Written By Jaka Kotnik

John Rush published something this week that I can’t stop thinking about.

In a post on X, the serial founder -who has shipped 30+ products over two decades -described how he collapsed 30 SaaS tools into a single AI agent. No more tab switching. No more dashboards. Just one chat interface that reads, writes, and updates across every tool he uses. His agent even scouts for cheaper alternatives automatically, cancels existing subscriptions, and migrates his data overnight without him lifting a finger.

His conclusion: the UX moat is dead. Brand loyalty is dead. Switching costs are dead. Tech margins are racing to zero, and the era of the pretty-UI CRUD app is over.

He’s right. Almost entirely. But the most important implication of what he described -he breezes past it in a single paragraph.

Read More Here

The Paragraph Everyone Skipped

Buried near the end of the post, almost as an aside, John writes:

“But it can’t download the collective intelligence of millions of users and bring it with me. Like... Stripe’s fraud detection. It’s trained on billions of transactions across millions of businesses. That data IS the product.”

He calls this “Big Data” and moves on. One paragraph. Then straight to “Boring Regulated Stuff.”

I want to stay in that paragraph for a while. Because I think it contains the entire story of what value creation looks like in an agent-first world -and most people building right now are completely ignoring it.

Your Agent Is Only As Smart As What It Can See

Here’s the thing about AI agents that the breathless enthusiasm tends to gloss over: a model’s reasoning capability and the intelligence it can actually access are two completely separate problems.

We’ve spent the last two years obsessing over the thinking problem -making models smarter, faster, cheaper, more capable. And we’ve made enormous progress. The thinking problem is largely being solved.

But there’s a second problem that hasn’t been solved at all: the seeing problem.

Your agent can reason brilliantly over the information it has access to. The question is -what does it actually have access to? In most cases, the answer is: far less than you think.

91% of AI deployments fail not because the model wasn’t smart enough. They fail because the intelligence the model needed was trapped somewhere it couldn’t reach -locked inside a PDF that was never parsed properly, sitting in a database with no semantic structure, buried in years of institutional knowledge that was never digitized at all.

The agent is brilliant. But it’s reasoning in the dark.

Data Isn’t the Moat. Intelligence Is.

This is where I’d push back slightly on John’s framing.

Raw data, by itself, is not a moat. It never really was. Storage is cheap. Copying is cheap. Aggregation is cheap. An agent could theoretically scrape, collect, and compile data at a scale no human team ever could.

What can’t be easily replicated is structured intelligence -data that has been transformed, contextualized, and made genuinely accessible to the systems trying to reason over it. Stripe’s fraud detection isn’t valuable because Stripe has transaction records sitting in a database. It’s valuable because decades of signal have been compressed into a pattern-recognition layer that improves with every new input.

The raw data and the derived intelligence are not the same thing. And the gap between them is exactly where durable value now lives.

Think about what this means for your business. If you have data that:

• Gets richer the more people contribute to it

• Becomes more accurate and more useful over time

• Cannot be meaningfully replicated just by having access to the same raw inputs

• Actively makes the AI agents that consume it better at their jobs…your agent will not switch away from you. It will seek you out.

This is a fundamentally different competitive dynamic than anything SaaS has operated on before. The old moat was friction -you stayed because leaving was painful. The new moat is gravity -systems stay because the intelligence you offer is genuinely irreplaceable.

The New Infrastructure Question

John ends his post with a call to arms: stop building tools, start building what comes after.

I agree. But I’d frame the question differently for founders right now.

The question isn’t just “what do I build?” It’s “what does the intelligence layer of this new stack actually look like?”

Because here’s what’s coming: an economy of AI agents that are constantly seeking out the best available intelligence to do their jobs. They will consume structured datasets the way today’s applications consume APIs. They will pay for verified, high-quality intelligence feeds. They will choose providers not based on UX or brand -but on signal quality, reliability, and provenance.

The picks-and-shovels opportunity in that world isn’t another model, another agent framework, or another orchestration layer. It’s the infrastructure that takes the world’s trapped, unstructured intelligence -the PDFs, the databases, the institutional knowledge -and makes it legible, trustworthy, and tradeable.

Data as an asset class. Intelligence as something that can be issued, distributed, and owned.

This Is Exactly What We’re Building

At Inflectiv, we started from a single uncomfortable observation: AI agents are failing not because the models aren’t good enough, but because the intelligence they need is invisible to them -trapped in documents, locked in silos, unstructured and unverifiable.

Our answer isn’t another agent framework or another model. It’s the infrastructure layer that sits underneath all of it -the place where raw data gets liberated, structured, and transformed into something an agent can actually reason over and trust.

We call it the intelligence layer. And we think it’s the most important piece of the next stack that nobody is talking about yet.

On Inflectiv, datasets aren’t just files. They’re issued with provenance, reputation, and economics attached. They’re discoverable by agents. They improve with every contribution. They can be owned, traded, and monetized by the people who created them -not just the platforms that happened to store them.

John’s agent can switch his email provider overnight. It can migrate his database by morning. But it will actively seek out the intelligence sources that make it smarter, more accurate, and more useful -and it won’t switch away from those lightly.

That’s the gravity we’re building toward.

If you’re sitting on data that genuinely compounds -proprietary research, specialized market signals, domain expertise that took years to accumulate -Inflectiv is being built for you. Not as a place to store it. As infrastructure to make it work in an agent-first world.

inflectiv.ai