Where AI is the architecture, not the feature.

The definition of a "thin wrapper" keeps moving. A year ago, building memory-enabled agentic workflows required real engineering talent and architectural decisions that took months. Today, a generalist with the right tools can replicate that in a weekend. Persistence and workflows used to be the moat. They're table stakes now.

We focus on AI-native applications where the moat compounds as models improve rather than eroding. That means companies capturing unwritten domain rules — the edge cases, exceptions, and judgment calls that never make it into documentation. It means integrating data in ways only insiders would know to combine: connecting contracts to case precedents, billing history, and communication patterns across systems. It means building memory architectures where context accumulates and every interaction makes the next one smarter.

The best AI-native products don't look like traditional software with a chat window bolted on. They rethink the interface itself — matching input modality to friction and output complexity to the task. The user's role shifts from clicking through workflows to supervising autonomous processes and making high-leverage decisions.

Our litmus test: Could a generalist builder replicate this in a weekend? If yes, the moat isn't real. We back founders where application specificity, proprietary data loops, and vertical expertise create advantages that widen over time — not ones that get commoditized with the next model release.

Ardent has been investing in financial services since the inception of the fund — Embedded Financial Services was one of our deepest thesis areas in Fund I. That early conviction gave us pattern recognition that we're now applying to a new generation of companies.

What's changed: AI is unlocking parts of financial services that software alone couldn't reach. Real-time consumer debt data. Clean energy project finance. Payments infrastructure for vertical software platforms. These aren't incremental improvements — they're new products serving markets that didn't have good solutions.

We've developed a specific edge here: regulatory complexity doesn't scare us. Our DC presence and network means we understand how financial regulation works, where the moats are, and which founders are building to last vs. building around the rules. Compliance-heavy markets are features, not bugs — they create durable barriers.

We look for founders rebuilding financial infrastructure at the layer that matters: the data, the API, the rails. Not the consumer interface on top, but the foundational plumbing that everyone else will build on.

For most of business history, a service required a person. Advisory, compliance, operations, ad management, customer support — these were labor-intensive by nature, not by choice. AI is changing that equation fundamentally.

Service-as-a-Software companies don't automate existing workflows. They rebuild the delivery model entirely: taking something that looked like a services business and making it behave like software — faster, cheaper, more scalable, with gross margins that compound over time.

The founders we back in this space think in systems. They understand the full workflow they're replacing — not just the obvious steps, but the edge cases, the institutional knowledge, the human judgment calls. And they build software that becomes as embedded and irreplaceable as the infrastructure it displaces.

These companies often reach revenue faster and require less capital than traditional SaaS. The trade-off is that the product judgment required is higher — you have to understand the service deeply before you can replace it.

Every major domain has had its software moment. The physical world is having its now.

General robotics foundation models have crossed a threshold. They're capable enough to make vertical specialization tractable, but not capable enough to make it unnecessary. Our conviction is that no single model will dominate all of robotics. The physical world is too varied, too demanding, and too unforgiving. These models provide a powerful baseline but what every production environment needs is a solution that closes the gap between baseline and reliable. The data, physics, and error tolerances of a precision manufacturing line are a completely different problem from a surgical suite, a logistics center, or a farm.

We believe this gap is structural. In robotics, deployment is data collection.

General manipulation skills and spatial reasoning transfer broadly across robots and environments. That is what foundation models are good at. Physical data is inherently local at the level that matters most for production deployment. The specific geometries, failure modes, and precision requirements of a given physical environment do not transfer. That last layer of specialization is where the vertical data flywheel lives, and where it compounds with every deployment. Specialization takes different forms. Reinforcement learning from real-world experience. Environment-specific geometric and physics modeling. Synthetic data generation. Approaches we haven't seen yet. We don't prescribe the method. We look for founders who have found the right one for their domain and are building the deployment base to make it defensible.

We invest across two tiers. The first is AI-native software building domain-specific robotic intelligence, systems that get measurably better with every deployment and build moats that are hard to replicate from the outside. The second is the infrastructure that makes that possible: synthetic training data generation, physical environment reconstruction, and the tools that reduce the cost of real-world data collection that every vertical company depends on. We think both tiers produce large, durable companies.

The founders we back are equal parts deep scientist and determined builder. Robotics software is not a domain where generalist engineers pick a vertical and figure it out. The hard problems demand genuine expertise in the underlying science. Founders who have spent years inside a specific physical problem and know not just how to build the software, but why the physics makes it hard.

AI is reshaping how people discover, communicate, transact, and make decisions. Most of the attention has gone to enterprise, but we believe some of the most important companies of this era will be built for consumers.

We opportunistically back founders with deep founder-market fit who are solving real problems around how people spend their time and money. We’re actively pursuing opportunities across marketplaces, social, and financial services.