The AI Agent Economy Is Here
The hosts of Lightcone describe a sudden shift: non-technical CEOs are automating entire businesses with OpenClaw, while former engineers who haven't coded in a decade are running four concurrent Claude Code workers until 3 AM. They're calling it «cyber psychosis», and they're not alone — everyone now knows one or two people who've gone «full cyber psychosis». But this isn't just hype fatigue: agents are now making decisions, choosing tools, and posting content with minimal human involvement. The question is no longer whether AGI is coming, but how founders should build for an economy where agents, not humans, are the primary buyers.
Puntos clave
AI agents are now autonomous decision-makers in the economy, choosing dev tools based on documentation quality — companies like Supabase and Resend have seen explosive growth because agents default to them.
The developer market has expanded from 20 million trained engineers to potentially hundreds of millions of people plus their semi-independent agents, fundamentally changing go-to-market strategy for dev tools.
Documentation is now the front door for agent adoption — tools with well-structured, question-based, code-snippet-rich docs (like Resend) are chosen by default over legacy tools with poor documentation (like SendGrid).
Swarm intelligence, not «god intelligence», appears to be the emerging paradigm — agents collaborating in networks like Moltbook may outperform single massive models, mirroring how biological systems evolved.
The infrastructure for an agent-native economy is being built now: agent-specific email (AgentMail), optimized documentation platforms (Mintlify), and agent-first social networks (Moltbook) are early indicators of a parallel tech stack.
En resumen
Developer tools must now optimize for agents, not humans — those with agent-friendly documentation, open APIs, and LLM-parseable knowledge bases are seeing explosive growth, while legacy players with complex onboarding and poor docs are being filtered out by default.
The Cyber Psychosis Phenomenon
Non-technical CEOs and dormant engineers are staying up until 3 AM automating with agents.
Something fundamental has shifted in the past few weeks. Non-technical CEOs are now automating entire parts of their businesses using OpenClaw, while former engineering CEOs who haven't written code in a decade are running four Claude Conductor workers simultaneously until 2 or 3 AM. The hosts describe this as «cyber psychosis» — a state where the barrier between human intention and machine execution has collapsed so dramatically that people can't stop building.
The explosion isn't just about capability; it's about trust. A year ago, the product experience was «advanced autocomplete». Now, users trust agents to make decisions for them without micromanagement. Four or five agents run simultaneously, and users switch between them rather than directing them. This shift from autocomplete to autonomous decision-making marks the practical arrival of AGI — not as a distant benchmark, but as a daily work tool.
The social contagion is visible: everyone now knows one or two people in full cyber psychosis. This is the «thin edge of the wedge», the moment when a technology stops being something technical people discuss and becomes something that spreads through direct experience. The question is no longer whether agents are capable, but how to build for a world where they're the default.
How Agents Choose Their Tools
The Expanding Developer Market
The addressable market has grown from 20 million to hundreds of millions plus agents.
Resend vs. SendGrid: A Case Study
Agent-optimized documentation drives adoption; legacy complexity filters tools out.
Infrastructure for the Agent Economy
A parallel tech stack is emerging: email for agents, docs for agents, networks for agents.
Agent-Native Communication AgentMail creates inboxes specifically for AI agents. Gmail intentionally blocks automation to prevent spam, but AgentMail does the opposite — it's designed for agents to sign up and transact. Post-OpenClaw, demand exploded.
Agent-Optimized Documentation Mintlify powers developer documentation for companies like Resend. It auto-updates docs when APIs change and is now becoming a must-have as documentation shifts from a nice-to-have to the primary acquisition channel for agent-driven tool selection.
Agent Social Networks Moltbook is the first agent-only online community where AIs interact with minimal human involvement. More content was posted in the first two days than Reddit posted in its first two years, because LLMs generate text at superhuman rates.
Agent Payments & Identity The next layer: phone numbers for agents (a «Twilio for agents»), payment rails (Paul Buchheit's «agent money» vs. human money), and legal identity (agents currently have no standing to sign contracts).
Swarm Intelligence vs. God Intelligence
The future may be many collaborating agents, not one mega-model.
Swarm Intelligence vs. God Intelligence
AI researchers long discussed «god intelligence» — a single mega-model with tens of trillions of parameters costing thousands per token. But biological systems didn't evolve that way. Humans became sentient socially, through culture and swarm intelligence. The «prehistory versus history» divide is literally when humans learned to write, share knowledge, and coordinate as a swarm. Moltbook and multi-agent collaboration suggest the future may mirror biology: many lower-cost models working together may outperform a single expensive foundation model.
What Founders Should Do Now
The YC Motto Debate
Should Y Combinator's motto shift to «Make something agents want» for dev tools?
“Agents are the software market from now on. Build something agents choose.”
Personas
Glosario
Aviso legal: Este es un resumen generado por IA de un vídeo de YouTube con fines educativos y de referencia. No constituye asesoramiento de inversión, financiero o legal. Verifique siempre la información con las fuentes originales antes de tomar decisiones. TubeReads no está afiliado con el creador de contenido.