Apr 23, 2026

O Moat, Where Art Thou?

How does one build a moat in the AI startup space? Or as VCs like to say: “Where does the value accrue?”

Here are some observations. I can’t quite thread them into a coherent narrative. The crystal ball is cloudy, and also the clouds are moving very fast.

• every AI startup wants to be a pure product, almost SaaS-like company. Almost all of them end up becoming bespoke solution builders, aka consultants. Nobody likes being called that, so they say “forward deployed engineer”. You want to build agents for companies, and every company is a snowflake.
• Palantir looked like a consultancy for a decade before it dawned on everyone that they had a durable product behind it. The startups are hoping to reproduce that playbook: structure the engagements to harvest transferable artifacts rather than bespoke deliverables.
• The major model vendors are moving aggressively into verticals. See recent product pushes such as “Claude for finance/law/healthcare”. So much for the 2023-25 thesis of finding shelter in vertical specialization.
• The major model vendors are aggressively hiring FDEs. FDEs are subsidized by the token business. They don’t have to pay for themselves — they exist to expand usage. A startup’s FDEs have to be explicitly funded. The vertical startup is competing with a cross-subsidy they can’t match.
• Capability is not access. Can the major model vendors really shmooze enterprise sales? Or: rizz is the only moat.
• Speed of erosion matters. If a startup can carve a niche that lasts 6 months before the giants catch up, maybe it can convert that into a trusted relationship that persists. In other words, move fast enough to turn a perishable asset into a durable one.
• maybe the very framing of “having a moat” is incorrect. When the ground under your feet shifts so fast, you probably want some notion of “velocity + customer intimacy”.