How AI Founders Pick a Winning Vertical, Not a Feature
For AI startup founders · Based on Greg Isenberg Startup Opportunity Scanner
// TL;DR
AI startup founders can use the Greg Isenberg Startup Opportunity Scanner to avoid the two biggest traps: bolting AI onto an existing UX, and building horizontal 'AI employee platforms.' Instead, go agent-first with an Action App where agents do work on the user's behalf, and verticalize on one job title. Build a Jobs-to-be-Done Stack of 30–50 tasks as your roadmap, position as 'Juniors' handling menial work, and validate the niche with the three-question test before building. The niche is the marriage; the model is the experiment.
Why is my horizontal AI product struggling?
Because horizontal loses to vertical. The Greg Isenberg Startup Opportunity Scanner is blunt: 'AI employee platform' competes with everyone and speaks to no one. Verticalization is the moat. Build an 'AI junior YouTube producer,' not an 'AI employee.' Build for one job title so precisely that the buyer feels the product was made for them.
The scanner also names two pitfalls AI founders fall into constantly: bolting AI on as a feature instead of going agent-first, and producing AI slop instead of top-1% AI-native quality. Both come from optimizing the wrong thing — features and volume instead of a specific niche served deeply.
What makes an Action App different from an AI feature?
An Action App is agent-first. The core UX is agents doing things on the user's behalf — clearing inboxes, booking calendars, filing expenses — not a screen of buttons for the human to click. Bolting AI onto an existing app keeps the old click-and-stare interface and adds a chatbot. That's the pitfall.
Reimagine the category instead. The dashboard shouldn't be a list of tasks for the human; it should show what the agents already did and surface only the 2–3 decisions requiring human judgment. Don't fear that full automation feels low-value — mistaking set-and-forget for low engagement is a listed trap. Humans always seek the path of least resistance, so convenience to the max is the goal.
How do I decide what my AI actually does?
Build a Jobs-to-be-Done Stack. Pick a specific persona — a junior podcast editor, a junior YouTube producer — and ask Claude to list all 30–50 tasks they perform: chapter generation, clip selection, thumbnail briefing, show notes, guest research, transcript cleanup. That list is your product roadmap.
Build agents to cover 2–3 jobs first, then expand to 10, then 50, until you've built a true digital employee. Position the product as 'Juniors' — replacing junior-level, repetitive, non-creative work, not senior talent. This lowers threat perception, makes the sale easier, and honestly matches what AI can do today. 'We handle the menial work at a fraction of a junior hire's cost' beats 'we replace your team.'
How do I validate the niche before building?
Run the three-question Niche Qualification Test on your target persona: are they underserved, do they have willingness to spend, and is the pain sharp enough that they already pay to patch it? Then apply 'Fish Where the Fish Are' — underserved verticals often beat crowded, trendy ones.
Stress-test with 'Date the Product, Marry the Niche.' If your first agent product fails, could you pivot to a different AI product for the same persona? If yes, the niche is sound and you iterate. If the whole business only works with this one model, reconsider whether you've actually committed to a niche or just to a clever technical demo.
How do I get my first users?
Use an acquisition wedge that dramatizes the pain. For an Action App, reverse-engineer a TikTok skit: a founder buried in busywork versus the same founder with five minutes of oversight. If you're building AI-native media around your product, study a successful AI-native creator in an adjacent niche and replicate the format — betting on top-1% quality with a human in the loop, not volume-driven slop.
What's my next step?
Pick one job title in an underserved vertical. Run the three-question Niche Qualification Test, then ask Claude for its full Jobs-to-be-Done Stack. Ship an agent-first Action App covering the top 2–3 jobs, position it as 'Juniors,' and dramatize the pain in a launch skit. Commit to the niche — let the model evolve.
// FREQUENTLY ASKED QUESTIONS
Should I position my AI product as replacing senior or junior workers?
Junior. The 'Juniors' positioning focuses on replacing junior-level, repetitive, non-creative work. It lowers threat perception, makes the sale easier, and honestly matches current AI capability. Claiming to replace senior talent triggers skepticism and resistance. 'We handle the menial work at a fraction of a junior hire's cost' is a far more credible and buyable pitch.
Isn't a fully automated app going to feel low-value to users?
No — that's the mistaking-set-and-forget-for-low-engagement trap. Humans always seek the path of least resistance, so maximum convenience is the goal, not a weakness. An Action App that quietly does the work and only surfaces the 2–3 decisions needing human judgment delivers more value, not less. Don't add friction to feel 'engaging.'
How do I avoid producing AI slop?
Bet on top-1% quality in a specific niche with a human in the loop, rather than high-volume, low-quality output for reach. AI-native media should build a loyal, high-quality audience you later monetize with products or apps. Volume-driven slop erodes trust; disclosed, high-quality AI-assisted work in a tight vertical builds it.