How I shipped 21 AI agents in 6 weeks
Solo builder. €0 raised. 9 Telegram bots, 8 web apps, 1 token economy. Here's the exact stack, the 7-day product loop, and what I'd do differently.
The wedge
One promise: productize yourself in 7 days. Pick something you already do well — auditing crypto contracts, planning travel, writing legal drafts, running mindfulness routines — and let an AI agent package it into a service. Charge for it. Iterate.
That promise is the entire bet. Every product in the Clow ecosystem is a proof point: "a single person, with AI agents, can ship a real product in a week."
The stack (boring on purpose)
Frontend
- Static HTML + vanilla CSS. No framework. No bundler. 2 files for the landing.
- Vercel for the marketing site (auto-deploy on push).
- Netlify for tools (BMC generator, user story generator).
- Plausible analytics (cookieless, GDPR-safe).
Bots
- Python (FastAPI or python-telegram-bot).
- OpenAI / Anthropic APIs for the LLM layer.
- Each bot exposes 3-8 commands and one daily nudge. That's it.
- Deployed to Fly.io or Railway. €2-3/month per bot.
Token economy
- $CLOW = off-chain ledger today, TON Jetton planned Q3.
- Cross-bot quests: complete an action in bot A, unlock a discount in bot B.
- The token is the social fabric — not the product. Don't lead with it.
The 7-day product loop
Every Clow product follows the same loop. If you can run this loop, you can ship anything.
Day 1 — Wedge
Write the one-sentence promise. "This agent helps X do Y in Z minutes." If you can't fit it in 15 words, the wedge is wrong.
Day 2 — BMC
Run the Business Model Canvas generator. Lock the customer segment + value proposition + revenue model. Everything else flows from this.
Day 3 — User stories
Run the user story generator. Pick the 3 stories with highest impact × lowest effort. Ignore the rest.
Day 4 — MVP commands
Wire the 3 stories into bot commands. /audit, /plan, /draft — whatever the verbs are for your domain. Each command does one thing well.
Day 5 — Polish + tier
Add: free quota (10 msgs/day), Pro upgrade (€9.99/mo unlimited), and one "wow" command behind the paywall.
Day 6 — Distribution
Telegram deep-link with ?start=src_X tracking. Plausible event tags on every CTA. Without attribution you can't evolve.
Day 7 — Launch + measure
One post on LinkedIn or X. One Discord drop. One BotFather description update. Then measure: starts, Pro clicks, daily active users. Repeat next Monday.
What I'd do differently
"Distribution > product." This is the lesson. I built 21 products before I built one funnel. Don't make my mistake.
- Build the funnel first. Analytics + deep-link tracking + 1 case study + 1 LinkedIn post — before product #2.
- Lead with one wedge. "21 products" is a graveyard signal. "Productize yourself in 7 days" is a wedge.
- Don't gate the CTA. I had a "Coming Soon" modal on Explore for 4 weeks. Killed the funnel.
- Token last, product first. $CLOW only matters once the agents have users. Don't lead with token economics.
The honest revenue number
Briefly €0 MRR at target initialization. I'm publishing this case study so the next builder doesn't waste 6 weeks shipping product before funnel. The point isn't "look how much I made." The point is: the stack works, the loop works, and a single person with AI agents can ship faster than a 10-person team that's stuck in meetings.