Imagine a hospital manager asking an AI agent to find software for coordinating outpatient referrals. The agent compares products, checks public documentation, reads security pages, reviews integrations, searches for case studies and returns a shortlist.
Which company wins a place on that list?
A large vendor with thousands of customers, a familiar brand and years of online documentation may have an obvious advantage. A younger company with a better solution for the hospital's exact problem may never make it into the recommendation. No person has rejected it. The agent simply found more machine-readable evidence for somebody else.
This is a hypothetical healthcare scenario. But it reflects a concern Matthew Prince raised during Web Summit 2026. Onstage with Lovable's Anton Osika and Recursive's Richard Socher, the Cloudflare co-founder argued that buying agents may behave differently from human customers, placing less weight on emotional affinity or local convenience and more weight on signals that tend to favour incumbents. He warned that agentic commerce could make the market more concentrated if new entrants cannot compete effectively.
The implications for HealthTech deserve attention because healthcare buying already places a heavy premium on reputation, references, risk management and compliance.
Why the buying journey may change
Today's HealthTech founder usually imagines a human journey. Someone discovers a post or attends a conference, visits the website, downloads a case study and eventually meets the sales team. The founder's story and credibility help start the conversation. Then come technical, clinical and procurement reviews.
As agents become more capable, some of the early research may be delegated. A buyer could ask an assistant to identify suppliers that meet a defined set of requirements and return a comparison. The agent might favour vendors with clear integration documentation, public pricing or a security certification that it can verify from accessible sources.
That could be beneficial. It may reduce the time buyers spend gathering basic facts and allow smaller procurement teams to examine more options. But the result depends heavily on which sources the agent can access and how it judges credibility. A model that treats “most mentioned online” as equivalent to “best fit for this hospital” could simply automate an existing market bias.
In a sector where the newest specialist company may have limited publicity but strong clinical understanding, that matters.
Healthcare already rewards established names
Hospitals have rational reasons to be cautious. Software may connect to sensitive data, affect clinical decisions or become embedded in services that cannot tolerate prolonged outages. Buyers want financial stability, support arrangements and evidence that another organisation has successfully deployed the product.
A startup must therefore clear several hurdles that a recognised supplier may find easier. It may not yet have a broad reference network, mature integration partnerships or a long history of responding to security questionnaires. Some of that is a genuine risk difference. Some is simply the difficulty of proving competence before enough customers have given the company an opportunity.
Now add an AI agent that ranks suppliers using available evidence. If the process rewards size without properly assessing product fit, the established vendor becomes even more visible. The startup receives fewer chances to demonstrate its strengths. Fewer opportunities mean fewer references, which in turn weakens its position in the next automated search.
That is the potential feedback loop behind Prince's concern. It is not evidence that hospitals are already handing procurement decisions to AI agents. It is a scenario to plan for as discovery and comparison tools change.
What becomes visible to an agent?
This should prompt a new kind of go-to-market audit. Look at a startup's website from the perspective of someone, or something, trying to establish the basics without a sales call.
Can a buyer understand the product's intended use? Is the target customer clear? Are supported integrations identified? Is clinical evidence accessible and accurately described? Are privacy practices and security responsibilities documented? Is the organisation transparent about where it operates and which services it supports?
In healthcare, the lack of clear information often gets disguised as mysterious positioning. A founder knows the product intimately, so the website becomes a string of ambitions: transform care, empower clinicians, redefine patient journeys. A human might tolerate that and agree to a meeting. An agent asked to compare six suppliers may have little usable evidence to extract.
This does not mean publishing confidential contracts or every commercial detail. It means creating a clear public account of the product's scope, evidence and operational commitments. A concise implementation description may be more valuable than another elaborate slogan.
The same principle applies to regulatory claims. If software is marketed for a medical purpose, founders must be precise about regulatory status and intended use. If a tool is purely administrative, calling it a clinically validated medical device would mislead rather than impress. Structured, verifiable information becomes a commercial asset.
The case for human relationships may get stronger
Prince's point does not mean founder-led sales will disappear. In complex healthcare purchases, a serious buyer needs to understand local workflows, risk tolerance, clinical priorities and what successful implementation will look like. Those conversations are difficult to reduce to a standard product comparison.
A well-run pilot can reveal value that a generic agent cannot infer from the open web. A discussion with an experienced clinical founder can uncover a service problem the buyer had not properly articulated. A transparent conversation about what a product cannot yet do may create more trust than a perfectly polished presentation.
This is where young companies can still compete. They can be unusually responsive, narrow in focus and willing to design around a specific operational challenge. The commercial goal is to make those strengths visible and measurable without relying entirely on the buyer already knowing the founders.
For an early-stage team, that may mean building fewer broad marketing claims and more evidence-rich case studies. Show the original workflow, the implementation, the limitations and what improved. Include enough detail for a potential buyer to judge whether the circumstances resemble their own.
But who decides the ranking criteria?
Automated procurement also raises a governance question. If an agent is instructed to favour low price, it may undervalue implementation support or clinical quality. If it favours market share, it may exclude innovative suppliers. If it uses a proprietary data source, entire classes of vendors could be underrepresented.
A hospital should therefore treat an agent-generated shortlist as a research aid, not a neutral truth. Buyers need to know what the system searched, what evidence it used, which criteria were weighted and whether gaps in the data affected its conclusion. Human procurement and clinical teams remain responsible for validating the shortlist against the actual need.
This is especially important when software performance varies across patient groups or use cases. The most visible vendor is not automatically the safest, and the least expensive proposal is not automatically the best value.
What HealthTech founders should do now
The practical response is not to optimise a website for every hypothetical AI agent. It is to publish a more complete, accurate and accessible account of the business. Product documentation, evidence, implementation information, security practices and honest use-case boundaries should be easy for both people and machines to find.
Founders should also diversify how trust is built. Clinical partnerships, conference presentations, peer-reviewed evaluation where appropriate, respected customer references and relationships with specialist communities will matter even if the discovery process changes. Human experts may increasingly be the people who question and refine the machine's shortlist.
Prince described the wider internet becoming more agent-oriented. He also argued that the goal should be a market in which many companies can compete rather than one dominated by a handful of incumbents.
For HealthTech, that goal is more than a competition slogan. Hospitals and patients benefit when promising new approaches have a credible route into the market. The challenge is to build automated buying systems that improve scrutiny without confusing familiarity with quality.
Startups should prepare for the possibility that their next first impression is not a meeting, a pitch deck or a conference booth. It may be a machine deciding which companies are worth showing to a human. Give that machine sound evidence to find, while building the relationships that make evidence meaningful.
Key Signals
If procurement agents shortlist vendors from public documentation, security pages and case studies, the advantage shifts to companies with deep, machine-readable evidence, which usually means incumbents. This is an extrapolation from Matthew Prince's warning, not an established fact about hospital buying today, but the direction is worth taking seriously. Startups should respond by making their evidence, integrations and security posture legible to machines as well as to people.
