Defensible Moats for AI Startups
Four durable places to build in AI where foundation-model labs are least likely to squash you.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 90%
Krieger's answer to 'where can AI founders play without getting crushed by OpenAI or Anthropic' is a four-part decision framework for durable positioning (on a one-to-three-year horizon). The moats are: deep vertical market knowledge, differentiated go-to-market rooted in customer relationships, a novel form factor that incumbents can't easily adopt, and raw startup intensity. Labs supply general models and coding/agentic strength; they generally won't build the intense, domain-specific solution for your niche.
Origin
Mike Krieger, drawing on founder-day conversations (YC, Menlo Ventures, Norwood) and examples like Harvey (legal) and biotech, answering Lenny Rachitsky's question about where AI founders should build.
Core principles
- 01Foundation-model labs won't build the deep, intense solution for every vertical
- 02Durability here is a 1–3 year bet, not a guaranteed 5–10 year moat
- 03Knowing the specific buyer beats knowing the company
- 04Novel form factors are hard for incumbents to adopt because users have fixed assumptions
- 05Startup existential intensity is a real, non-replicable advantage
How to run it
- 1
Own a differentiated vertical market
Build deep, specific knowledge of how a particular industry actually works — legal, biotech, healthcare, insurance — encoding real workflows a generalist would never invent from scratch.
Pro tip Harvey's UI encodes a specific flow lawyers do that you 'never would have come up with from scratch.'
Watch out Domains that 'don't sound sexy' (compliance, healthcare) are often exactly where large durable companies get built.
- 2
Build a differentiated go-to-market
Cultivate relationships that let you know not just the company you're selling to but the specific person — CIO vs CTO vs CFO vs general counsel vs the engineering team choosing an LLM.
Pro tip Start that first buyer conversation early even if deep empathy is hard to build in a 3-month accelerator; or co-found with someone from that world.
Watch out You can't fake buyer empathy — a generic 'sell to the company' motion loses to someone who knows the individual decision-maker.
- 3
Invent a novel form factor
Take a genuinely different bet on how people interface with AI — something that looks weird or power-user-only today but could become huge as models improve.
Pro tip Incumbents struggle to adopt new form factors because their users already have fixed assumptions about how the product works.
Watch out These bets look 'out there' at first and won't have broad appeal on day one — that's the point.
- 4
Harness startup intensity
Operate with the existential, 'you against the world' urgency of a tiny team that must win — a mode of working, not an area of building.
Pro tip Krieger notes this is what he had at both Instagram (2 people) and Artifact (6 people).
Watch out You can't instill this with OKRs — it has to be felt, and it evaporates if you stop treating the problem as existential.
In the wild
Krieger spent time with the Harvey team, who showed him UI encoding a very specific flow that lawyers actually do — not something a generalist would design from scratch, but the way the work truly gets done.
→ Illustrated vertical market knowledge as a moat: labs supply the model, but the domain-specific application is where a startup stays defensible.
Common mistakes
Building a thin wrapper any lab could replicate
If your product is just an off-the-shelf model with light prompting, a foundation-model company can absorb it — you need real vertical depth, buyer relationships, or a novel form factor.
Only knowing the company, not the buyer
Selling generically to 'the company' misses that different individuals (CTO, CFO, general counsel) make the decision; not knowing the specific buyer forfeits the go-to-market moat.
Is it for you?
Best for
Founders and operators starting AI companies who worry about being squashed by OpenAI, Anthropic, or Google.
Not ideal for
Teams seeking a guaranteed decade-long moat — Krieger explicitly caps these as 1–3 year durable, not permanent.
From the transcript
“One is understanding of a particular market”
“paired with that is like um differentiated go to market which is the relationship that you have with those companies”
“don't underestimate how much you can think and work like a startup and feel like it's you against the world”
“I get excited about startups that will get started that have like a completely different take on what the form factor is by which we…”
From the episode
Anthropic’s CPO on what comes next
Mike Krieger (co-founder of Instagram)