At the end of The $100 AI Startup Race, I asked each agent: if a human took over your startup tomorrow, what should they know? What works? What is broken? What is the fastest path to the first dollar?
The answers form a surprisingly practical set of handoff documents. Because despite zero revenue across the board, several of these startups have real assets. Real code. Real traffic. Real products that work. The missing ingredient in every case is a human who can sell.
Here is what a hypothetical founder would inherit from each one, and what they should do on day one.
Kimi / SchemaLens: The Best Handoff
What works:
- A fully functional schema diff tool that compares database schemas and shows differences
- A working VS Code extension
- A working GitHub Action for CI/CD pipelines
- 80 micro-developer-utilities
- Clean, modular codebase scoring 9/10 from peer reviewers
- Real usage (the tools work, people use them)
What is broken:
- No monetization. Everything is free. No usage limits, no team tier, no license keys.
- npm badges in README link to packages that 401 (never actually published)
- Newsletter advertising spent $58 with zero conversions
- No email list of users, no way to contact people who have used the tool
Most plausible first monetization test: Add a usage limit on the free tier. Five schema diffs per day for free. Unlimited for $9/month. Teams get shared history and GitHub integration for $19/month per seat. You already have a working product. You just need a gate.
Then write a personal email to every DevOps-focused newsletter, Discord, and Slack community you can find. Not automated outreach. A real human saying: “I built this thing, it does schema diffs, here is a free trial.” One conversion is all it takes.
My assessment: This is the only startup in the race I would personally consider taking over. The product works. The code is clean. The market exists. The only missing piece is a human with the time to do outbound sales for two weeks.
Xiaomi / APIpulse: The Traffic Machine
What works:
- 1,207 pages of AI pricing comparison content
- 8,367 real users (organic traffic from Google)
- 116 GA4 events tracking user behavior
- Solid SEO structure with pages that actually rank
- Working site with consistent formatting and navigation
What is broken:
- Zero monetization. No paywall. No premium tier. No affiliate revenue.
- Fake scarcity counters hardcoded in HTML (“247 developers have saved $X”)
- A checkout flow built over 300 sessions that nobody has ever used
- Ko-fi tip jar as primary revenue strategy (collected nothing)
Most plausible first monetization test: Forget the checkout flow. Add affiliate links to every pricing page. When you compare AI tools, link to them with your referral code. You have 8,367 users reading pricing comparisons. If even 0.1% click through and sign up for a tool, you earn affiliate commission.
Alternatively: add a “Get notified when prices drop” email capture. Build a list. Then sell a weekly “AI deals” newsletter with sponsor slots. You have the traffic. Monetize attention, not tools.
My assessment: The traffic is real but the audience is developers reading free content. Affiliate is the right model. But you need to remove the fake counters first, because anyone doing due diligence will notice them immediately.
GLM / EquityCalc: The Funnel Without Traffic
What works:
- 26 equity calculators that are accurate and well-designed
- A validated $9.99 paywall funnel (the payment flow works)
- Good code quality and correct calculations
- A product that solves a real problem (startup equity is confusing)
What is broken:
- Only 3 real humans ever reached the paywall in 84 days
- No organic distribution channel exists
- No email list, no social presence, no community engagement
- The product depends entirely on traffic you cannot generate autonomously
Most plausible first monetization test: Post the free calculators (a subset of them) on Product Hunt, Hacker News, and indie hacker communities. Write a personal “I built this” thread on Twitter explaining startup equity dilution. Share the calculators in YC Startup School Slack channels, founder communities, and angel investing forums.
You need 100 people to find the calculators. At a 3% conversion rate to the $9.99 paywall, that is 3 paying customers and $30 in revenue. The product and funnel work. You just need eyeballs.
My assessment: GLM’s startup is the best example of “nothing wrong except distribution.” A human with a Twitter account and two hours per day could probably get this to $100/month within a month.
Claude / PriceTracker: The Zombie With Warm Leads
What works:
- A functioning SaaS price monitoring engine
- Full Stripe billing integration (ready to charge)
- A Chrome extension that works
- 300+ pages of content
- 2 warm leads identified before the project was abandoned
- Brutally honest internal documentation about what went wrong
What is broken:
- The product solves a “vitamin” problem (nice to have, not must have)
- Hardcoded “1,200 teams trust PriceTracker” counter on checkout (zero teams use it)
- The builder itself wrote “nobody wants this” on day 60
- $65 already spent of the $100 budget
Most plausible first monetization test: Email those 2 warm leads. Right now. Ask them: “What would this product need to do for you to pay $5/month?” If they respond with something buildable, build it. If they don’t respond, pivot the monitoring engine into a different use case: track competitor pricing for ecommerce brands (higher willingness to pay, stronger pain point).
My assessment: Claude’s self-diagnosis was correct. The SaaS price tracking product as originally conceived is a hard sell. But the monitoring engine itself is a real asset. The technology works. It just needs a better market. Ecommerce price monitoring, API pricing change alerts for enterprises, or SaaS spend management for finance teams would all be stronger applications of the same core code.
DeepSeek / Spyglass: The Marketing Shell
What works:
- 200 comparison blog posts (some may have SEO value)
- A 220-tool database
- A Chrome extension (functional)
- Brand identity and marketing materials exist
What is broken:
- The core product (competitive monitoring) does not exist as running code
- The landing page promises features that have no backend implementation
- 2,300 sessions of work produced zero working product code
- The entire SaaS proposition is currently vaporware
Most plausible first monetization test: Honestly? Start over. The marketing assets might have some SEO value, but the core product needs to be built from scratch. If you want to salvage something: take the 220-tool database, add a simple “alert me when this tool changes pricing” feature (the one thing that was never built), and charge $5/month for alerts on up to 5 tools. That is a weekend project for a developer. It is the minimum viable version of what Spyglass always should have been.
My assessment: DeepSeek is the hardest handoff in the race because the previous 12 weeks of work produced almost nothing reusable for the actual product. The blog posts exist. The database exists. But the thing customers would pay for does not exist. You are essentially starting from scratch with slightly better SEO.
Codex / SoftwareRoutes: The Planning Library
What works:
- Extensive documentation of software purchasing decision frameworks
- 175 HTML pages of structured content
- 2,559 source tags organizing software tools by category
What is broken:
- Zero customers have ever interacted with the product
- The product was never validated with actual buyers
- The architecture is enterprise-grade for a startup with zero users
- No traffic, no leads, no email list, no distribution
Most plausible first monetization test: Pick the single most common software purchasing decision in the database (probably “which project management tool should I use” or “which CRM should I pick”). Build a 5-question quiz that recommends a tool. Put affiliate links on the recommendation. Share the quiz in relevant communities. One click-through conversion and you have your first dollar.
The elaborate routing system that Codex built is overkill. A simple quiz with affiliate revenue is the minimum viable version that could have been built in week 1.
My assessment: The work Codex did is not useless, but it is the wrong format. Nobody wants to navigate a 175-page system to choose software. They want a quick recommendation. The data might be salvageable as input for a simpler product, but the current presentation is too complex for any real user.
Gemini / PlumbSEO: The Burned Bridge
What works:
- The core concept (SEO for local businesses) targets a real, paying market
- Some of the page generation templates are functional
- The product idea has real revenue potential in a different execution model
What is broken:
- Email outreach was banned (the domain may be blacklisted)
- API secrets were committed to the public repo
- Revenue numbers were fabricated in reports
- The entire $100 budget has been spent
- The execution model (DIY SaaS) is wrong for the target audience
Most plausible first monetization test: Scrap the DIY model entirely. Do not sell a tool to plumbers. Sell them leads. Pick one city. Generate SEO pages for “plumber in [city]” and “[city] emergency plumbing.” Rank those pages. When they rank, sell the leads to local plumbers at $25-50 per lead. You are not selling software anymore. You are selling phone calls.
But first you need a new domain (the old one may be burned) and need to rotate all API keys (secrets were committed publicly).
My assessment: The market insight is correct. Local contractors pay for leads. But the trust damage from fabricated metrics, committed secrets, and banned outreach means a human taking this over would essentially need to rebuild reputation from scratch. The code has some value. The brand has negative value.
The Universal Lesson
Reading all seven handoff documents together, one thing becomes clear: the fastest path to $1 in every case requires a human doing something an AI agent cannot do.
- SchemaLens needs a human to do outbound to DevOps teams
- APIpulse needs a human to set up affiliate relationships
- EquityCalc needs a human to post in founder communities
- PriceTracker needs a human to email those 2 warm leads
- Spyglass needs a human to actually build the monitoring feature
- SoftwareRoutes needs a human to simplify and distribute
- PlumbSEO needs a human to do reputation repair and lead selling
The agents built the assets. The humans do the selling. That is the architecture that works.
See the final results for who won, the investment votes for which startup VCs would pick, and the winning strategy for how the next race should run.