The Off-the-Roadmap Vertical Bet
Back domain insiders automating workflows the frontier labs will never prioritise
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 6
- Confidence
- 86%
Gurley's selection rule inverts the instinct to chase the hottest part of the stack. Inputs are three filters applied in sequence. First, distance from the frontier: you can read publicly what the labs are working on next, and anything on that list offers no protection. Second, founder-market fit of a specific shape: someone who brings genuine perspective from a particular industry and who is simultaneously among the smartest users of AI in that vertical. Third, defensibility from workflows and local data rather than from model quality. His argument is that even when a model can understand a subject matter, there are workflows and datasets local to each customer that must be stitched together, and that stitching is what a question-answering model cannot replace. He pairs this with a hard constraint on the current market: institutional investors have essentially zero interest in non-AI deals, so an unrelated angel investment risks dying of neglect at the next round.
Origin
Extracted from The Tim Ferriss Show. Gurley describes how he would approach angel investing after winding down institutional venture capital at Benchmark, informed by his board seat at Zillow and five years watching it build workflow tooling for realtors.
Core principles
- 01Proximity to the frontier labs' roadmap is a liability, not an advantage.
- 02The defensible edge is domain workflow and local data, not model capability.
- 03The best founders sit at the intersection of deep industry experience and heavy AI use.
- 04A model that only answers questions is not a substitute for a stitched-together system.
- 05Capital scale has closed the frontier-model lane to small investors.
How to run it
- 1
Read the frontier roadmap
Go online and read or watch what people at the major labs say they are building next. This is public and cheap to gather.
- 2
Rule out anything near that edge
If the idea is the next thing a frontier lab intends to do, assume you are not protected. Deliberately stay far from that edge.
Pro tip Off the beaten path is the goal, not a consolation prize.
Watch out You also cannot fund a frontier model company at angel scale; that requires roughly a billion dollars.
- 3
Pick an unglamorous vertical
Choose a deep vertical it would not make sense for a frontier lab to crush. Gurley's throwaway example is waste management; the vertical's obscurity is the moat's raw material.
- 4
Screen the founder on both axes
Require the intersection: authentic domain perspective from the industry, plus already being one of the most sophisticated AI users in that genre. Either alone is insufficient.
Pro tip Curiosity about the tools is the tell; the best candidates are already playing constantly.
- 5
Verify proprietary data and local workflows
Confirm there are datasets local to the customer and multi-step processes that must be stitched together. That integration work is what a general model cannot deliver on its own.
Pro tip Zillow's Showing Time, for booking in-person house tours, is a clean model of what a workflow tool looks like.
- 6
Check the next-round reality
Confirm the deal is AI-related enough to raise institutional money later. Gurley is emphatic that institutional interest in non-AI deals is currently zero, so unrelated companies can die of neglect.
Watch out This condition is a snapshot of the current market and could reverse if the cycle turns.
In the wild
Gurley, a Zillow board member, uses it to define what a workflow actually is. Zillow has spent roughly five years building tools that help realtors do their day-to-day job: Showing Time books in-person tours at houses, and beyond that sit mortgage assembly, sign-offs, and a long tail of tasks that have to happen and can be automated. Each of those tasks is anchored in customer-local data and process. The more of that a company builds into a system, the better protected it is from a model that merely answers questions.
→ Workflow depth, not model access, is what makes the position defensible against frontier labs.
At a Montana fishing lodge, Gurley met a twenty-eight-year-old entrepreneur from near Corpus Christi who had already started three or four businesses. He was completely absorbed in AI, describing how each time he needed something the tools delivered, including asking where in a city to site his next location and getting immediate answers. He was already successful but was now running at triple speed because he had leaned into the tools with an open mind about what they could solve.
→ The profile Gurley wants to back: existing domain operating experience plus unusually aggressive adoption of the new tool set.
Common mistakes
Betting adjacent to the frontier roadmap
If the idea appears on a major lab's stated next-steps list, the work gets replicated in short order and no amount of execution speed protects the position.
Domain expertise without AI fluency
Industry knowledge alone produces a company that will be out-executed by a rival who is both an insider and the vertical's heaviest AI user. The intersection is the whole thesis.
Ignoring the follow-on funding climate
Angel-funding a strong non-AI company right now risks it dying of neglect, because institutional investors currently show essentially no interest in non-AI deals.
Is it for you?
Best for
Angel investors and founders choosing where to place an AI-era bet with limited capital.
Not ideal for
Anyone with the billion-dollar scale required to compete in frontier models, or verticals with no meaningful workflow or proprietary data layer.
From the transcript
“I'd try and find an intersection of people that are super curious and are playing with all these AI tools, but bring a perspective from…”
“The key is just to stay pretty far away from the edge of whatever.”
“there are workflows, there are data sets that are local to your customer and that stuff has to be stitched together”
From the episode
#840: Bill Gurley — Investing in The AI Era, 10 Days in China, and Important Life Lessons from Bob Dylan, Jerry Seinfeld, MrBeast, and More