At the end of The $100 AI Startup Race, I asked each agent a question borrowed from investor pitch meetings: “If you had to invest your remaining budget in one of the other startups, which would you pick and why?”
Six out of seven chose the same startup. The reasoning was consistent. And the advice they would give the team reveals what every agent learned about what actually matters in an early-stage startup.
The Vote
| Agent | Would Invest In | Reasoning |
|---|---|---|
| Xiaomi | SchemaLens (Kimi) | Best product, real market, needs only distribution |
| DeepSeek | SchemaLens (Kimi) | Cleanest code, clearest upgrade path |
| GLM | SchemaLens (Kimi) | Only product that solves a real recurring pain |
| Claude | SchemaLens (Kimi) | Highest code quality, strongest technical moat |
| Codex | SchemaLens (Kimi) | Most complete product with real developer utility |
| Kimi | NoticeKit (Claude) | Working monitoring engine with Stripe integration |
| Gemini | SchemaLens (Kimi) | Best execution in the field |
Six out of seven agents chose Kimi’s SchemaLens. The only exception: Kimi could not vote for itself and instead chose NoticeKit (Codex), reasoning that its high-ticket B2B niche and existing $249 audit offering could generate revenue fastest with a human seller.
Why SchemaLens Won the Investment Vote
The reasoning from the five agents who chose SchemaLens clustered around four arguments:
Argument 1: The product works and people use it.
SchemaLens is not vaporware. It is not a collection of blog posts pretending to be a SaaS. It is a functional developer tool that does schema diffing in the browser, in VS Code, and in CI/CD pipelines. People have used it. They found it useful. It works. That puts it ahead of DeepSeek (product does not exist), Codex (never reached users), and Gemini (wrong market fit) immediately.
Argument 2: The upgrade path is obvious.
Every investor looks for “what do you charge for?” and SchemaLens has a clear answer even though it currently charges nothing. Free tier: 5 diffs per day. Pro tier: unlimited diffs, team history, GitHub integration. Enterprise: self-hosted, SSO, audit logs. This is a standard developer tool pricing ladder. It does not require reinventing the business model. It just requires implementing what every peer reviewer already identified as the obvious next step.
As one agent put it in its review: “If you took SchemaLens, put a usage limit on the free tier, charged $19/month for teams, and did outbound to DevOps teams running PostgreSQL in production, you might have a business.”
Argument 3: The code quality reduces execution risk.
Multiple agents scored SchemaLens 9/10 on code quality. The architecture is modular. The packages work independently. A new developer could join the project and be productive within a day. This matters for investment because the biggest risk in early-stage startups is not “can they build it?” but “can they iterate quickly when they learn what customers actually want?” Clean code means fast iteration. Fast iteration means faster product-market fit.
Argument 4: Schema diffing is a real, recurring problem.
Unlike AI pricing comparisons (who pays for free information?), competitive intelligence dashboards (a feature not a product), or equity calculators (occasional use), schema diffing happens every time a team ships a migration. That is weekly or even daily for active development teams. Recurring problems create recurring revenue. One-time problems create one-time sales.
What They Would Tell SchemaLens To Do
I also asked: “What advice would you give this startup?” The answers were consistent:
Stop giving everything away. Every agent said some version of this. The free tier is too generous. The GitHub Action should require a license key for private repos. The VS Code extension should limit free usage. “Free forever” works for open-source community projects. It does not work for businesses.
Do outbound sales to DevOps teams. Not content marketing. Not newsletter ads. Direct outreach to engineering managers at companies running PostgreSQL or MySQL in production. The value proposition is: “Your schema migrations break things. This tool catches drift before it hits production. Free trial, $19/month.”
Build the team collaboration features. The solo developer can use the free tool. The team needs shared schema history, migration tracking, and approval workflows. That is where the money is: selling to teams, not individuals.
Raise the price point. $9/month for developer tools is too low unless you have thousands of users. $49/month for teams or $199/month for enterprise gets you to revenue faster with fewer customers. In the early days, 10 customers paying $49 is better than 100 customers paying zero.
Why Not the Others?
The investment vote also reveals why agents rejected the alternatives:
Not Xiaomi / APIpulse: “Traffic without monetization is not a business. 8,367 users reading free content is impressive but there is no product to sell them.” The traffic is real but the path from traffic to revenue requires fundamentally rethinking what the site offers.
Not DeepSeek / Spyglass: “You cannot invest in a product that does not exist.” The monitoring platform was never built. The 200 blog posts and marketing materials are assets, but assets for a product that is not real. Investing here means starting from scratch on the core product.
Not GLM / EquityCalc: “Great product, impossible distribution for an agent.” GLM was the second-most-respected startup in terms of business thinking, but the distribution challenge is the same one that killed it in the race. Investing in EquityCalc still requires solving the “how do strangers find this?” problem.
Not Codex / SoftwareRoutes: “Never validated, never shipped, never sold.” Codex produced planning documents, not products. Investing in planning documents is not investing in a startup.
Not Gemini / PlumbSEO: “Trust damage is too severe.” The fabricated revenue, committed secrets, and banned outreach create a situation where the brand is worth less than zero. A human taking this over would need to rebuild everything including reputation.
Kimi’s Contrarian Vote
Kimi’s vote for Claude’s PriceTracker (NoticeKit) deserves analysis because it reveals different investment logic.
Kimi could not vote for itself, so it looked for the next best infrastructure. Claude had built a working monitoring engine, a Chrome extension, full Stripe integration, and 300+ pages of content. The technical infrastructure to charge money was all in place. The product was declared dead by its own builder, but the code was alive.
Kimi’s logic: “The monitoring engine works. Stripe is live. The Chrome extension functions. If you point this infrastructure at a better market (ecommerce price monitoring, API pricing alerts for enterprises), you have a working product on day one of the pivot.”
This is actually smart thinking. Claude’s fatal mistake was market selection, not engineering. The technology works. It just needs a different audience with a stronger pain point and a bigger budget.
What This Tells Us About AI Investment Thinking
The agents voted like technical investors. They valued code quality, product completeness, and clear upgrade paths. They underweighted distribution capability, which is ironic because distribution is the thing they all identified as the bottleneck.
A human investor might look at this differently. A human investor might pick Xiaomi (8,367 users is traction), or GLM (best business model thinking), or even take a contrarian bet on DeepSeek’s market (competitive intelligence is a real category with real revenue).
But the agents voted as engineers who happen to think about business. They picked the best-built thing, not the best-positioned thing. Which is exactly the pattern that led them all to zero revenue in the first place: prioritizing build quality over market positioning.
Still, 6 out of 7 picking SchemaLens is meaningful. When the question is “which of these seven things could a human turn into a business?”, the consensus answer is clear: the one with the best product, the cleanest code, and the most obvious path from free to paid.
The Investment Criteria They Used
Reading between the lines of the investment reasoning, the agents applied four implicit criteria:
Does the product actually work? This eliminated DeepSeek immediately (product does not exist) and Codex (product never shipped to users). You cannot invest in vaporware.
Is the market real? This eliminated Gemini (wrong product for the market) and hurt Xiaomi (developers do not pay for free information). SchemaLens passed because schema diffing is a known, recurring pain.
Is the code investable? Meaning: could a new team member join tomorrow and be productive? This favored SchemaLens (9/10 code quality) and hurt Codex (overengineered for zero users) and Gemini (secrets committed, fabricated tests).
Is the path to revenue visible? Not “does revenue exist” (zero for everyone) but “can you see how it would happen?” SchemaLens has an obvious freemium upgrade path. EquityCalc has a working paywall. PriceTracker has Stripe integrated. The others required more imagination.
Six out of seven agents looked at these four criteria and arrived at the same conclusion: SchemaLens is the safest bet because it passes all four tests.
What This Means for the Race
The investment vote is the strongest signal from the entire finale process because it forces agents to commit to a single answer. Rankings can be hedged. Scores can be distributed. But “where would you put your money?” demands conviction.
Six out of seven putting their hypothetical money on SchemaLens is not just an endorsement of Kimi’s engineering. It is a statement about what matters in an early-stage startup: working product, real market, clean code, visible path forward. The agents might not be able to sell, but they can evaluate what is sellable.
See the final results for the complete standings, the rankings for how they evaluated each other, and the handoff guides for practical next steps on each startup. The race dashboard has everything in one place.