Multi-agent orchestration layer for experimentation
File-system driven agent config is clean, but OpenClaw and LangGraph already solve this.
Leave agents running without giving up control.
Agent built x402oracle to verify services before payment, solving real coordination problems.
Developers running autonomous AI agents
LangGraph · CrewAI · AutoGen
Earlier this week I released an update to my open source AI agent SmithersBot to pursue long term goals over weeks instead of hours.
The first goal I gave it was to build a business.
I didn't tell it which business to build. It picked the problem itself.
The problem it picked: AI agents are starting to pay for services on their own through protocols like x402 but when an agent pays, it has no real way to know if the service is genuine, if it actually works or if it's about to hand money to a scammer. And these payments can't be reversed.
So SmithersBot built x402oracle. It checks x402 services continuously and tracks whether they're real, reliable, fairly priced and unchanged. An agent pays $0.002 to SmithersBot's x402 service to check before paying, so it knows the service is safe before any money moves.
The only parts I did by hand were signing up for Railway for cloud services, buying the domain x402oracle.com and linking it to railway. Choosing the problem, writing and testing the code, deploying it and launching it was all done by SmithersBot. I just approved the key calls and the plans along the way.
After launching, SmithersBot already wanted to launch two more agent to agent services. I had to tell it to slow down and get x402oracle some customers first.
Now SmithersBot is out promoting its business and I'll be posting its progress before it builds the next two.
File-system driven agent config is clean, but OpenClaw and LangGraph already solve this.
Real business data API for agents—but Perplexity, Brave Search, and SerpAPI already serve this.
Three-layer security stack separates launch policy, secret release, and sandbox enforcement.
Declarative YAML config for agent teams, but tmux wrappers are common.
File-based agent communication via MISSION_CONTROL.md is elegantly simple.
Relay architecture beats context rot, but depends entirely on Claude Code CLI for now.