A lot of AI business content stops at the revenue screenshot. This episode is more useful because Andrew Warner brings the examples, and Hiten Shah keeps asking the adult questions: where did the customers come from, what are the costs, what churn is hiding underneath the number, and is this actually software or a service wearing a software hat?
That is the right lens for 2026. AI makes building faster. It makes prototyping cheaper. It lets a solo founder ship things that used to require a small team. But it does not make distribution, margins, trust, retention, or judgment disappear.
Source note
The episode comes from The Next New Thing AI. The supporting video hub is here: How they make money vibe coding. The case-study revenue numbers below are discussed in the episode and source hub; treat them as reported founder claims unless a company publishes audited financials.
Hiten Shah is a useful guest for this topic because his background is not "AI tourist." Crazy Egg is a long-running website optimization company whose site lists heatmaps, recordings, A/B testing, web analytics, surveys, and conversion tools. His newer product, Typeahead, is a local Mac autocomplete tool that says it works across apps and keeps writing on the user's machine.
The core idea: AI lowers build cost, not distribution cost
The mistake is thinking "I can build it now" means "customers will come." AI changes the first half of the equation. It helps you prototype faster, write code, make videos, scrape data, build calculators, generate landing pages, and test niche products. It does not automatically answer:
- Who wakes up already wanting this?
- Where do those people spend attention?
- What search query, community, job role, or painful workflow reveals intent?
- What does it cost to deliver the product after AI compute, support, and rework?
- Why will the customer stay after the first month?
That is why Hiten keeps returning to distribution. The product is easier to build. The market is not easier to fool.
Examples from the episode
1. The AI power dialer: AI inside a defined business problem
The episode starts with an AI-enabled power dialer, described in the source hub as Clairvo, that reportedly reached $1M ARR. The interesting part is not only that it was built with Claude Code. The interesting part is how AI was aimed at a specific metric: improve call pickup rates and sales-call volume.
That is a better use of AI than "give me a startup idea." The founder already had a problem space. Claude helped generate and test implementation ideas inside that container.
Lesson: AI is strongest when pointed at a real business constraint, not a blank page.
2. Shipper / app builders: category demand is real, but differentiation is hard
The source hub describes Shipper as an AI app builder doing roughly $50K/month gross and $307K ARR. The attraction is obvious: people want to build websites, web apps, mobile apps, Chrome extensions, bots, and internal tools without assembling a full engineering team.
Hiten's pushback is the important part. If big players like Lovable, Bolt, Replit, and other app builders define the category, a smaller builder can draft behind them. But that also means the product surface is easy to compare and hard to defend.
Lesson: app builders can work, but only if you have distribution, a sharper niche, or a workflow the broad tools underserve.
3. Fractional Chief AI Officer: probably services, and that is fine
One example is a Fractional Chief AI Officer / AI implementation business described as doing millions in recurring revenue. Hiten questions whether that should really be called ARR if the business is service-heavy.
That distinction matters. Services revenue is not bad. In fact, for AI, services may be the best starting point. Companies do not just need a model. They need workflow mapping, tool selection, data cleanup, permissions, training, change management, review gates, and somebody accountable for the result.
Lesson: do not hide from services. In AI deployment, services are often where the value is.
4. AI video tools: revenue can be real, but margins matter
The episode discusses Revid, an AI video tool reported in the source hub as doing more than $600K/month. Hiten's question is exactly the one founders should ask: what are the generation costs, gross margins, support costs, and churn?
AI video can be an excellent market because the pain is visible. People want short videos, ad creatives, explainers, and social content. But if each output relies on expensive model calls and repeated generations, revenue alone is not enough to judge the business.
Lesson: AI media businesses need margin math, not only impressive demos.
5. CutCoach: niche beats broad when intent is sharp
The CutCoach example is a niche mobile app for combat athletes managing weight cuts. The source hub reports $20K/month. What makes the example useful is not that it is huge. It is that the audience is specific, the pain is concrete, and the distribution channel is reachable through organic social and athlete communities.
A niche like this may not become a venture-scale company. It may still become a strong solo business.
Lesson: small markets can be good markets when the pain is expensive, urgent, and easy to target.
6. AI explainer videos built with Base44: productized workflow knowledge
The Base44 example is framed as an AI explainer-video product built quickly with no-code and AI tools. The deeper lesson is arbitrage. A founder who understands how to combine AI video tools, prompts, landing pages, and user needs can package that know-how for buyers who do not want to learn the workflow.
That is not fake. Many useful businesses begin as "I know how to do this annoying new thing, and you do not."
Lesson: workflow expertise can be productized before it becomes pure software.
7. Luxury restroom trailer directories: boring SEO still works
The directory example is wonderfully unglamorous: luxury restroom trailer rentals. This is the kind of niche that looks ridiculous until you remember that local commercial searches often have high intent and fragmented supply.
AI helps with the hard parts: gathering structured data, creating city pages, comparing vendors, writing helpful content, and maintaining the directory. But the business still depends on search intent and data quality.
Lesson: boring directories can work if the search query is valuable and the page genuinely helps the buyer.
8. Simple utility sites: the app is easy, the traffic is not
Mortgage calculators, typing tests, name generators, and other utility pages are easy to build with AI. The harder question is whether the page can rank, retain attention, earn links, and monetize without annoying users.
If a user searches for a calculator, gets the answer, and leaves satisfied, that can be a good business. If the page is thin AI filler around a commodity widget, it is just another page waiting to be ignored.
Lesson: utility sites need intent match and distribution, not just generated code.
The Hiten filter for AI businesses
If you are evaluating an AI business idea, run it through this filter before building:
- Distribution: where do customers come from without wishful thinking?
- Margin: what does it cost to deliver after model calls, video generation, hosting, support, and rework?
- Retention: why does the customer come back next month?
- Category: is this software, services, marketplace, content, directory, or arbitrage?
- Defensibility: what do you know, own, or reach that a broad AI tool does not?
- Truth in metrics: is the number monthly revenue, gross revenue, ARR, annualized services revenue, or profit?
- Workflow depth: does the product solve a real job or just demonstrate a tool?
I would add one more JQ AI SYSTEMS question: can this become a repeatable system a client would pay to own?
The best opportunity: AI deployment services
If I had to pick the easiest serious AI business from this episode, I would pick the same direction Hiten points toward: AI deployment services.
Not because it sounds trendy. Because it matches the market gap. Companies are buying AI tools faster than their teams can turn them into reliable workflows. That creates demand for people who can:
- map a workflow from intake to delivery;
- choose the right model and tool stack;
- connect email, CRM, docs, spreadsheets, Slack, Zapier, or internal data;
- build review queues and permissions;
- train teams on what to trust and what to check;
- turn repeatable work into reusable automations.
The first version can be a service. The second version can be a playbook. The third version can become software or a managed system.
This is also where tools like Zapier MCP matter. The value is not that AI can talk. It is that AI can act inside controlled permissions across real business apps.
Case-study video wall
The source hub includes eight supporting case-study videos. I am embedding them here as a compact research wall so you can inspect the examples directly.
Builder checklist
Before you build another AI app, answer these questions in writing:
- Customer: who has the painful workflow?
- Trigger: what makes them look for help now?
- Distribution: search, social, community, outbound, partnerships, marketplace, or existing audience?
- Workflow: what exact job are you improving?
- Margin: what does each result cost to generate, support, and fix?
- Retention: why would they keep paying after the novelty fades?
- Review: where does a human approve outputs before risk appears?
- Moat: data, distribution, domain expertise, process, trust, or brand?
- Service-to-software path: what parts repeat across clients?
Vibe coding is real leverage. But the winners will not be the people who can generate the most apps. They will be the people who know which problems deserve an app, which problems deserve a service, and where the customers already are.