Six Big Shifts in AI Over the Last Year — And What They Mean for MedTech SaaS Providers
I argued that SaaS economics are breaking — per-seat licensing collapses when AI agents do the work, and the platforms that win price on validated outcomes and expose themselves through MCP. Today is a left turn. The first four shifts were about what’s changing inside your customers’ companies. This one is about what’s changing outside — how their customers discover products, how they interact with them, and who the actual buyer in the room turns out to be.
The series:
- Frameworks ate prompting
- The human-AI boundary moved
- Development taught us everything other functions now have to learn
- SaaS economics started breaking
- The customer experience is being rewritten
- Your data became your most strategic asset — and your most strategic liability
Today: The customer experience is being rewritten. And the rewrite has already happened — most B2B vendors just haven’t noticed yet.
How Customers Discover Products Now
For twenty-five years, the way a B2B buyer found a vendor was essentially one thing: Google search. You optimized for it. Your marketing team built SEO strategies around it. If you weren’t on page one, you weren’t considered. That was the game.
That game ended faster than anyone expected. G2 ran a survey in August 2025. Half of B2B software buyers now begin their purchasing journey in an AI chatbot instead of Google.Not “occasionally use” — begin. The default front door to vendor research shifted from a search engine to a language model, and most software companies didn’t notice until their organic traffic started falling.
Gartner’s 2025 B2B Buying Survey showed a related number. 67% of B2B buyers now use AI assistants during their vendor research process. A year earlier that figure was 23%. A 3x jump in twelve months. That’s not a trend you have time to study. It’s a transition you’re already behind on.
A new software category appeared on G2 specifically to address this — Answer Engine Optimization, or AEO. Also called Generative Engine Optimization, GEO. The category grew over 2,000% in a single year — an entire tooling ecosystem materialized out of nothing because the problem is real and urgent. SEO agencies are rebranding. Marketing teams are reorganizing. The skill of “making sure you’re in the answer” is now its own discipline.
The 2X AI Visibility Index, released in April 2026, put hard numbers on this. They analyzed 70 B2B companies to measure how often brands appeared across generative AI environments buyers use for research. The finding: only 4.3% of companies maintain a healthy AI discovery funnel. 96% are effectively invisible at the earliest stage of the buying journey — appearing only when the buyer already knows the company name.
What separated the 4%? Two factors. Technical readiness for AI discovery — meaning structured, machine-readable content that AI systems can ingest. And authority signals — domain expertise that positions a vendor as an authority in a specific problem space. The companies scoring highest were classified as “Authority Leaders”: vendors with deep domain expertise expressed in formats AI systems discover and incorporate into their reasoning.
Here’s the question every medtech SaaS vendor has to answer. When a regulatory affairs director asks ChatGPT or Claude or Perplexity — “what’s the best quality management system for a Class II MedTech company?” — are you in the answer? Because if you’re not, the buyer will never know you exist. They aren’t going to check page two of Google. They aren’t going to double-check by searching again. The answer they get is the consideration set. And the vendors who appear are the ones whose technical documentation, validated case studies, and regulatory authority are structured in ways AI systems recognize and amplify.
The Customer May Never Open Your Product
The discovery change is only half of the story. The other half is what happens after discovery — how the customer actually uses your product.
Think about how customer experience used to work. Marketing gets the customer interested. Sales closes the deal. Product onboarding teaches them the interface. Customer success makes sure they adopt the features. Every team along that chain is designing for a human who opens your application, logs in, looks at buttons, reads labels, clicks through flows. That was the assumption every enterprise software product was built on.
Get Will Macdonald’s stories in your inbox
Join Medium for free to get updates from this writer.
Now walk through what actually happens when your customer is an AI-native user. A regulatory specialist at one of your medtech customers needs to research predicate devices for a 510(k) submission. Two years ago, she opens your regulatory software, navigates to the predicate search, runs queries, exports results, copies them into her submission workflow. She spends thirty minutes inside your product.
This year, she asks Claude to find predicate devices for her specific device and indication. Claude reaches through the MCP layer — yours or a competitor’s — pulls the same data, synthesizes it, drafts the substantial equivalence section of her submission, and hands her a working draft. She never opened your software. She never saw your logo. She doesn’t know if the answer came from your product or a competitor. If your data was accessible through her agent, you’re in the submission. If it wasn’t, you’re not.
“Easy to use” used to mean legible to a human. It now means reachable by an agent that has never seen your marketing.
That’s the shift. Everything you spent a decade optimizing — funnel conversion, landing page layout, onboarding flow, in-app tutorials, release-note communication — is built for an audience that may never show up. Not because those things are wrong, but because the audience shifted. The human may still use your product some of the time. The agent representing the human is now doing most of the work.
For medtech specifically, this compounds with the regulatory dimension. The agent doesn’t just need to reach your data — it needs to know whether your data is regulator-grade. When the agent drafts a 510(k) substantial-equivalence argument, does it cite predicate device data from your platform with audit-trail evidence? Or does it pull from an open database and hope for the best? Your platform’s auditability becomes the agent’s preferred source — but only if the agent can verify the audit trail without leaving the workflow.
The practical question is concrete: what parts of your customers’ work are moving into the agent, and what parts stay human? And for the parts that move to the agent — how does your product reach them?
When the Agent Is the Buyer
Here’s the reframing that matters most strategically. For as long as enterprise software has existed, winning a customer meant convincing a human. You spent money on brand. You built relationships. Your field sales team took people to dinner. Your CEO met with their CEO. The entire GTM motion of enterprise software was built on the premise that at some point in the decision, a human was looking at you and deciding whether to trust you.
When the buyer is an agent, none of that matters.
Agents don’t care about your brand. They don’t remember your dinner. They don’t know who your CEO is. They choose based on what they can verify: is your data accurate, is your API stable, is your latency acceptable, is your output legible enough to use without reformatting.That’s the evaluation.
And the early evidence confirms this isn’t theoretical. B2B marketing teams deploying AI agents for vendor evaluation are already finding that agents systematically prioritize vendors whose technical documentation is comprehensive and accurate over vendors with stronger brand recognition but weaker technical information. It’s a return to technical merit as the primary differentiator — but with the mediation layer being AI agents rather than human experts. Your technical merit has to be expressed in ways AI systems can access and interpret accurately.
This sounds alarmist. Let me be specific about what’s actually happening and what isn’t. Humans aren’t going away as decision makers — not for major purchases, not for strategic platform choices, not for regulated-industry procurement. The human signs the contract. But the shortlist that reaches the human is increasingly curated by agents acting on the human’s behalf.The agent eliminates candidates before the human ever sees them. If you’re not in the shortlist, your brand can’t save you.
So the question for every medtech SaaS company is the one to put on a whiteboard tomorrow: how do you get to the top when AI agents are the ones doing the searching? The answer isn’t better SEO. It isn’t more sales. It isn’t better brand campaigns. The answer is: does your product give the agent something it can verify?
For medtech, the verification surface is unusually rich. Validated workflows. Documented predicate device libraries. Audit trails. Regulatory-submission evidence. Quality system documentation. These artifacts were created for regulators and procurement. They turn out to be exactly the kind of structured, authoritative content AI systems weight heavily when assembling shortlists. Your regulatory documentation isn’t just a compliance asset. It’s your AEO surface.
What This Means for MedTech SaaS
The GTM motion that worked for the last twenty-five years was built on three assumptions: that buyers find you through Google, that buyers use your product through your GUI, and that the person making the purchase decision is human. All three assumptions are weakening at the same time, and the platforms still operating on them are losing visibility faster than they realize.
The practical reorientation has three components.
First, treat regulatory and technical documentation as AEO content, not just compliance overhead. Your 510(k) case studies, validation summaries, audit-trail capabilities, and integration documentation are the structured authoritative content AI systems weight when answering “what’s the best platform for X?” If your documentation lives behind a login wall or as a marketing PDF, the agent can’t see it, can’t cite it, and won’t shortlist you.
Second, make your platform reachable by agents. Article 4 covered the four levels of accessibility — GUI-only, REST API, MCP layer, agent-native. The MCP layer is the new product surface. If your competitor ships MCP first and your data is locked behind a GUI, the agent talks to them. Your customer never knows you were an option.
Third, instrument what agents are asking about your platform. If an agent reaches your API or your MCP layer, you have logs. Treat those logs the way you used to treat search analytics. What queries are agents running against your platform? Which ones return rich answers, and which ones return nothing useful? That’s your new SEO research — and it tells you exactly where your AEO investment will pay off.
The platforms that pivot now will be the ones in the shortlist when a customer’s agent asks “what’s the best regulatory submission platform for our device class?” The platforms that don’t will spend the next eighteen months wondering why pipeline is collapsing while their competitors’ is growing.
Next article — the series close. Your data just became your most strategic asset — and your most strategic liability. Three layers of data moat, the closed-loop dynamics that compound, why HIPAA / GDPR / MDR turn the moat into a litigation surface if you get governance wrong, and the three Monday-morning actions every medtech SaaS leader should take.
