Most commentary on AI markets is broad enough to be safe and useless. We think about it differently. The question we ask when evaluating a market is not whether AI matters there — it matters everywhere — but whether the market has a structure that lets an AI-native operator out-compound the incumbents quickly. Some markets do. Most do not. Below are the ones we are most actively underwriting against.
01 — AI agents that operate businesses
The largest software opportunity of the next decade is not a better SaaS app. It is the agent layer underneath SaaS apps — intelligence that operates a business across the tools its owner already uses. Bookings, invoicing, support routing, contract review, vendor selection, hiring filters: all of it becomes the work of an agent rather than the work of a person clicking through a dashboard.
This space rewards companies that have a deep view of how a small business actually runs — because the agent has to translate messy human intent into discrete operations across many fragmented systems. The opening is for operators who have built operating tools for SMBs before and know where the seams are.
02 — Knowledge and intelligence systems for organizations
Every company has institutional knowledge that lives in documents, decks, recordings, Slack threads, and the heads of the five people who have been there longest. The current generation of search and document tools captures almost none of it usefully. The next generation will — and the companies that build the layer between organizational memory and the people doing work today will become the new operating system underneath knowledge work.
The opening here is durable: every organization wants its collective intelligence to compound rather than evaporate. Almost no one currently sells them the tool that does it well.
03 — Education that scales judgment, not just content
Online education solved the content problem fifteen years ago. It did not solve the judgment problem — the part that a private tutor or a senior practitioner used to provide, tailored to the specific person learning. AI changes that ratio. A learner can now have access to private-tutor-grade feedback at scale, in language they understand, on schedule they can actually keep.
The companies that win this space will not look like content libraries. They will look like coaches. The underlying business model is durable — people will pay to become more capable, especially when the alternative is falling behind in a labor market that is rewriting itself.
04 — Productivity software for the small operator
Most modern productivity software is built for enterprises and adapted, badly, for individuals and small teams. The unit economics of AI flip that assumption: a tool that costs almost nothing to operate per user can profitably serve a market that has historically been unreachable. Solo founders running real companies, two-person agencies, family operators, professional services firms — these are markets that the SaaS era under-served and that AI-native tooling is suddenly economic to serve.
This is one of the spaces where two and a half decades of building for entrepreneurs gives us the strongest information advantage. We know what they actually need — and what they will quietly never use, no matter how much marketing it gets.
05 — Future of work: the operating layer for hybrid teams
Most software for distributed teams was bolted onto tools built for the office. The next generation of work tooling will be designed natively for teams that mix humans and agents, run across time zones, and operate against asynchronous decisions as the default. The opening is to define what an operating system for that kind of team looks like — not a chat tool with AI features, but a coordination substrate that assumes intelligence is part of the team.
This category has been over-funded and under-built. The version that wins will look very different from any current incumbent.
06 — Personal-finance and financial-planning tools that actually advise
For decades, the gap between the financial advice rich people get and the advice everyone else gets has been enormous. AI collapses it. A working professional can now have access to planning intelligence — tax optimization, asset allocation, debt structuring, college funding, retirement modeling — that previously required a $300/hour conversation. The companies that meet that demand at the right price will compound for decades.
The opening here requires regulatory care and tasteful product design. It rewards builders who treat the customer as an agent making real decisions, not a target to be monetized.
07 — Marketplaces that connect AI-augmented humans
The supply side of marketplaces is being rewritten by AI: a freelancer with the right tooling now produces what a team used to produce. The marketplaces that organize that supply — and translate it into trustable outcomes for buyers — become more valuable as the underlying labor shifts. Software development, design, copywriting, analytics, legal review, accounting, customer support: all of these markets have an opening for the marketplace that guarantees outcome rather than billing time.
What we are deliberately less interested in
We are not actively pursuing horizontal foundation-model companies, hardware-heavy AI plays, or consumer-app chatbot wrappers. The first two are not where our edge lies; the third is mostly noise that will compress hard inside three years. We are also cautious about AI-for-developers tooling — a category that is already well-funded and where the moats are narrow.
The common thread
Every category above shares the same shape. There is a durable customer with a real budget, a workflow that AI can rewrite end to end, an opening for a small operating team to ship a sharper product than the incumbents, and a curve that compounds for many years rather than peaking at year three. Those four conditions matter more than any surface-level AI signal.
When a founder pitches us a company that meets them, we pay attention.