agents parsing page structure.

DocuSign, like many enterprise platforms, relies heavily on interactive elements that reveal functionality progressively. Dropdown menus, modal dialogs, and multi-step wizards all create challenges for agents attempting to map available actions. The homepage audit methodology specifically examines how these patterns manifest at the critical first-contact moment.

What Works: Machine Readability

Not all principles showed poor performance. Machine readability averaged 71 across the 100 products, the highest score of any principle. This suggests that basic semantic HTML and structured markup have become standard practice, even if products haven't optimized specifically for agent interaction.

This finding offers encouragement: the foundation for agentability already exists in most products. The gap isn't primarily technical infrastructure, but rather design intention and information architecture.

The Tier Distribution

This distribution indicates the industry stands at an inflection point. More than three-quarters of products score above 20, suggesting baseline awareness of structured interfaces, but fewer than one in four have reached the threshold where agents can operate with reasonable effectiveness.

DocuSign in Context

DocuSign's position in this landscape reflects broader patterns in enterprise SaaS. Products designed during the mobile-first era optimized for touch interfaces, responsive layouts, and progressive disclosure—all sensible choices for human users. But these same patterns create opacity for AI agents.

Electronic signature workflows involve complex state management, legal requirements, and multi-party coordination. These inherent complexities amplify the agentability challenge. An agent attempting to initiate a signature request must navigate recipient management, document upload, field placement, authentication requirements, and notification preferences—each potentially hiding behind progressive disclosure patterns.

Practical Guidance for Product Teams

The audit data points to clear priorities for teams looking to improve agentability:

Start with transparency

Add machine-readable capability documentation to your homepage and primary entry points. A simple JSON-LD schema describing what your product does, what inputs it requires, and what outcomes it produces can dramatically improve agent comprehension. This is the highest-leverage, lowest-effort improvement most products can make.

Audit shadow-UI patterns

Inventory interface elements that require hover, click, or other interaction to become visible. Consider whether critical functionality could be exposed in the initial page structure, even if styled to appear progressive for human users. The goal isn't to eliminate dynamic interfaces, but to ensure agents can discover available actions.

Document defaults explicitly

With an average score of 35, defaults represent a mid-tier opportunity. Make default values visible in form fields rather than using placeholders. Document which fields are required versus optional in machine-readable attributes, not just visual styling.

Improve status communication

At an average of 47, status communication performs better than most principles but still leaves room for improvement. Ensure loading states, process progress, and error conditions appear in the DOM structure, not just as visual indicators.

Test with automation

The most direct way to identify agentability barriers is to run a free audit or attempt to automate common workflows using headless browsers or agent frameworks. Where automation struggles, agents will struggle.

Looking Forward

The agentability gap won't close by accident. It requires intentional design decisions that balance human user experience with machine interpretability. DocuSign and its peers in the enterprise SaaS category face particular challenges given their workflow complexity, but the principles apply universally.

As AI agents become more capable and widespread, agentability will shift from a nice-to-have to a competitive requirement. Products that make themselves accessible to automated workflows will integrate more easily into agent-driven processes. Those that don't risk becoming islands in an increasingly connected automation landscape.

The good news: the foundation exists. With machine readability already strong across the industry, the path forward focuses on information architecture and intentional disclosure rather than fundamental technical restructuring.