The AI-Native Illusion
We're really talking about Interface-Native....
Most companies marketing themselves as AI-native aren’t.
They’re Interface-native. There’s a difference, and the difference matters more than anyone in the current hype cycle wants to admit.
Strip the AI-native label off most of the companies proclaiming this and look at what’s actually happening.
For example, the AI SDR startup that raised at a billion-dollar valuation is orchestrating LinkedIn Sales Navigator for identity, Apollo or ZoomInfo or Clay for enrichment, Outreach or Salesloft for delivery, and writing back to Salesforce for the system of record.
The AI is doing the personalization and the orchestration. But the underlying systems are entirely traditional. Pull any one of those underlying systems out and the AI SDR has nothing to operate on.
The same is true of the AI revenue intelligence platforms ingesting from Salesforce and Gong, the AI sales coaching tools analyzing what Gong already captured, the AI agentic platforms automating workflows between SaaS tools that have existed for fifteen years.
This isn’t a critique of these tools, some are quite useful. The personalization is real. The orchestration is real. The efficiency gains are real. But none of it is transformation, and the “AI-native” marketing by these organization is positioning them as transformation.
We aren’t actually buying AI-native systems, we are buying an interface layer. It is a much slicker, smarter, faster way to access systems that already existed. These tools provide much easier ways to integrate disparate databases and workflows. Using traditional tools, it takes months or years to achieve the same level of integration. So these Interface-native tools can be very powerful.
But these Interface-native tools are not the future of enterprise software, though many of them imply this in their marketing. The reason is, without the current enterprise software platforms, they would crumble. They would cease to exist.
The traditional platform vendors are making similar claims, but from a slightly different perspective. Salesforce shipped Agentforce and has recently announced Headless 360. HubSpot shipped Breeze. ServiceNow has its agents. Microsoft has Copilot threaded through everything. Each was launched with the marketing language of fundamental rethinking.
Architecturally, each is an agent interface on top of the same data model, the same workflow engine, the same business logic that existed before. Salesforce in 2026 with Agentforce is still Salesforce. The object model didn’t change. The fundamental approach to managing customer relationships didn’t change. What changed is how you access and act on what’s already there. Basically, a slicker user interface. There’s value in that. Sometimes substantial value. But it’s interface value, not architectural value.
And the vendors marketing it as transformation are doing the same thing the startups are doing, selling efficiency disguised as something it isn’t, at least yet.
Here’s where it gets interesting for buyers. The startups in the first category, the AI SDRs, the AI revenue platforms, the AI coaching platforms, the agent orchestration companies aren’t just dependent on the underlying platforms. They’re competing with those same platforms.
When Salesforce ships interface capabilities that overlap with what an AI SDR startup, the AI revenue platforms, or the AI coaching tools provide, the startup’s value proposition compresses fast. The startup has no defense because its entire architecture depends on Salesforce, and Salesforce can offer the same interface layer at a marginal cost. Buyers who chose the startup for AI capability now find those capabilities in the platform they already pay for. A lot of these companies aren’t going to survive what’s coming.
This isn’t speculation. Watch what’s happening in adjacent categories. The AI meeting summary startups are being absorbed by Zoom and Teams adding the capability natively. The AI email assistants are being absorbed by Gmail and Outlook. The AI document tools are being absorbed by Microsoft and Google. The interface layer is the easiest layer for the platform vendor to absorb, and the platform vendors are absorbing it.
So when you’re filling your stack with Interface-native tools, you’re not just paying premium prices for AI capabilities. You’re accumulating fragility. Every Interface-native tool is a dependency on the the underlying vendor surviving the absorption cycle.
Companies that buy the platform vendors’ agent layer get some efficiency lift without architectural change, which is fine if you know that’s what you’re buying. It’s dangerous if you think you’re transforming.
We’re seeing the buyers of these tools recognize this. Many are refusing to license the tools in the ways traditional SaaS licensing worked, insisting on shorter commitments, freeing them to move from interface-tool to interface-tool.
None of this is happening because vendors are uniquely deceptive or buyers are uniquely gullible. It’s happening because the buying market and the vendor market have settled into a story line that fits both the interface providers and the users.
Boards want AI stories. CEOs want to tell the board they’ve deployed AI. CROs want to show the CEO they’re modernizing the GTM. Interface-native tools give everyone a credible answer without forcing the harder strategic conversation about whether the entire GTM strategy must change.
The vendors are giving the buyers what the buyers are rewarding. The buyers are getting what they want, which isn’t transformation, it’s the appearance of transformation.
This same story line is playing out inside companies. Sales managers building their own AI coaching tools. Reps wiring up call-prep agents. Marketing teams standing up workflows that duplicate what rev ops is already doing. Each project feels like a win. Almost none of them move the actual customer outcomes the company exists to produce. The pattern at the buying level is motion that resembles transformation but produces only efficiency. The same as what we see at the operating level. Different layers of the stack. Same, but nothing new.
And here is where the data starts to matter, because the story line is increasingly hard to defend with results.
In August 2025, MIT’s NANDA initiative published *The GenAI Divide: State of AI in Business 2025*. The headline finding is that 95% of enterprise generative AI pilots produce no measurable impact on the P&L, against $30–40 billion in enterprise spending. Only 5% deliver significant value.
McKinsey’s 2025 State of AI report finds that nearly nine in ten companies have deployed AI in at least one business function, while 94% report not seeing significant value from those investments.
The National Bureau of Economic Research surveyed roughly 750 corporate executives in late 2025 and early 2026 and found that, where productivity gains are measurable, they sit in the range of 0.4 to 0.8%, modest by any historical standard for a technology being marketed as transformative.
Daron Acemoglu’s macroeconomic model published the same year puts the ceiling on AI’s contribution to total factor productivity at less than 0.66% over a ten-year horizon.
The numbers tell a consistent story. Massive adoption. Massive investment. Marginal returns.
The Interface-native crowd has an explanation for this; adoption gaps, integration challenges, organizational learning curves, the productivity J-curve that makes new technologies look weak before they look strong. All of those are real. None of them are the whole story.
The whole story is that you can’t get transformative results from non-transformative deployments. If the architecture is interface on top of legacy, the gain is interface gain. Faster access. Cleaner search. Smarter summarization. A few hours saved per rep per week, sometimes more. That’s worth something, but it isn’t business model change, and the data is showing exactly that.
The companies seeing 0.8% productivity gains aren’t doing it wrong. They’re doing exactly what the Interface-native marketing told them to do, and getting exactly what that approach can produce.
There is a third category. A small number of companies are building from the ground up around what AI makes possible rather than what was possible in 2005, when most enterprise software architecture was set.
Some of the legal tech work, new contract analysis and legal workflow platforms, is closer to this than to Interface-native, because they’re rebuilding legal workflows rather than wrapping existing legal tech.
Some of the engineering and code generation work, the agent-first development environments now emerging — is building toward something new rather than putting copilots into existing development environments.
The medical imaging and diagnostic companies are AI-native by definition, because their core capability couldn’t exist without the model.
Some of the customer service work being done by newer entrants is rebuilding the support experience around what an AI agent can actually do, rather than putting an interface layer over a ticketing system.
Most of even these companies are still partly Interface-native. The AI-first marketing outpaces the AI-first architecture. But the AI-first architects are trying to redesign the underlying work, not just the access interface to existing work. That’s a meaningful distinction even when execution is incomplete.
The honest assessment is that almost none of the visible AI-native vendor space is doing transformation. They’re doing efficiency. The AI SDR tool makes outbound cheaper and faster, but the outbound itself is the same activity, targeting the same buyers, with the same value propositions, through the same channels. That’s efficiency.
Transformation would be reaching buyers in ways that weren’t possible before, or eliminating the need for outbound entirely by changing how demand is created. Almost no one is doing that work, and it’s not because the technology can’t support it. It’s because building genuinely new AI-first platforms is hard, slow, capital-intensive, and the market hasn’t been trained to value it. Building interface layers is fast, cheap, demos well, and rides the hype cycle. The capital markets reward the second behavior. The buying patterns reward the second behavior. So the field is full of interface work, not because that’s where the real opportunity is, but because that’s where the easy money is.
This is where buyers have to make a choice that almost no one is making consciously. You can keep buying interface layers, get some efficiency lift, accept that it is probably a temporary step, and tell the board you’re AI-enabled.
That’s a reasonable strategy as long as you’re honest about what you’re doing. Or you can start asking what AI makes possible for your business, not what it lets you do faster, but what it lets you do that you couldn’t do before. What customer outcomes does it open up that weren’t reachable? What work could be eliminated entirely rather than accelerated? What does the business look like if you started from the AI capability rather than retrofitting it onto what already exists?
Those questions are harder. They don’t have a vendor answer. No one is going to send you a deck. The Interface-native crowd can’t help with them because the questions point past what the Interface-native crowd is doing. The traditional platform vendors can’t help with them because the answers might require leaving the platforms behind. They are searching for those answers themselves. The genuinely AI-native companies are too quiet, too early, and too far from where most enterprise buying happens to drive the conversation yet.
But that’s where the actual possibility is. And it’s where the conversation has to go if AI is going to mean anything more in five years than a slightly faster way of doing what we already do. The data we have so far suggests it won’t mean much more than that for the companies that don’t ask the harder question. A 0.8% productivity gain is not what the next decade was supposed to be about.
The hype cycle will run its course. The Interface-native startups will mostly get absorbed or fail. The platform vendors will have added agents to everything and called it transformation. The companies that win the next decade will be the ones that asked the harder question early, not what can AI do for our existing business, but what could the business be if we built it around what AI makes possible.
Most leaders aren’t going to ask that question. They’ll buy the interface layer, check the AI box, and tell themselves they kept up.
The ones who do ask it will build the businesses that make the rest look like they were standing still.
Afterword: This is the first in a series about how we rethink AI—and how we rethink businesses leveraging AI.
Afterword: I love the discussion and examples the speakers in this AI generated discussion of this article. These examples help visualize the issues I’m discussing very nicely. Enjoy!