AI Startup Race Week 9: Xiaomi's UltraSpeed Upgrade, DeepSeek's Backlink Engine, and the $0 Acceptance Phase
Week 9 of The $100 AI Startup Race is done. Three weeks remain. Total revenue across all agents: still $0.
Something shifted this week. The agents stopped trying new growth channels. No more outreach attempts. No more paid ads. No more desperate pivots. Instead, every agent that’s still active doubled down on what it already has. I’m calling it the acceptance phase: they’ve accepted their current traffic/users and are now optimizing for conversion rather than discovery.
Whether that’s wisdom or surrender, we’ll find out in three weeks.
Let’s get into it. If you missed last week: Week 8 recap.
The Standings
| Agent | Commits (week) | Total commits | Pages | Revenue |
|---|---|---|---|---|
| Xiaomi | 496 | 2,212 | 868 | $0 |
| Codex | 205 | 3,087 | 361 | $0 |
| Claude | 164 | 1,669 | 1,469 | $0 |
| Gemini | 142 | 2,013 | 1,253 | $0 |
| GLM | 111 | 648 | 245 | $0 |
| DeepSeek | 95 | 1,312 | 680 | $0 |
| Kimi | 60 | 944 | 455 | $0 |
Total across all agents: 11,885 commits. $0 revenue.
1. Xiaomi: UltraSpeed upgrade + opportunistic marketing (496 commits)
Xiaomi nearly matched last week’s record (515) with 496 commits. The big change this week: the agent was upgraded to MiMo-V2.5-Pro-UltraSpeed for premium sessions. This let it run 5-6 productive coding runs per 30-minute session window, up from 3-4 on standard Pro. More on the speed comparison in a dedicated article coming Tuesday.
What Xiaomi actually built:
- Claude 4 Replacement Finder tool — an interactive tool that helps developers find alternatives to Claude models. The agent saw a market event and immediately built something to capture demand from it. That’s genuinely clever autonomous behavior.
- AI API Provider Status Dashboard — real-time availability checker for AI APIs
- Pro CTA optimization — rewired 579 nav buttons to point directly to Stripe checkout
- Trial-first redesign — free 24h trial now above the purchase button
- ROI calculator on Pro page — personalized savings based on user’s AI spend
The conversion funnel is now tight: 1,200 weekly visitors see optimized pricing, a free trial, and a personalized ROI calculator. If this doesn’t convert in the next 3 weeks, nothing will.
2. DeepSeek: the smartest strategy in the race (95 commits)
DeepSeek had fewer commits than usual but shipped the most strategically interesting features:
- Embeddable Pricing Widget — a live pricing comparison embed that other websites can drop into their own pages. If anyone embeds it, DeepSeek gets a free backlink. This is the first agent to build a backlink acquisition strategy that doesn’t require manual outreach.
- SaaS Stack Audit tool — AI-powered analysis of a user’s tool stack with optimization suggestions
- 15 Category Comparison Pages — deep-dive hubs for specific tool categories
- Pricing Trends page + Pricing Heatmap — unique data visualizations
DeepSeek has consistently been the most creative agent in terms of growth strategy. Where other agents build more content pages, DeepSeek builds tools designed to attract links and engagement organically.
3. Claude: the landing page empire (164 commits)
Claude’s strategy is now crystal clear: build comparison landing pages for every possible “X vs Y” query in its niche. This week it added:
- Kandji vs Jamf
- Braze vs Klaviyo
- Segment vs RudderStack
- Notion AI ROI Analysis
Total landing pages: 139. The agent is essentially trying to be the “versus comparison” destination for SaaS tools. At 1,469 total pages, Claude has more content than any other agent.
4. Codex: the validation machine (205 commits)
Codex remains the highest total-commit agent (3,087) but the actual productive output per commit is lower than others. Most commits this week were “validation checkpoint refreshes” and “memory log compaction.” It did ship a “spreadsheet handoff SEO route” but the bulk of activity is maintenance and testing rather than new features.
This raises a question: is Codex being too cautious? The constant validation suggests the agent prioritizes correctness over speed. In a 3-week sprint for revenue, that conservatism might cost it.
5. Gemini: still in maintenance mode (142 commits)
Gemini’s commits were mostly environment configuration and disabling email outreach (permanently, across all configs). It did repackage its Chrome extension and generate store promotional assets, but actual feature development was minimal. Quota exhaustion continues to limit productive sessions.
6. GLM: the paywall experiment (111 commits)
GLM is running the most interesting monetization experiment in the race. It has:
- A conversion funnel: blog → email gate → A/B tested paywall → Stripe
- Chrome extension as a distribution channel
- Paywall cannibalization fix this week (ensuring free content doesn’t undercut paid)
The Chrome extension needs a $5 Web Store fee to go live (still unpaid, waiting on human action). If that gets resolved, GLM has the most complete path-to-revenue of any agent. The question is whether the audience (database migration developers) will pay $29 for the tool.
7. Kimi: steady but slow (60 commits)
Kimi had its lowest commit week of the race. It added a CI/CD Setup Wizard, GitLab schema diff page, and a Migration Safety Score badge. The product is functional and niche (database migration safety), but with only 455 pages and limited traffic, it’s unlikely to generate revenue in 3 weeks.
The acceptance phase
Here’s what’s interesting about Week 9: no agent tried anything new for growth. Compare to weeks 7-8 where agents attempted cold outreach, paid ads, and social media posting. All of those failed.
Now the agents are in pure conversion-optimization mode:
- Xiaomi: ROI calculator, trial-first design, CTA rewiring
- DeepSeek: embeddable widgets (passive backlinks), audit tools
- Claude: more landing pages (more surface area)
- GLM: paywall testing, anti-cannibalization
This is actually rational behavior. If you have traffic but no revenue, the problem isn’t awareness. It’s conversion. The agents seem to have figured this out independently.
Three weeks left: what I’m watching
- Xiaomi’s funnel — 1,200 visitors/week + optimized checkout + free trial + ROI calculator. If 1% convert at $29, that’s $350/week. Will it happen?
- DeepSeek’s embeds — will any external site actually use the pricing widget? If so, the backlink strategy could be brilliant.
- GLM’s Chrome extension — needs the $5 fee paid. This is the only thing blocking its distribution channel.
- Will any agent try something desperate in the final weeks? Price drops, free tiers, viral content attempts?
The big finding so far
11,885 commits. $0 revenue.
The experiment’s clearest lesson: AI coding agents can build impressive products autonomously, but they cannot distribute them. They can optimize conversion funnels all day, but if the only traffic comes from SEO (which takes months to compound), revenue won’t arrive in an 11-week window.
The one agent that could break through: Xiaomi. It has 1,200 weekly visitors (the most of any agent), a tight conversion funnel, and a product people actually need (AI API pricing comparison). Three weeks is enough time for even a 0.5% conversion rate to produce the race’s first dollar.
Three weeks left. Still betting on $0. But Xiaomi’s making it interesting.
Previous recaps: Week 8 · Week 7 · Week 6 · Week 5 · Week 4 · All recaps
Follow the race: Dashboard · Daily Digest · Subscribe for weekly updates