What’s Happening
SurfaceGX, a startup positioning itself as an AI Visibility Repair Infrastructure platform, has published its first benchmark study scanning 26 recognized B2B SaaS brands across six dimensions of AI discoverability. The report, titled “The Agent Readiness Gap,” finds that the cohort has largely mastered the technical foundations of AI search readiness, averaging 76.2 out of 100 against a B2B SaaS sector average of 58. But the study exposes a uniform and striking failure at the next layer: not one company in the 26-brand group has exposed a discoverable agent interface at its primary domain, and primary-domain AI guidance scores averaged only 29.8 out of 100. Three brands, Sprout Social, Monday.com, and Asana, achieved perfect overall scores, while the agent readiness score for 24 of 26 was zero.
The Bigger Picture
The SurfaceGX findings matter well beyond the 26 brands in the sample. They illuminate a structural inflection point in how enterprises need to think about their digital presence: one where the audience is increasingly a machine, not a human.
The GEO and AEO Gap Is Real, and This Data Quantifies It
For years, enterprise marketing and web teams have optimized for human readers arriving via search engines. That discipline produced the strong citation readiness scores SurfaceGX observed, a cohort average of 77.6, reflecting well-structured content, clean crawlability, and solid page experience signals. The good news for ITDMs is that this investment was not wasted. It transfers, at least partially, to AI answer engines that share many of the same technical requirements Google confirmed in its May 2026 generative AI search guidance.
The problem is the next layer. AI agents, including autonomous systems built on Model Context Protocol (MCP), need machine-readable discovery signals at predictable primary-domain endpoints: WebMCP manifests, mcp.json files, agent.json declarations. These are small files with outsized consequences. They tell an AI system what a brand’s agents can do, how to invoke them, and what scope of authority they carry. None of the 26 companies studied had published one at their primary domain.
This is not a capability gap. It’s a discoverability gap, which makes it both more embarrassing and more fixable. Several cohort brands, including Asana, Gong, Klaviyo, Sprout Social, Webflow, and Monday.com, have already shipped documented MCP servers. Those servers exist on developer subdomains or behind authentication. The primary domain simply does not announce them. The infrastructure exists. The front door is unlocked but unmarked.
What This Means for ITDMs
For IT decision-makers, the business risk here is straightforward. As AI agents become the primary interface through which buyers research, compare, and interact with enterprise software, brands that are not discoverable by those agents will be invisible in an increasing share of decision-making workflows. ECI Research’s 2025 AI Builder Summit survey found that two-thirds of enterprise AI leaders have already implemented multi-agent collaboration in live or pilot workflows. That figure reflects the demand side of this equation: enterprise buyers are deploying agents that will query, evaluate, and potentially integrate with vendor products programmatically. If a vendor’s agent capabilities are not discoverable from the primary domain, those workflows will route around them.
The economics are blunt. A brand can have a fully functional MCP server with well-documented capabilities and still lose agent-driven discovery to a competitor that has published a simple agent.json at its root domain. This is not a theoretical future risk. The SurfaceGX scan was run today, and the gap is open today.
ITDMs evaluating their own organization’s AI presence should treat primary-domain agent discoverability as a new category of technical debt. It sits alongside structured data hygiene and crawl access in the hierarchy of requirements, but unlike those disciplines, which took years to normalize, the remediation here is relatively low-effort and high-signal.
What This Means for Developers
For developers, the findings land differently. The MCP infrastructure question is not “can we build this?” The question is “have we wired it to the front door?” The SurfaceGX methodology specifically checks primary-domain endpoints because that is where AI systems look first. A server running on docs.vendor.com or behind an OAuth wall is invisible to an AI agent doing cold discovery.
The practical implication is that platform and web engineering teams need a new checklist item in their deployment and release processes. Publishing and maintaining an agent.json or WebMCP manifest at the primary domain is not a complex engineering task. It is a configuration and governance task, one that requires coordination between the teams building agent capabilities and the teams managing the primary domain. That coordination appears to be the real missing link, not engineering skill.
ECI Research’s 2025 AI Builder Summit data found that 44% of enterprise AI leaders have only moderate confidence that AI agents can act autonomously without human intervention. That hesitancy is understandable given where tooling maturity sits today, but it does not reduce the urgency of making agent capabilities discoverable. Discovery is a prerequisite for trust-building, not a consequence of it.
Looking Ahead
Agent Discoverability Will Become a Procurement Criterion
The SurfaceGX benchmark is a first-generation instrument measuring a first-generation problem. As AI agent ecosystems mature, expect this category of readiness to migrate from “interesting benchmark” to “standard vendor evaluation criterion.” Enterprise procurement workflows increasingly involve agentic research steps, and vendors that surface their capabilities via standardized machine-readable contracts will receive preferential treatment in automated shortlisting. The brands in this cohort that act in the next 6–12 months will build a compounding advantage: AI systems that discover and successfully invoke their agents early will reinforce that discoverability through citation and recommendation feedback loops.
The Standards Layer Will Consolidate Quickly
Model Context Protocol has emerged as the leading candidate for agent discovery standardization, with major platforms and tooling vendors coalescing around it through 2025 and into 2026. The current fragmentation between WebMCP manifests, mcp.json, and agent.json signals reflects an early-stage standards environment. ITDMs and developers should expect consolidation within 12–18 months, which will lower the cost of compliance and raise the visibility cost of non-participation. Organizations that begin publishing and maintaining primary-domain agent contracts now, even in draft form, will be better positioned to adapt to whatever final standard emerges than those starting from zero.
The practical recommendation is direct: audit your primary domain for agent discoverability today, map which of your MCP or agent capabilities are deployed but not announced, and assign ownership to close that gap. This is not a roadmap item. It’s a configuration task with a disproportionate strategic return.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.
