Credit: This article was inspired by a LinkedIn post by Ian Seddon and a follow-up exchange with David V Duccini of Silicon Prairie Holdings. The original thread can be found here. What follows is an elaboration and technical expansion of those ideas, because the thread opened a door worth walking through.
The Problem No One Is Naming Correctly
There are 41 million small and medium businesses across the US and UK. Most of them have never pitched to a VC, never attended a demo day, and never had a relationship with an angel investor. They run dry cleaning shops, independent law firms, boutique manufacturers, local service businesses. They are the productive backbone of both economies.
They are also about to be blindsided.
Not by AI itself, but by AI discoverability. As autonomous agents increasingly mediate commercial interactions, businesses that are not findable, verifiable, and trustworthy to an AI agent effectively stop existing as far as new customer acquisition is concerned. This is not a prediction for 2030. It is already the operational reality for every business that depends on search traffic, and it is accelerating as LLM-powered browsing agents replace keyword queries with intent-based task execution.
The question Ian Seddon posed in his original post is deceptively simple: who do these 41 million businesses trust to guide them through this transition? And crucially, who has the structural incentive to actually do it well?
The answer, he argues, is not a SaaS company with a sales team. It is someone the business owner already knows, who is financially invested in the outcome, whose interests are therefore aligned by construction rather than by contract.
That framing is correct. But it leads somewhere more architecturally interesting than a capital formation model. It leads to a question about reputation infrastructure.
The Limitations of Identity, and Why Reputation Is the Harder Problem
David Duccini's response to Ian's post introduced a reference that deserves unpacking: Pretty Good Privacy, specifically its Web of Trust model.
PGP was designed in 1991 to solve secure communication without a central certificate authority. Instead of trusting Verisign to tell you that a public key belongs to a specific person, you trust the transitive endorsements of people you already trust. If Alice trusts Bob's key, and Bob has signed Carol's key, Alice has a basis for trusting Carol, even though they have never met. The web of trust is a reputation graph, not an identity registry.
Duccini's insight is that the internet largely solved identity (badly, through centralized login silos) but never solved reputation. Every major platform built its own reputation system as a moat: Amazon seller ratings, Airbnb reviews, Uber driver scores, GitHub contribution graphs. These systems are not portable. They do not compose. If you are a trusted seller on Etsy, that reputation is invisible to a buyer on eBay. Your reputation is imprisoned inside the platform that captured it.
This is not an accident. Reputation portability would destroy the lock-in that makes these platforms valuable. A seller with five years of verified five-star reviews on Amazon has no leverage to leave. Strip the reputation, and they start from zero anywhere else.
The coming shift is that AI agents do not respect platform boundaries the way human browsers do. An agent executing a purchase task on behalf of a user will query across sources, synthesize signals, and make a trust judgment that is not constrained to a single platform's rating system. The question of who a business is, and whether it is trustworthy, will be computed across a reputation graph that no single platform controls.
This is the structural opening. If reputation becomes portable and composable, the businesses and individuals who have accumulated it gain leverage they have never had before. And the infrastructure that enables that portability becomes foundational.
How the Web of Trust Actually Works as Infrastructure
The PGP web of trust is a directed graph where nodes are entities (people, organizations, services) and edges are signed attestations. An attestation is a cryptographic claim: "I, Alice, who you already trust, assert that this key belongs to Bob, and that I have verified this to a degree I am willing to sign." The degree of verification is part of the attestation. Marginal trust and full trust are distinguished in the model.
The critical property is transitivity with decay. Trust propagates through the graph, but attenuates with each hop. Direct endorsements carry the most weight. Third-degree connections carry very little. This mirrors how human social trust actually works, which is why it proved durable as a model even though PGP itself never achieved mass adoption.
What PGP never solved was the user experience problem: key management is operationally painful, the tooling is archaic, and the concept of signing someone's key in person at a key-signing party never scaled to a consumer population. The model was architecturally sound but practically unusable for most people.
The coming decade of cryptographic tooling, decentralized identity standards (W3C Verifiable Credentials, DID specifications), and AI agents that can reason over graph-structured data resolves most of those barriers. The agent does the graph traversal. The human provides the social relationships. The cryptographic layer provides the integrity guarantees.
A concrete architecture looks like this:
Entity Registry (DIDs)
│
▼
Attestation Graph (signed claims, verifiable credentials)
│
├── Direct endorsements (weight: high)
├── Transitive endorsements (weight: decaying by hop count)
└── Platform-imported signals (weight: configurable)
│
▼
Agent Query Layer (trust scoring at inference time)
│
▼
Decision (route purchase, recommend service, extend credit)
The agent does not ask "is this business on Google's verified list?" It asks "how many paths of verified trust connect me to this business, and how strong are those paths?" That is a fundamentally different query, and it produces fundamentally different outputs.
Capital Formation as a Trust Graph Problem
Ian's original insight, that financial investment and active advocacy should be structurally the same role, maps cleanly onto this architecture.
Traditional early-stage investment treats investors as passive capital providers who receive information rights and liquidation preferences. The ambassador role, if it exists at all, is informal and motivationally misaligned: an investor who has already written a check has declining marginal incentive to hustle for the company's success once the check has cleared.
The structure Ian is exploring, a hybrid SAFE with revenue share provisions, changes the incentive topology. An investor whose return is directly indexed to revenue has a permanent reason to introduce customers, accelerate deals, and deepen the relationship between the business and the people they know. The financial incentive and the distribution role collapse into a single node.
In graph terms: the investor becomes an edge in the trust graph, not just a node in the cap table. Their attestation, "I use this, I believe in it, and I am invested in it," carries more signal than advertising precisely because it is costly. They have skin in the game. Costly signals are credible signals. This is the core insight from signaling theory that most startup marketing ignores.
The governance design challenge Ian describes, protecting investor rights without creating voting complexity that blocks execution, resolves through a tiered structure where revenue-share participants hold economic rights without governance rights until a defined conversion threshold. The cap table stays clean. The trust graph gets dense. Both objectives are served.
What This Looks Like in Practice, Near Term
The pricing compression already underway in AI services illustrates exactly how these dynamics play out at the product layer.
Scenario A: The horizontal SaaS play, increasingly broken
A company builds an AI discoverability tool and prices it at $199/month per business. They hire a sales team, run Google Ads, attend trade shows. Customer acquisition cost is $800. Churn is 4% monthly because the business owner does not understand the product deeply enough to see the value. Lifetime value is roughly $1,500. The unit economics are marginal, and as AI pricing compresses further, the tool's perceived value erodes faster than the company can innovate.
Scenario B: The trust-graph distribution model
An operator with relationships across 200 local businesses in a regional market becomes a revenue-share participant. They are already known and trusted by their network. They introduce the tool. Onboarding happens through a conversation, not a sales funnel. The operator's financial interest means they stay engaged post-sale, answering questions, troubleshooting, advocating internally. Customer acquisition cost approaches the cost of onboarding one operator. Churn drops because the relationship layer absorbs friction that would otherwise cause cancellation. The operator earns a percentage of the revenue they generate, indefinitely.
The second model is not just cheaper to operate. It is structurally more durable because the trust that enables acquisition is the same trust that ensures retention.
Scenario C: Agent-to-agent trust negotiation
By 2027, the marginal case is not a human operator introducing a product to a human business owner. It is an AI agent, operating on behalf of a purchasing organization, querying a reputation graph to determine which service providers meet a trust threshold before initiating a commercial engagement. The businesses that have accumulated verifiable endorsements through human networks in 2025 and 2026 will score higher in those agent queries. The infrastructure advantage compounds.
This is not hypothetical. Autonomous procurement agents are already deployed in enterprise contexts. The SME version follows the enterprise version, typically with a three to five year lag. That lag is the window.
What Developers and Builders Are Misreading
The dominant framing in AI infrastructure circles treats discoverability as a technical problem: build better embeddings, optimize for semantic search, fine-tune retrieval. This is correct but incomplete.
Discoverability is ultimately a trust problem. Technical signals (schema markup, structured data, MCP server endpoints) are necessary but not sufficient. An agent deciding whether to route a $50,000 purchase order through a small vendor needs more than a well-structured API. It needs evidence that this vendor is trustworthy, reliable, and endorsed by parties the agent's principal already trusts.
The technical layer and the social layer are not separable. The builders who treat them as separable will build tools that work beautifully in demos and fail in production because they have no answer to the trust question.
The deeper misread is about who the real infrastructure is. Ian's 41 million SMEs are not the users of the infrastructure. They are the nodes in the graph. The real users are the operators who traverse the graph on their behalf, whether those operators are human ambassadors or AI agents or some hybrid of both. Build for the traversal, not just for the nodes.
What Comes Next
Three developments are worth watching closely.
W3C Verifiable Credentials and Decentralized Identifiers are achieving slow but real adoption in regulated industries. Healthcare and financial services are forcing the infrastructure to mature. Once the tooling is production-grade in those sectors, the cost of deploying it in SME contexts drops significantly.
MCP (Model Context Protocol) and similar agent communication standards are creating the query layer that will make reputation graphs machine-readable at scale. An agent that can query a business's verifiable credential graph in the same call that it queries their product catalog is a qualitatively different buying agent than anything that exists today.
Revenue-share capital structures are getting legal scaffolding. The SAFE instrument democratized early-stage equity investment. The next instrument, whether it is called a SAFE-R or something else, will democratize revenue-indexed participation. The legal infrastructure for this is being built now, and the first cohort of businesses using it will have a structural advantage in recruiting high-quality advocates.
The pieces are assembling. The builders who understand that trust is the scarce resource, and that reputation graphs are the infrastructure that allocates it, will be building the right things. Everyone else will be optimizing the wrong layer.
Originally inspired by a LinkedIn post from Ian Seddon and a follow-up exchange with David V Duccini of Silicon Prairie Holdings. Published on SnackOnAI. Thoughts, corrections, and extensions welcome.
Sponsored Ad
If you enjoy practical AI insights, check out SnackOnAI and support the newsletter by subscribing, sharing, and exploring our sponsored ad—it helps us keep building and delivering value 🚀
The News Source 2.3 Million Americans Trust More Than CNN
Tired of spin? The Flyover delivers fast, fact-focused news across politics, business, sports, and more — free every morning. No agenda. No paywall. Join 2.3 million readers who trust us to start their day right.