Gong Earns 49/100 Agentability Score as Conversation Intelligence Platforms Face Agent Integration Challenges

Conversation intelligence platforms like Gong occupy a unique position in the SaaS ecosystem: they already process vast amounts of unstructured data, apply AI to extract insights, and surface recommendations to human users. Yet when it comes to enabling AI agents to operate their interfaces, these platforms face many of the same challenges as the broader software industry.

According to data from The Agentability Project's audit of the top 100 SaaS products, Gong's homepage earns an agentability score of 49/100, placing it in the Agent-Ready tier and above the industry average of 38.3. This score reflects how well the product's interface can be operated by AI agents—software that autonomously navigates applications, extracts information, and completes tasks on behalf of users.

Understanding Agentability

Agentability measures how easily AI agents can interact with software interfaces. The metric evaluates products across eight Agent Factors Engineering (AFE) principles, each scored 0-100:

• Machine readability: Whether UI elements are programmatically identifiable and semantically meaningful
• Transparency: Whether the application exposes its state, schema, and capabilities in machine-readable formats
• Shadow-UI avoidance: Minimizing dynamic overlays, modals, and client-side rendering that obscure content
• Defaults: Sensible preconfigured values that reduce decision points
• Control: Mechanisms for rate limiting, permissions, and safe operation
• Chunking: Breaking complex workflows into discrete, resumable steps
• Status: Clear feedback about operation state and progress
• Clean handoffs: Smooth transitions between agent and human control

Products scoring 45 or above qualify as Agent-Ready, indicating that while challenges remain, the fundamental architecture supports agent interaction. Gong's 49 places it among 22 products in this tier, compared to 54 in the Developing category (35-44), 17 Lagging (20-34), and 7 Agent-Blind (under 20).

Industry-Wide Transparency Crisis

Transparency measures whether applications expose structured information about their capabilities, data models, and current state in formats that agents can programmatically consume. This might include API schemas, sitemap annotations, structured data markup, or explicit capability declarations.

The near-total absence of transparency signals across the SaaS industry represents a fundamental mismatch between how modern web applications are built and how agents need to interact with them. Products designed exclusively for human users provide visual affordances—buttons, labels, color coding—but rarely expose equivalent machine-readable metadata about what actions are possible or what data means in context.

For conversation intelligence platforms specifically, this creates a paradox. Gong processes sales calls, emails, and meetings to extract structured insights from unstructured conversations. Yet the platform itself does not expose structured metadata about its own interface, forcing agents to rely on the same visual parsing techniques that humans use.

The Shadow-UI Problem

Shadow-UI refers to interface elements rendered dynamically through JavaScript—modals, dropdowns, tooltips, and overlays that appear on demand rather than being present in the initial page structure. These patterns create significant challenges for agents, which must trigger specific interactions to reveal hidden content, handle timing-dependent rendering, and navigate layered interface states.

Modern SaaS applications lean heavily on these patterns for reasons that make sense in human contexts: they reduce visual clutter, provide contextual help, and create fluid interactions. But they impose substantial complexity on agents, which must maintain state models of what UI elements exist conditionally and execute precise interaction sequences to access them.

Where Gong Excels

Despite sharing industry-wide weaknesses in transparency and shadow-UI avoidance, Gong demonstrates relative strengths that contribute to its above-average score. The homepage audit evaluates machine readability at 71/100, well above the industry average for this principle.

Machine readability reflects the quality of HTML semantics, ARIA labels, consistent naming conventions, and structural clarity. Higher scores indicate that agents can more reliably identify interactive elements, understand their purpose, and predict the outcomes of interactions. For platforms with complex feature sets like conversation intelligence, this foundation matters significantly.

Interpreting the Broader Context

The audit reveals dramatic variance across AFE principles. Machine readability averages 71/100 across all products—a relatively strong showing that reflects years of web accessibility efforts and semantic HTML advocacy. Control mechanisms average 48/100, and status feedback averages 47/100, both approaching adequacy.

But the other 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

This distribution suggests that current web development practices accidentally support some agent needs (through accessibility work that improves machine readability) while actively undermining others (through architectural choices that assume human operators).

What This Means for Product Teams

For teams building conversation intelligence platforms or similar complex SaaS products, several concrete steps can improve agentability:

Expose structured metadata. Implement schema.org markup, OpenAPI specifications for web interfaces, or custom capability declarations that tell agents what your application can do. Even basic structured data about page purpose and primary actions provides value.

Audit dynamic UI patterns. Identify modals, dropdowns, and overlays that hide functionality. Where possible, provide alternative access paths that don't require hover states or precise click sequences. Consider progressive enhancement approaches that work without JavaScript.

Document state machines. Complex workflows like deal review, forecast analysis, or pipeline management involve multiple states and transitions. Make these state models explicit and exposable, not just implicit in UI behavior.

Design handoff points. Identify where agent automation should pause for human judgment. Make these boundaries explicit through API design, UI structure, or policy declarations.

Test with automation tools. Use browser automation frameworks to interact with your interface. Points where automation struggles often indicate agent challenges. Teams can run a free audit to establish a baseline measurement.

The Road Ahead

Gong's 49/100 score places it ahead of most SaaS platforms in agent compatibility, but the absolute number reveals how far the entire industry must progress. An agentability score below 50 indicates that more than half of the design and architectural decisions that affect agent operation remain unaddressed.

As AI agents become more prevalent in enterprise workflows—handling research, data entry, monitoring, and routine tasks—software that accommodates agent operation will deliver compounding advantages. Products that remain navigable only by humans will require expensive custom integrations or force organizations to maintain hybrid workflows that undermine automation benefits.

The Agentability Project's the Agentability Index provides benchmarks across 100 products, revealing that agent-ready software remains the exception rather than the rule. For conversation intelligence platforms positioned at the intersection of AI capabilities and human workflow, this represents both challenge and opportunity.