The 8 top take-aways from Jacob Effron appearance on Nick Moran’s Full Ratchet podcast.
1. Great wedges in AI —> The strongest wedges are products where AI creates a step-change in the user experience. In verticals like healthcare and legal, the teams with real traction are the ones solving high-ROI use cases that end users actually love, not ideas pushed by “innovation teams.” The best wedges also create room to expand into a broad surface area of product over time.
2. Evolved thinking around “owning the data” —> Early on, we expected defensibility to come from proprietary data (own the data, own the model). What’s become clear is that you often don’t need massive datasets to apply RL effectively; data alone rarely creates a moat. Understanding evals, building great workflows, and iterating quickly matter far more than sitting on unique data.
3. Where we’re seeing the most traction —> Coding, customer support, healthcare, legal, logistics, and voice interfaces are showing real product-market fit and meaningful revenue scale. These are categories where AI already delivers material improvements in accuracy, speed, or cost. Other areas, like go-to-market, likely need another step-change in model quality before the value fully lands.
4. Why healthcare and legal are great markets —> Because quality truly matters—no one wants the “80% right but cheaper” hospital or law firm. That dynamic pushes the market toward the highest-performing systems, not the lowest bidder. Both sectors also face deep access challenges, which creates room for AI to expand capacity rather than simply replace existing spend. S/o Abridge and Legora
5. Reaction to “T2D3 is dead” —> Hyper-growth benchmarks aren’t dead, but AI has reset expectations. Some AI companies are growing so fast that the historical T2D3 profile now signals “good,” not “category-defining.” But businesses growing at traditional T2D3 levels can still be strong; they just may not represent the deepest product-market fit in AI.
6. Domain experts versus young, cracked teams —> Both archetypes win. The scrappy, technical founders have an easier time getting into rooms that used to be accessible only to deep domain experts, but domain-native CEOs still have an advantage when it comes to understanding second and third product acts.
7. What’s overrated in AI —> The idea that software disappears and everyone simply generates UIs on the fly is highly overestimated. Building software is hard—but maintaining it across teams, standards, and organizations is even harder. Incumbent software models will evolve, but the notion of a “post-software world” is unrealistic.
8. Biology is underrated —> The most interesting breakthroughs are coming from adjacent model types that rely on new kinds of data, not just the LLM progress everyone focuses on. He sees real research advances and unprecedented capital flowing into these fields, which together could produce significant model breakthroughs over the next 5–10 years.
