
Full recording, 1h 19m. Plays from youtube-nocookie.com. Watch on YouTube
Jump to a chapter (28 chapters)
- Welcome and plan for today
- Intros: how everyone uses AI
- Community demo: an AI Mandarin speaking tutor
- The dream: "Claude, earn me $100 while I sleep"
- Diego joins: 15 years in IT, from Argentina
- Paxel: a profile of how you build with agents
- Open lab: Claude Code sessions as constellations
- The same idea for LangGraph and LangSmith
- How a lab wins clients: show your skills
- Story: AI as CTO and marketing team for a small business
- Agentic protocols lab: MCP, agent-to-agent, agent-to-business
- x402 pay-per-call lab
- Agentic Academy: free, open agent courses
- "Vibe coding" is now agentic engineering
- Diego's startup: a cyborg company run by agents
- Human in the loop: approvals with passkeys
- Q&A: guardrails when agents move money
- How L402 could fit a cyborg company
- Live demo: an agent pays for data over Lightning (L402)
- Use cases for agent payments
- How L402 works: a robot at the shop
- Will the agent economy beat the human economy?
- Q&A: is anyone using L402 for real yet?
- A Nostr meme bot that earns zaps
- Learning resources and the tools Diego uses
- How Code Orange uses Claude
- How to reach Diego
- Wrap-up and next sessions
TL;DR
- Agents can now hold and spend money: on the call an agent bought five pieces of live Bitcoin data for 10 sats each over Lightning, using L402.
- Diego's "cyborg company" puts AI agents at desks in a 3D office. They raise a hand when they need a human, and the human signs off with a passkey.
- Show your work to win work: an open lab of experiments is Diego's portfolio, and one small-business rebuild became a case study for the next client.
- L402 is still experimental. Homework: build something small that uses it and show it next month.
Homework for next month0 / 3
Mission board0 XP
Two badges to earn on this page
Play the agent that pays, then play the human who has to say yes or no.
Highlights
1Diego's open lab: agent sessions you can read like a night sky
Observability, made visual. Guest speaker Diego (Argentina, 15 years in IT) built everything he showed with Claude Code. His open lab takes the telemetry a coding agent already produces and draws each session as a galaxy: the orchestrator in the middle, sub-agents around it, tools and hooks firing as lights between them. Colours mean something. A lot of red means too many guardrails are firing and the work is stuck. There is a second version for LangGraph.
Inside the lab
How it pays off: the lab is free. Clients see it, then ask for the same thing inside their own company. “You have to give something away in order to have something in return.”
2A cyborg company: agents at desks, humans accountable
A company you can walk through. Diego's startup shows a business as 3D buildings with departments inside. Each desk has an agent with a face and a job. You can see what it is doing, talk to it by voice, and read reports by department. The goal is to let non-technical people, and people with disabilities, work with agents.
Human in the loop
Guardrails today: hard budget caps per department, and hooks that stop an agent when a rule is broken (yellow is mild, red is severe). Agents do not send payments yet. The money they spend is model tokens.
Mini-game 1✋ Human in the loop
An agent raised its hand. Approve or stop it?
Four requests. Get them right to earn the badge.
3An agent pays for data over Lightning
Real sats, live. keypleb vibe coded a small demo before the call. The agent's job: write a live Bitcoin network briefing. To do it, it needs five answers (block height, fees, price, mempool, hashrate) and each one costs 10 sats. A phone wallet paid a 50-sat invoice, and the agent bought each answer by itself. No account, no card, no API key.
L402: a robot walks into a shop
Try the real one: codeorange.dev/agent-pay works with real sats or in practice mode.
Mini-game 2⚡ Paying agent
You are the agent. Buy the five answers.
A practice run of the demo. No real sats, and the answers are sample values.
Pick a question. Each call costs 10 sats.
“I feel like it's more experimental, and it's up to us to make it mainstream.”
keypleb, on L402
A meme bot that earns zaps. The last demo was a Nostr account that posts memes made by an agent, with a Lightning address on its profile. If a post makes you laugh, you zap it and the sats land in an ecash wallet. Not L402, but the same idea: value for value, with a bot on one side.
Key takeaways
1. Agents can already hold and spend sats. Paying per call fits how agents work: model calls, data feeds, web search, compute, paywalled content, even a tiny fee to keep spam away.
2. Keep a human accountable. Budget caps, hooks that block broken rules and a signature for anything that matters. The more an agent can do, the more this counts.
3. Show your work to win work. Diego's lab is his portfolio. keypleb rebuilt the website, a dashboard and the ads for a small business with an agent, and that story is now the pitch for the next client.
L402Lightning Labs. HTTP 402 plus a Lightning invoice. Paid in sats.
x402Same 402 idea, paid with stablecoins. Diego's lab runs it on a testnet.
ZapsTips on Nostr over Lightning. What the meme bot earns.
Test yourself
Resources shared
Tools
- Agent pay demo, Code Orange: the demo from the call: an agent buys five pieces of live Bitcoin data for 10 sats each over Lightning, with real sats or in practice mode
- Paxel, Y Combinator: free tool that reads your local Claude Code, Codex or Cursor sessions and gives you a profile of how you build with agents
- Claude Code, Anthropic: the coding agent Diego used for everything he showed, about 95% of the time in the desktop app
- LangGraph, LangChain: agent framework; Diego built a second constellation view on top of its tracing tool, LangSmith
- OpenRouter: one API for many models, with free tiers; used as a fallback in Diego's projects and for the Mandarin tutor
- Antigravity, Google: agent-first code editor one developer uses for projects he does not plan to maintain by hand
Reading
- L402, Lightning Labs: the protocol behind the demo: an HTTP 402 "payment required" answer, a Lightning invoice, and a token that proves you paid
- Lightning agent tools, Lightning Labs (2026): "The agents are here and they want to transact": the toolkit that Pi402 and the Wavelength wallet build on
- x402: pay-per-call for agents over HTTP 402 with stablecoins; Diego's lab runs it on a testnet next to MCP-style protocols
- Model Context Protocol (MCP): the standard way to give an agent tools; one of the protocols animated step by step in Diego's protocols lab
Wallets
- Minibits: ecash wallet with a Lightning address; it receives the sats in the agent pay demo and the meme bot's zaps
- Cashu.me: ecash wallet in the browser, used for the 21-sat zap test
- Wallet of Satoshi: the phone wallet that paid the 50-sat invoice live on the call
Code Orange
- BOSS Challenge: free Bitcoin open-source challenge; the next one starts in January
- Agentic Engineering workshop: every 3rd Wednesday, 12:00 UTC on Discord. Build something and show it next month
- Decoding Bitcoin cohort: Mondays 7pm UTC+8 on Discord, join any week
Also on the call: a community member showed a voice tutor she built to practise Mandarin, with ChatGPT for the logic, Gemini for the design and free model keys from OpenRouter. Ideas for next time: pay 21 sats over L402 to join a workshop, a paywalled Substack-style page on Nostr, and an L402 experiment inside Diego's cyborg company with Code Orange as a design partner.
These notes are also on GitHub: code-orange-dev/session-notes. Free to share and reuse (CC0).