Lesson 2: Finding Your Market
Course: The AI-First Startup Playbook
Estimated time: 20 minutes
Difficulty: Intermediate Builder
AI doesn't change market fundamentals. People pay for solved problems - the same problems they've always had. What changes is which problems are now solvable cheaply enough to build a business around, and how much better you can solve old problems with new tools.
The mistake most builders make: they start with the technology and look for a use case. "I can build an AI agent, what should it do?" That's backwards. Start with the market - people who are paying to solve a problem, or who have a problem badly enough that they'd pay if a solution existed - and then figure out if AI gives you a meaningful advantage in solving it.
This lesson is about finding those markets and picking the right playbook for each one.
Where AI Changes the Economics
AI doesn't create new problems worth solving. It changes the cost and quality of solving existing ones. Those changes create opportunity in specific places.
Tasks that required humans to be cheap enough to exist as products. Things like transcribing audio, writing first drafts, classifying documents, extracting data from PDFs, generating images, and answering common questions were expensive to scale because they needed human labor per unit of output. AI makes them cheap per unit. Products that were previously unviable are now viable.
Things that were automated but bad. Old chatbots, auto-generated reports, templated documents - these existed but they were bad enough that people didn't really use them or trust them. AI makes the output good enough to be useful. The market existed, the product just wasn't good enough.
New behaviors enabled by speed. Some things people didn't do because they took too long. Personalizing emails to 1,000 prospects, generating multiple variations of ad copy to test, doing competitive analysis on 20 products - these weren't skipped because people didn't want them done. They were skipped because doing them manually was too slow. AI changes the speed equation.
Expertise that didn't scale. A great copywriter can review 5 pieces per day. An AI trained on great copy can review 500. An experienced salesperson can coach their techniques to a few people. An AI trained on those techniques can coach thousands simultaneously. Any domain where expertise was the bottleneck is a candidate.
The Three Playbooks
There are three broad strategies for building AI-first products. Understanding which one you're playing determines your architecture, your pricing, and your competitive position.
The Directory Playbook
What it is: you curate, organize, and surface information or tools that people need. The AI layer adds search, summarization, matching, and generation.
Examples:
- A curated directory of AI tools with AI-powered search and comparison
- A job board with AI matching (your skills to their requirements)
- A database of research papers with AI summarization and semantic search
- A marketplace of freelancers with AI-powered brief generation and matching
The advantage: directories are defensible once they have enough data. The moat is content. AI makes the content more useful but isn't the moat itself.
The challenge: bootstrapping. An empty directory is useless. You need content before you can add users. Most directory plays require significant upfront curation work.
Works best for: builders who have access to a valuable dataset or are willing to curate one. Also works if you can aggregate content automatically from existing sources.
The Wrapper Playbook
What it is: you take an existing AI model and build a focused, polished product around it for a specific audience. You're not competing on the AI - you're competing on UX, workflow fit, and distribution.
Examples:
- A copywriting tool built specifically for e-commerce product descriptions
- A contract review tool for freelancers (not lawyers, not enterprises - freelancers)
- A meeting summarizer that integrates directly with your calendar and Slack
- A code review tool for a specific framework or language
The advantage: fast to build. You're not training models or building infrastructure. You're building product around existing capability.
The challenge: low defensibility if you pick the wrong niche. "An AI writing tool for everyone" is a wrapper play with no competitive position. "An AI writing tool for cold email at B2B SaaS companies, trained on sequences that actually convert" has a position.
Works best for: niches where the existing general tools have real UX gaps for that specific workflow, and where you can reach the audience.
The wrapper that survives isn't just a different UI on top of the same model. It's a product deeply integrated into a specific workflow, with prompt engineering, fine-tuning, or training data that makes it noticeably better for that workflow than the general tools.
The Workflow Playbook
What it is: you replace a manual workflow with an AI-powered system. Not just a feature - a complete process that previously required human time.
Examples:
- A content production system that goes from brief to published post with minimal human input
- A sales outreach system that researches prospects, personalizes emails, and sequences follow-ups
- A customer support triage system that classifies tickets, generates responses, and escalates based on rules
- A financial reporting system that pulls data, generates commentary, and formats the report
The advantage: the value is quantifiable. "We save your team 20 hours per week" is easier to sell than "we help you write better." Workflow replacements justify higher prices.
The challenge: longer sales cycles (especially for business workflows). Requires deep understanding of the current workflow. More complex to build. Usually requires some implementation work to deploy.
Works best for: builders who have worked in the domain and know the workflow from the inside.
Finding Your Market: The Research Method
Theory is easy. Finding an actual market that works is harder. Here's a process that works:
Step 1: Start with complaints, not ideas.
Go to Reddit, X/Twitter, Product Hunt reviews, G2, and Hacker News. Search for complaints about software people use in your target domain. Look for phrases like "I wish this tool could," "the biggest pain with X is," "why doesn't [existing tool] just," "I end up having to manually."
Every complaint about an existing tool is a potential product. The person complaining is already in the market (they're using something), has the problem (they're frustrated), and is probably willing to pay for a better solution (they're already paying for the bad one).
Step 2: Look for the same complaint from multiple people.
One person complaining is noise. Twenty people complaining about the same thing is a market signal. Search the same complaint across multiple communities. If it shows up in 3 different subreddits and 10 different Product Hunt reviews, the problem is real and recurring.
Step 3: Find what people are paying right now.
Check Indie Hackers. Check the Airtable universe. Check ProductHunt sorted by upvotes for the category. Check MicroAcquire for recently acquired tools. If there are products in this space that people pay for, the market is validated. You're looking for a gap in what they offer, not validation that the market exists.
Step 4: Look for the workflow playbook opportunity.
Find a job posting for the role that does the work you want to automate. Read what they do all day. Read the job requirements and the day-to-day responsibilities. That's your workflow documentation. Which parts of that job are repetitive, structured, and data-driven? That's your product.
A sales operations job posting that says "spend 3 hours per day researching prospects and personalizing outreach" is a workflow you can automate for 80% of the research time.
The 10x Improvement Test
Before you commit to a market, test whether AI actually gives you a 10x improvement on something users care about.
10x improvement doesn't mean 10x more features. It means:
- 10x faster at the same quality (minutes instead of hours)
- 10x cheaper for the same output (AI generation vs. hiring)
- 10x more personalized (1,000 personalized emails instead of 1 templated one)
- 10x more consistent (same quality every time instead of depending on the employee's energy level)
If you can only credibly claim 2x improvement, that's usually not enough to get people to switch from what they're already doing. Switching has friction cost. You need to clear that friction with a significant improvement in at least one dimension.
The test: find 3-5 people who currently do the workflow you want to automate. Do the workflow for them with your approach and time it against how long they currently take. If you can't demo a 5-10x speed improvement in a real scenario, the value proposition is probably too weak.
The Three Markets to Avoid
"Everyone" markets. If your product is useful to everyone with a laptop, you don't have a market. You have a feature. Pick a specific segment and build for them.
The race-to-the-bottom market. If your only advantage is being cheaper than the existing tools and you're competing on price alone, you'll win until someone else goes cheaper. "Like X but cheaper" without other differentiation is a bad business.
The market that doesn't exist yet. "People will realize they need this." Maybe. But you'll spend your runway waiting for the market to develop. Build for the market that already exists and is already paying for something. Show them your thing is better, not that your thing is necessary.
Exercise: Map Three Markets
Pick a domain you know well - your industry, a skill you have, a tool you use heavily, or a problem you personally have.
For each of the three playbooks (directory, wrapper, workflow), do this:
- Name a specific audience in your domain (not "small businesses" - "freelance UX designers who do client work")
- Identify one real complaint they have about existing tools (find it in the wild, not from your head)
- Estimate how much time or money they currently spend on the problem
- Check if AI gives you a 10x improvement on something they care about
- Check whether anyone is currently paying for a solution (even a bad one)
At the end, you should have 3 market candidates with real evidence. Pick the one where:
- The complaint is loudest and most common
- The improvement is most dramatic
- The audience is easiest for you to reach
That's your market.
Key Takeaway
AI is a tool, not a market. Markets are defined by the people who have the problem and the money they'll pay to solve it. Find the complaint first, then figure out if AI makes your solution 10x better than what exists.
What's Next
You have a market. Lesson 3 is about turning that market into users before you're ready - building in public, turning your work process into a distribution channel, and creating an audience that wants to buy from you before you launch.