Defining Intelligence: What Actually Wins in an Agentic World?

Every conversation about AI defensibility eventually arrives at the same place: data. Whoever has the most proprietary, hard-to-replicate, longitudinal data wins. Data is arguably the single most defensible asset a software company can have and the recent surge in valuations for companies sitting on decades of structured information is rational, not exuberant. The market is right to pay up for genuine data scarcity.

But “data is the moat” has become a thought-terminating shortcut. It points to an asset rather than defining what makes that asset useful. The more interesting question, and the one I want to address here, is what it actually means for a software company to be intelligent.

At FTV Capital, we’ve spent nearly three decades watching how data assets compound and evolve into intelligence at high-growth software companies. In an agentic world where foundation models are converging, the difference between simply owning data and being intelligent matters more than ever. The two aren’t the same thing: data is raw substrate (signals, transactions, observations, etc.), while intelligence is the layer that adds context to turn those inputs into something useful.

I believe intelligence shows up in four distinct ways. All four involve some form of proprietary asset, but they differ in how data is transformed and how the resulting intelligence compounds. Here’s how I think about it:

1.      Source Intelligence. Being the most complete record of something the world needs to look up.

Bloomberg has observed financial markets in granular detail for 40 years. S&P has tracked the constituents of the global economy with unmatched depth. Workday sits inside the transactional fabric of the employment relationship for tens of millions of workers. The intelligence here comes not just from the data being proprietary but from the record being comprehensive, granular, and longitudinal enough to function as a source of truth. This is the cleanest and most durable form of intelligence. It’s also the rarest; comprehensive records take decades to build, and starting from zero against an incumbent archive is a losing proposition.

2.      Interpretive Intelligence. The ability to turn raw signals into decisions.

The underlying data may come from various sources, but how the data is used to confidently drive action is the real differentiator. Windward, one of our portfolio companies, is a great example of this. Windward enriches maritime positional data with decades of labeled vessel behavior, sanctions enforcement outcomes, fraud patterns, and risk signatures. The result is mission-grade intelligence that cannot be reproduced with raw data feeds; it takes years of intentional observation and pattern identification to build. With this type of intelligence, the primary data is typically inert, but the fusion and interpretation layer drives action, gets sharper as customer usage increases, and underpins prediction and forecasting. The proprietary assets here are the algorithms that transform data inputs and the derivative dataset that results. I’d argue this is one of the most misunderstood and exciting categories of intelligence right now.

3.      Network Intelligence. Data that exists because the network exists.

Highway, another FTV portfolio company, illustrates this well. As the leading carrier identity and compliance platform in trucking, Highway helps freight brokers prevent fraud. Part of what makes the company special is its interpretive intelligence (i.e., turning fragmented carrier data into real-time fraud decisioning), but Highway also benefits from network effects. By working with over 90 of the top 100 freight brokers in the U.S., it generates behavioral and transactional insights as a byproduct of wide participation. More importantly, as the network operator, only Highway can see patterns that span the full graph (cross-participant fraud, emerging risk signatures, structural shifts in behavior), which no single broker could observe on its own. That dataset cannot be replicated without first replicating the network, and the network has already coalesced. This is different from most datasets because the data is not an external truth the company captured; it is an internal truth the company manufactures, analyzes, and understands. The value is in the company’s positioning as the venue, not the archive.

4.      Codified Intelligence. Encoding expertise into a system that delivers it at scale.

This category is structurally different from the others. The first three start with discrete datapoints and ask what intelligence the data enables. Codified intelligence starts with proprietary expertise, often tacit know-how held by a small group of specialists, and turns it into something replicable. The proprietary asset isn’t a dataset in the traditional sense; it’s the structured representation of how experts actually do the work. Several tech-enabled services businesses in the FTV portfolio have capitalized on this. Lean Solutions Group is one of them. Over the last decade, Lean has built the largest nearshore workforce in logistics, embedded inside hundreds of transportation companies. That front-row view of how back-office work in freight actually gets done and how to make it more efficient is now codified in operational playbooks and AI-enabled tools that let Lean’s teams deliver the work better than customers could deliver it themselves. The intelligence here is rooted in expertise that becomes significantly more powerful when paired with AI.

The fault line in software isn’t whether a company is “AI-native vs. legacy SaaS.” It’s between companies that have something genuinely scarce and companies that don’t. In an agentic world, the model is no longer the moat. On its own, data isn’t either. Proprietary data is still a precursor to defensibility, but differentiation requires turning that data into a form of intelligence, which is more scarce than the substrate itself. A company sitting on decades of data without a complete record, an interpretive layer, a network dynamic, or codified expertise is increasingly just a vendor with an archive. A company with data but no intelligence won’t be able to power the agents that come next. The most intelligent agents will be the ones built on top of intelligent companies.

Going a step further, the companies that will grow their advantage the fastest are the ones stacking two or three forms of intelligence at once. Windward provides another good example of this dynamic. With the recent launch of its Maritime Intelligence Operations Center (MIOC), the company added a codified intelligence layer on top of the interpretive intelligence it already built into the platform. The interpretive intelligence routes the right cases to the MIOC team, and the team’s expert analysis becomes training data that makes the interpretive intelligence more accurate. The two compound because each one’s output is the other’s input.

This raises the question every founder building today has to answer: where does your intelligence actually live? Maybe it’s one or two of these four categories. Maybe it’s something we haven’t even named yet. But the companies that come out on top will be the ones that can answer the question clearly.