The AI-native companies we want to back

Where we are in the cycle

The AI infrastructure buildout is largely done.

We’re entering what Carlota Perez would call the deployment phase, the point after a technology’s installation period where value shifts from building the foundational technology itself, to the businesses that run on top of it. The internet followed a similar arc in the 90s: a period of fibre and broadband buildout, followed by the creation of companies such as Amazon and Google. None of these giants reinvented the wiring underneath them, however they turned out to be some of the most significant and defensible businesses in history.

The frontier models (e.g. OpenAI, Anthropic) have spent the last few years building this foundational layer of technology for the AI era. That phase is largely behind us now.

Where Tall Order Is Investing

Tall Order’s third fund is built around what comes next. We will invest primarily in the companies built on top of this foundational layer, and that solve problems previously out of reach for software until now.

We’re still early in that shift. As AI moves deeper into different sectors, customers are moving from experimental usage to the commitment of incorporating it into their budgets. That’s where we think the next several years of value creation get unlocked: market by market.

The shift in AI capability changes broader economics too.

If an AI product delivers a quantifiable business outcome, it gets priced against something bigger than the software cost it displaces. It’s also the value of the time it frees up for the person who previously had to do that work by hand.

Take an aged care family liaison coordinator. The record-keeping software they use might cost a few hundred dollars a year. However, most of what fills their day: logging medication schedules, following up with families for updated paperwork, fielding the same questions over the phone, isn’t why they were actually hired. This manual work keeps them from the parts of the job only a person can do: spending time with a resident who’s had a rough week, walking them through their care plan, or knowing when a family needs a phone call instead of an email.

The AI product that absorbs this work is not competing with a few hundred dollars of admin. It’s being measured against what that coordinator’s reclaimed time is worth. That’s more like a fraction of a salary instead of just a subscription fee. Charge a reasonable share of the value that time creates, and in that scenario, $20,000 a year for something that used to sell for a few hundred isn’t really a stretch. It’s a fundamentally different category of business, and why markets that previously considered too small for venture are worth a second look.

Why The Old Rules Don’t Apply

Opportunity aside, this is a trickier landscape to invest well in. A product can look unbeatable today and lose its edge in a matter of months. ARR growth numbers that would’ve had investors racing to sign a term sheet now warrant a higher level of scrutiny.

It’s an extraordinary time to be investing, but the way we used to invest pre-AI doesn’t fully translate into this new environment.

Competition has changed

The elephant in the room is that the competitive landscape has shifted.

First, the frontier models themselves. Anthropic and OpenAI don’t need to be explicitly pursuing a startup’s market for this to be a problem. They just need to be good enough (and embedded enough) in someone’s day, that the customer decides what they’re already using is close enough. Canva spent 13 years building a moat, but that didn’t stop Claude Design from getting into their customers heads.

You also have other competitors in the mix. The first wave of well-funded AI-native players, building at a pace that used to take a team of 50 engineers. Then you have the deeply entrenched incumbents who’ve had to create or adapt their AI strategy (hello, acquisitions!). Their UX might be dated (sorry Marc Benioff), but they’re sitting on decades of proprietary customer data which counts for something.

This makes us particularly interested in net-new market categories. Those where there was never a serious incumbent tool to begin with, or where incumbents have hit a natural ceiling. Products that can meet customers where they’re at now (not where they were over a decade ago). We think these sectors are hugely underrated, and potentially in industries that nobody is bothering to talk about.

Moats have changed

For a long time, one of the safest bets in software was integration complexity. Once a customer’s data became embedded enough in your system, it got too hard to rip them out. Getting that data out meant painful exports and months of pain, so everyone just stayed put. Fast forward today and AI migration tools with the right permissions can automate exactly that kind of data migration.

The same goes for interface quality. Companies could’ve had best in class UI/UX. Customers use plenty of products purely out of habit, but if those product’s main point of difference today is a nicer interface, that’s quite a fragile moat. AI-generated interfaces can now be achieved in a fraction of the time, and if agents are increasingly the ones making decisions rather than a human clicking through multiple screens, interface quality might actually not matter as much as a founder thinks it does.

Data moats purely built on volume isn’t the safe fallback it used to be either. Having more data than anyone else used to be a serious advantage. It isn’t anymore, and the real question is whether that data gets smarter and more useful the longer it’s used.

How Tall Order Evaluates AI-Native Companies

We’ve had to spend time thinking through how we will identify what is truly defensible and can be enduring, now that the frameworks of the traditional software era no longer apply.

Given all of that, here’s a quick overview of the framework we run through for every AI opportunity.

People

We believe domain experience matters even more in this AI era than the one before it. The founders who know a workflow deeply enough can tell us exactly which parts: (1) are ready to be handed to a model, or (2) still require human judgement. These founders can walk us through the entire procurement for their category: who initiates the evaluation, who influences it, who holds the budget (or can quietly block a sale), and what change management needs to happen internally before an AI product gets fully adopted and trusted to do the work. They usually have relationships and credibility with the buyer, which matters more if they are being trusted by the customer to hand over specific judgement calls in their process. A litmus test is if we could picture someone better positioned than this founding team to solve this exact problem in this specific industry.

Strategy (starting with market selection)

The companies we get most excited about have deliberately chosen to be different, not just marginally better. Are they going after a market that’s too small, specific or weird for a frontier model or well-funded player to bother contesting? Horizontal applications are the ones most exposed to displacement by the labs building models underneath them, but a solution for managing complex geotechnical data (hi Tablogs!) is a very different proposition to another AI project management tool. We also want to see if there is a material barrier to entry (i.e. regulatory, technical innovation, compliance-based) that keeps other entrants out, and if the founder has a credible plan to secure them before anyone else does.

Defensibility

This is one of the most consequential dimensions for AI-native companies. We believe that the ones who will endure are the ones that get harder to compete with over time.

Primary Moats: What we look for from Day 1

These are examples of moats founders can start building from day one and are central to our investment decision.

Moat	What We Evaluate
Codified Innovation	Does the product embody a technical breakthrough built on years of R&D that a competitor cannot quickly replicate? Is the underlying science or engineering problem genuinely hard, not just a clever application of an existing idea?
Data intelligence	What will the product understand about a specific customer’s environment after 1,000 interactions that a new entrant with access to the same raw data would not? How does that understanding translate into measurably better product outputs and outcomes for that customer/user over time?
Workflow depth	How deeply is the product embedded into the customers’ day-to-day operations? Does the product sit within a workflow where failure carries material financial, compliance, or operational consequences? What is the “tax” or “level of disruption” involved in switching away from this system? Do other systems used increasingly depend on this product to better understand and act on customer operations, or is it just one of several interchangeable data sources?
Flywheel potential	Does data or feedback by one customer measurably improve the product for other customers/users over time? Does that advantage accelerate over time? (i.e. does customer #1000 make the product meaningfully harder to replicate than customer #100?)

Structural Moats: What matters with scale

These show up as companies scale over time. They are not necessarily seed-stage decision criteria, however, founders should be able to articulate how they fit into the company's long-term competitive strategy.

Moat	What We Ask
Distribution reach	Has the founder identified a distribution channel native to their domain? Will each new customer acquired through that channel make the next acquisition easier? Can the company achieve the volume and scale required where it is significantly more difficult for others to reach it?
Ecosystem leverage	Is the architecture designed to become a platform that other third parties can build on and depend on? Does it create value for parties beyond the direct customer?
Brand trust	How much does buyer risk tolerance and trust play a role in procurement success? Does the market entry strategy prioritise credibility and domain authority from the outset?
Scale economics	Does the business model contain unit cost dynamics that improve structurally with volume? Can competitors at a smaller scale match those economics over time?

Where that leaves us

This isn’t a checklist every early-stage AI-native company needs to tick every box on. Building domain experience, choosing the right ground to compete on, and engineering moats can take years. We don’t expect any team to have this all locked down on day one when we invest.

The founders that make us lean forward can share a coherent plan for how they’ll build one or more of these moats over time and have clearly thought about what happens when they scale.

So if you’re building these moats around your castle, I’d love to hear from you.