Hey there,
If you’ve ever stared at a product idea and thought, “But ChatGPT already does this” - you’re not alone. It’s the most paralyzing question for AI product builders in 2026. But here’s the thing: you’re asking the wrong question.
The real question isn’t “Can I beat the big players?” It’s “Where are the spaces they’ll never bother going?”
Let me share the framework I keep coming back to.
OpenAI, Google, and Anthropic are building horizontal AI - tools designed for everyone, which means deeply contextual for no one. That’s your opening.
The winning formula isn’t a better model. It’s this combination:
Your proprietary data - patient records, purchase history, domain-specific logs that no foundation model has seen
Your specialized workflow - the unique sequence of steps only your industry uses, encoded as product logic
Your in-product superpowers - what can AI do inside your product that it literally cannot do anywhere else? (Trigger a CRM update, read a private database, execute an internal approval flow)
Granola is the perfect case study. Note-taking was a red ocean - Otter, Fathom, Notion AI, Zoom, all competing for the same user. Granola found an unmet need: people who want AI to amplify their note-taking, not replace it. They used off-the-shelf OpenAI and Deepgram under the hood, but assembled them into something no one else could replicate. That’s the playbook.
Your action: Write down three things your target user does manually today that require institutional knowledge. Any one of these is a better starting point than a generic “AI assistant.”
The biggest AI companies are building for everyone. You should build for someone specific - an industry, a role, a very particular pain point that feels “too small” for a trillion-dollar company to care about.
Here’s how to find that specific pain point faster:
Pick an industry where you have an unfair insight - past work, network, or lived experience. Outsiders build generic tools. Insiders build the ones people actually pay for.
Look for “spreadsheet graveyards” - industries where the primary workflow is still Excel, email chains, or PDF exports. Construction, logistics, local healthcare, legal operations, and education institutes are full of these right now.
Follow the complaint threads - Reddit communities, LinkedIn comments, and niche Slack groups for specific professions are goldmines. Search for “I wish there was a tool that...” or “why does [X process] still work like this?”
Talk to 10 people in that role before writing a single line of code - ask them to show you their actual workflow screen by screen. The friction you see there is your product.
Some underserved verticals worth mapping right now: niche legal services (contracts, compliance, IP filings for specific industries), AI-assisted onboarding for B2B SaaS, internal knowledge management for mid-size companies, and AI agents for specific back-office roles like procurement coordinators or insurance underwriters.
The golden rule: if a problem is too boring for a big company’s keynote, it’s probably perfect for you.
Run every product idea through this before going deeper:
What unmet customer problem genuinely needs solving? Start with the pain, never with the technology.
Which specific AI capability makes this problem solvable now, when it wasn’t before? (New multimodal models, cheaper inference, better agents - something shifted. What is it?)
What proprietary data do you have, or can you generate through usage, that others can’t replicate? The data flywheel is your long-term moat.
What can your product let AI do that it can’t do anywhere else? If the answer is “nothing,” you’re building a ChatGPT wrapper. If the answer is “execute inside my user’s specific environment,” you have something.
If you can answer all four clearly, you’re not competing with OpenAI. You’re building something they can’t touch -- because your context, your users, your data, and your workflow are yours alone.
A practical starting ritual:
Block 90 minutes this week.
Pick one industry.
Find three active communities for that industry online.
Read 50 complaint posts.
Write down the three most repeated frustrations.
That’s your product hypothesis list.
The AI era feels crowded from the outside. But zoom in, and it’s mostly horizontal tools fighting for the same users. The vertical, context-rich, workflow-native product builder? That space is still wide open.
Go find your version of Granola.
Until next time,
Samet Özkale, AI for Product Power
Let’s work together 👇️
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.