In 1998 Larry Page had an insight that changed the internet: a link from one website to another is a trust signal. Not all trust signals are equal, a link from a credible, established source transfers more trust than one from a new site with no history. Trust flows through the network. Accumulate enough credible endorsements and you become part of the trusted infrastructure of the web.
That insight built the backlink economy, shaped search rankings for twenty-five years, and created competitive moats that took years to develop and were very hard to displace.
The same logic applies in the agent economy. The currency has changed. The underlying mechanism has not.
What Made Backlinks Powerful
To understand what is happening in the agent economy, it helps to understand precisely why backlinks were such a durable competitive signal.
They were hard to fake at scale. They came from external, independent sources making their own decisions. They transferred trust from an established node to a less established one, trust that had been built over time and could not be manufactured quickly. And crucially, once earned, they kept working passively, compounding their benefit without ongoing effort.
These properties made backlinks valuable as a signal precisely because they were resistant to gaming. Anyone could stuff a page with keywords. Not everyone could earn genuine endorsements from hundreds of credible independent sources. The difficulty was the point.
Tool integrations in the agent economy share every one of these properties.
What It Means When an Agent Calls Your Service
When an autonomous agent has been configured to use your API for a specific task, retrieving data, processing documents, verifying information, executing a transaction, that configuration is a trust decision. Someone evaluated your service, decided it was reliable and appropriate, and incorporated it into a system that will call it repeatedly, automatically, without any further human decision at the point of execution.
That is not a one-time visit. It is a recurring endorsement from an automated system making decisions about who to work with. Every time the agent runs, your service is called. Every time the workflow executes, your capability is invoked. The business relationship is structural, embedded in infrastructure, not dependent on a human remembering your name.
A backlink might send a human visitor to your site once. A tool manifest integration sends automated work to your service every time the agent runs. The compounding dynamics are not comparable, they are categorically different in scale.
The Tool Manifest as First-Page Ranking
Getting into an agent's tool manifest, the defined set of capabilities an agent has permission to use, is the agentic equivalent of getting on the first page of Google results in 2005. The difference is that a search ranking might send a human visitor to your site once, at their discretion. A tool manifest means the agent calls your service automatically, every time it needs what you provide.
The businesses that achieve this position are not just visible. They are infrastructure. And infrastructure businesses have different economics from visibility businesses: lower churn, higher retention, predictable usage patterns, and a compounding advantage that grows as the agent completes more tasks and their tool selection becomes more entrenched.
Most businesses are not thinking about this yet. Which is precisely the window for the ones that are.
ChatGPT now refers approximately 10% of Vercel's new user signups, up from 1% just six months prior. AI agents drove roughly 20% of retail sales during the 2025 holiday season. The commercial channel is already significant and growing faster than any acquisition channel since mobile.
Vercel / Salesforce, 2025–2026
The Protocol Layer Being Built Right Now
In late 2024 Anthropic released the Model Context Protocol, an open standard for connecting AI systems to external tools and data sources. MCP defines a consistent interface for how agents discover, authenticate with, and call external services. The significance is hard to overstate.
Before standardized protocols, integrating a service into an agentic workflow required custom engineering on both sides, expensive, slow, and dependent on the specific agent implementation. MCP changes that. A service that publishes an MCP-compatible server can be discovered and integrated by any agent built on the protocol, without bespoke integration work for each new implementation.
For businesses, this is a publishing opportunity. Building an MCP server for your service is the equivalent of submitting your site to Google's index in 1999. The infrastructure for agent discovery is being built right now. The services that publish compatible interfaces early will have a discoverability advantage that compounds as the ecosystem grows, the same first-mover dynamic that early web businesses captured when search engines were indexing a less crowded web.
The API Gap in Most B2B Businesses
The Agent Action Score (AAS) dimension of the AVS framework evaluates specifically whether a business can be acted upon by an autonomous agent, whether an agent can access pricing, initiate a transaction, call a documented API, and integrate a service without human intervention at any step.
Across the Agentiview dataset, AAS scores are consistently lower than ARS scores in every category analyzed. The median AAS score across enterprise companies is approximately 25 out of 100, meaning the typical enterprise SaaS company, even those with reasonably good machine readability, is almost completely inaccessible to autonomous agents trying to act on what they find.
"Contact us for pricing" is an AAS score of zero on pricing transparency. A documentation site behind an email gate is an AAS score of zero on documentation accessibility. A trial provisioning process that requires a human approval step is an AAS score of zero on action interface availability. Most B2B companies, evaluated honestly against what agents actually need to transact, are commercially invisible in the agent economy.
What This Means for Competitive Positioning
The backlink analogy has a warning embedded in it: the businesses that built strong backlink profiles early had advantages that were extremely hard for later entrants to close. Years of link building, carefully accumulated authority, citation patterns that reinforced themselves, these created asymmetric competitive moats.
The same dynamic is forming in agent tool integration and API accessibility. The services that are discoverable by agents, callable by agents, and reliably available to agents today are building citation patterns in AI systems that will become harder to displace as the ecosystem matures. The services that enter the market in two or three years, when agent-mediated commerce is obviously and visibly significant, will face competitors that have been accumulating this trust infrastructure for years.
The Agent Viability Score (AVS) measures the combination of machine readability (ARS) and commercial actionability (AAS), the full picture of how a business appears to autonomous agents evaluating and transacting. The gap between ARS and AAS scores is where most businesses lose: they can be found by agents, but cannot be acted on. Being visible without being actionable is commercial inertia in the agent economy. The AVS framework was built to measure exactly that gap, and the full assessment reveals what it would cost to close it.