DEV Community

Sanjay Divakar
Sanjay Divakar

Posted on

I'm 15 and building an AI C-suite that runs a business for you. Here's the prototype

#showdev #ai #javascript #buildinpublic

I'm 15 and I've been building Founderless: you bring a business idea and some cash, and a team of AI agents (CEO, CTO, CFO, COO, CMO) builds the product, sells it and markets it. You either approve the big decisions from your phone (Cockpit mode) or switch on Autonomous mode with a spend limit.

It's a working prototype with a waitlist right now, and public launch is planned for May 2027. Here's how the prototype works, plus the parts I'm unsure about.

The flow

  1. New company: type an idea, or do a 3-question interview and the CEO agent suggests 3 ideas. Pick Cockpit or Autonomous (with a spend threshold) and fund a wallet.
  2. Brief: the CEO streams a company brief (name, tagline, customer, problem, offer, pricing, 30-day goal). Nothing else happens until you edit and approve it.
  3. C-suite: the five exec agents stream their first-week plans.
  4. Landing page: the CTO writes the copy, which fills a tested template you can preview and download.
  5. Launch posts: the CMO drafts launch posts and a launch email (drafts only).
  6. Wallet + spend: the CFO files spend requests. In Autonomous mode anything at or under the threshold auto-approves and anything over it goes to an Approvals inbox. In Cockpit mode everything goes to the inbox.
  7. Activity log of every agent and founder action.

In the prototype, money, payments, ads and posting are simulated. Nothing is actually bought or posted yet. Every agent step is a real model call streamed live to the UI.

Stack (kept boring on purpose)

  • Node 22 with the built-in node:sqlite, so there are no runtime dependencies
  • Vanilla JS front end with a hash router and Server-Sent Events for streaming
  • A provider adapter, so it runs on a local llama.cpp model or on Claude/OpenAI
  • On the local 9B model, JSON output is grammar-constrained with response_format: json_schema, so it can't emit broken structure. I also parse partial JSON so the UI can preview output while it's still streaming
  • A Playwright end-to-end run of the whole flow in both modes

Where I'd love input

  • Spend guardrails: a threshold + approvals inbox is the first layer. What else would you want before trusting agents with real money? Per-category caps? A daily budget? A "two agents must agree" rule?
  • Autonomous vs Cockpit: would you ever actually turn autonomous on?

Demo video of the prototype: https://founderless.io/media/founderless-demo.mp4

If you want to follow along, the waitlist is at https://founderless.io/?src=devto (you get a referral link after signing up, and if 3 friends join you're in the first cohort).

Thanks for reading. Brutal feedback welcome.

Top comments (0)