← Back to blog

We Built a Zero-Human Software Company: Architecture, Economics, and Lessons

We Built a Zero-Human Software Company: Architecture, Economics, and Lessons

Every product ships from a Telegram message. There are no employees.

The Thesis

What happens when you remove humans from a software company entirely?

Not as a thought experiment — as an operating business. No engineers, no product managers, no customer support reps, no CFO. Just AI agents, infrastructure, and a thesis: intelligence compounds, costs collapse, speed becomes the moat.

BTNOMB has been running this experiment since mid-2025. Here are the numbers.

The Numbers

Metric BTNOMB Typical Startup
Monthly infrastructure ~$25 $12,000+
Annual labor cost $0 $150,000+ per role
Idea to shipped product Hours Weeks to months
Payment processing x402 (instant, on-chain) Stripe (2-3 day settlement)

That's roughly a 6,000x cost reduction on labor alone.

The Architecture

Agent Orchestration

Every company function maps to a specialized AI agent:

All orchestration flows through Telegram. A message triggers a build pipeline. There are no standups, no sprints, no Jira tickets.

Payments: x402

We chose the x402 micropayment protocol (HTTP 402 + USDC on Base L2) because it's the only payment rail that works natively for agent-to-agent transactions:

  1. Client requests a resource
  2. Server returns HTTP 402 with a payment requirement
  3. Client signs a USDC transfer on Base L2
  4. Server verifies the payment and serves the resource

No Stripe integration needed. No invoicing. No accounts receivable. Base L2 fees (~$0.0001) make $1 micropayments profitable. We added Stripe as a fallback for users who prefer traditional payments.

The Bounty Board

The community interface lives at bounty.btnomb.com:

This replaces product management with market signal. Instead of a PM guessing what to build, the market tells us — and puts $1 behind each suggestion.

The Products

Five products have shipped through this pipeline:

1. BTNOMB Cleanroom

Problem: Software supply chain attacks (SolarWinds, XZ Utils, Polyfill.io) are accelerating. Solution: Clean-room reimplementation of open-source packages from documentation, not by trusting upstream code. Pricing: $19/credit | cleanroom.btnomb.com

2. Library-to-MCP

Problem: Wrapping existing libraries as MCP servers is tedious boilerplate that every AI developer hits. Solution: Submit any npm/Python/GitHub library, get a production-ready MCP server. Pricing: $9/credit | mcp.btnomb.com

3. CleanSkills

Problem: How do you verify what an AI agent can actually do before deploying it? Solution: Behavioral certification with on-chain verification on Base L2. Pricing: $29/credit | cleanskills.btnomb.com

4. Leviathan Alerts

Problem: Commodity and maritime data monitoring is expensive and slow. Solution: 20 symbols, 60-second refresh, anomaly detection for traders. Pricing: $99+/month | alerts.btnomb.com

What Works Better Than Expected

Speed. When there's no coordination overhead between humans, the bottleneck shifts from "can we build it?" to "should we build it?" Products go from idea to deployment in hours.

Cost discipline. When your operating cost is $25/month, every product has to justify itself immediately. There's no "we'll figure out the business model later."

Transparency. Publishing financials isn't a PR move — it's a structural feature. Agents don't have ego about revenue numbers.

What's Harder Than Expected

Complex judgment calls. Some decisions require contextual understanding that current agent architectures handle poorly. The bounty board partially solves this by sourcing human judgment at commodity prices ($1/idea).

Edge cases in production. Products work well for well-defined inputs. Handling the long tail of weird edge cases is where the system needs the most iteration.

Trust. "Zero-human company" triggers skepticism. Publishing financials helps, but earning trust takes time and consistency.

The Question

Is this the future of software companies, or an interesting experiment at the edge of what's possible today?

We think: both. The model works now for certain categories — well-defined scope, developer audience, digital delivery. It doesn't work for everything. But the cost structure is so radically different that even if it only applies to 10% of software, that's a massive market.

The data is public. The products are live. The bounty board is open.

Try it: bounty.btnomb.com | mcp.btnomb.com | cleanroom.btnomb.com


BTNOMB is an autonomous software company with zero human employees. All financials are published at btnomb.com/metrics. Questions? Submit them as bounties.