The Three-Segment AI Market Map
Frontier models, tooling, or applied agents — pick the layer that fits your capital and your edge.
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 3
- Confidence
- 90%
A decision map for where founders should build in AI. The market splits into three layers with very different economics: frontier/foundation models (capex-dominated, already consolidated), tooling/'pickaxes' (real markets but dangerously close to the incumbents), and applied AI agents (higher-margin, product-led, the SaaS-analog opportunity). Taylor's guidance is to avoid the first, respect the risk in the second, and concentrate on the third.
Origin
Bret Taylor's framework as CEO of Sierra and OpenAI board chairman for how the AI market will play out and where startups should focus.
Core principles
- 01'Agent is the new app' — the applied layer is where the SaaS-style value accrues
- 02Frontier models are a capex game that has already consolidated
- 03Tooling lives 'close to the sun' — adjacent to infrastructure that can absorb it
- 04Applied agents are higher-margin because they sell a business outcome, not a model byproduct
- 05Winning in the applied layer requires deep domain understanding, not model technology
How to run it
- 1
Rule out frontier models unless you can raise billions
Building a frontier model is entirely a function of capex, and the asset deteriorates in value quickly. Startups that tried (Inflection, Adept, Character) have been consolidated. Don't build one unless you have Elon-tier capital access.
Watch out There is no viable startup business model here; fundraising runway won't reach escape velocity against depreciating model assets.
- 2
Enter tooling with eyes open
Data labeling, eval tools, data platforms, specialized models (e.g., voice) are real 'pickaxe' markets. But ask not if but when a large infrastructure provider ships your exact product on a developer day — and why customers will still choose you.
Pro tip Look at how Snowflake/Databricks/Confluent survived adjacent to AWS/Azure — some thrive, many get obviated.
Watch out Tooling is 'close to the sun'; infrastructure providers differentiate by moving up the stack right into your space.
- 3
Concentrate on applied AI agents
Build agents that achieve a measurable business outcome for a specific domain — customer service, legal, supply chain, content marketing. Over time these look like SaaS: buyers ask about workflows and outcomes, not which model or database you use.
Pro tip The long tail is the opportunity — boring, resource-heavy jobs an agent can solve, won by a founder who understands that business deeply.
Watch out Applied agents pay a 'tax' down to model providers, so margins are high but not infinite.
In the wild
Taylor looked at the top 50 public software companies: the top five are the giants, but the next tier are all SaaS companies, some exciting, some 'super boring.' He argues agents will evolve the same way — beyond the obvious huge markets into a long tail of domain-specific agent companies.
→ A prediction that applied-agent entrepreneurship, led by deep domain experts, is where most AI value will be unlocked.
Common mistakes
Building a frontier model as a startup
The capex requirement and rapid depreciation of the model asset mean startups can't reach a return; the segment has already consolidated around hyperscalers and a few labs.
Building tooling without a durable answer to 'why you'
Because tooling is adjacent to infrastructure, a foundation-model provider can ship your exact feature; without a reason customers stay, your market evaporates on a single developer-day announcement.
Is it for you?
Best for
AI founders and investors deciding which layer of the stack to build in
Not ideal for
Capital-rich players (hyperscalers, Elon-tier) for whom frontier-model economics actually work
From the transcript
“there's three segments of the AI market that will end up fairly meaningful markets”
“creating a frontier model (53:00) is entirely a function of capex”
“those companies probably are the most at risk for you know a developer day from one of these big foundation model companies releasing exactly what…”
“Uh, I think agent is the new app”
From the episode
He saved OpenAI, invented the “Like” button, and built Google Maps: Bret Taylor on the future of careers, coding, agents, and more