Defensibility in the Age of AI
May 2026
Alline Akintore & Eric Rosenfeld
In March this year, public software did something it had not done in more than 2 decades: it began trading at a discount to the S&P 500. About $2 trillion in SaaS market cap has evaporated since the fall and the median revenue multiple for public SaaS hit 3.4x, the lowest since the metric has been tracked.
One of the reasons for this repricing has been the thesis around seat compression; if an AI agent can do the work of five people, enterprise customers might buy 100 licenses instead of 500. The per seat pricing model that made SaaS the most valuable business model of the last decade is evolving. As early-stage investors our deal flow sits directly downstream of this repricing, which is why we spend a lot of time thinking about defensibility with this AI platform shift.
Furthermore, in 2025 we were talking about eval suites as one of the moats for the wave of agentic AI, but that has changed too – in the last six months foundational models have commoditized the eval layer especially for general tasks.
At OVF we spend time looking for patterns where defensibility is structurally durable. Other than IP-intensive players, the application companies still building real moats in 2026 aren't winning on product features but rather on assets that AI can't generate such as regulated market access, embedded distribution, and operational depth that takes years to accumulate.
Mission-critical depth where mistakes cost real money:The gap between a working demo and a reliable production system in regulated, high-consequence domains is an order of magnitude or more of accumulated effort, and that gap is the moat. Enterprises are unlikely to adopt AI agents with generic AI tooling and evals are only artifacts to support depth of understanding of the workflow including the edge cases. For example, Eclypsium protects firmware and embedded software (the layer beneath the OS), where supply-chain attacks and persistent threats hide. AI helps the product detect anomalies faster, but the value sits in years of accumulated edge cases, vendor relationships, and technical and operational expertise across enterprise environments. There are other examples in clinical-grade healthcare, critical infrastructure (e.g. Customer.io), security operations (such as AiStrike, C1.ai), and finance. Hamilton Helmer refers to this as process power in 7 Powers.
Distribution that AI can't replicate: When product is cheap, channels are scarce. Hydrolix runs natively on Akamai's edge – a structural distribution position that compounds with every customer Akamai brings. Another portfolio company, Platformr, holds a similar position with AWS. These aren't marketing partnerships but embedded relationships where the partner's success depends on the startup's success.
Cornered regulatory access:Kalshi operates as a CFTC-designated contract market, a federally regulated prediction market. That status took years of compliance work, legal infrastructure, and regulatory relationships that no AI agent can shortcut. A competitor with a perfect product clone has nothing to sell until they secure the same designation. This is the pattern wherever the right to operate is the scarce asset: regulated finance, defense, clinical trials, controlled substances, healthcare reimbursement. AI commoditizes the application; it does not commoditize the license.
One objection to our posture might be, “aren't these moats just lag? Within 24 months, won't AI commoditize say, regulated workflows too?” Our answer: maybe, but unlikely. If one is building to solve a real use case, our thesis is that compounding defensibility (as listed above) outside the software stack is where true competitive advantage lies.
We continue to be bullish on how AI is shaping industries and we look to partner with founders building with a thesis on where their durable moat lies. If that's you, our colleague Deepthi Madhava would be glad to meet — Deepthi@OregonVentureFund.com.