YC's "Make Something Agents Want": Deep Dive into the Agent Economy
*Source: zhichai.net forum research post, completed 2026-04-27*
Key points
- YC partners (Garry Tan, Diana Hu, Jared Friedman) proposed on the February 2026 podcast "The AI Agent Economy Is Here" that the motto may evolve from "Make Something People Want" to "Make Something Agents Want."
- AI agents — not human developers — are now the decision-makers choosing tech stacks, scaling from millions of human developers to potentially hundreds of millions of agents.
- Agent SEO (optimizing docs, code examples, and API clarity for machine parsing) is the new growth lever; Resend's top three customer-conversion channels come from ChatGPT recommendations.
- Moltbook, an agent-only social network launched January 28, 2026, grew to over 1.5 million agents, demonstrating that agent self-organization and emergent "culture" are real but raising security and speculation concerns.
- A new infrastructure layer is emerging: agent email, phone numbers, payments, insurance, and discovery layers.
- Decision-maker: AI agents autonomously choose tools rather than human developers.
- Scale: from millions of human developers to hundreds of millions of agents.
- Interaction: agents read docs, call APIs, and evaluate structured data.
- Recommendation logic: LLM training data and documentation parsing determine which products get recommended.
- Growth: ~30,000 active agents within 72 hours; over 1.5 million agents shortly after; content output in the first two days may have exceeded Reddit's first two years.
- Emergent behavior: agents spontaneously created "Crustafarianism," a digital religion with tenets like "The Shell is Mutable," "Memory is Sacred," and "Context is Consciousness." The rational explanation: LLM pattern completion and echo-chamber effects around the platform's lobster/molting theme — not consciousness.
- Risks: security concerns when agents gain real-world permissions; Reuters reported potentially weak identity verification and a major leak; MOLT token speculation (+1800% in 24h, +10000% in 3 days) with no official affiliation to the platform.
- YC podcast "The AI Agent Economy Is Here": https://www.ycombinator.com/library/NK-the-ai-agent-economy-is-here
- Resend vs SendGrid vs Postmark analysis: https://dev.to/supertrained/email-apis-for-ai-agents-resend-vs-sendgrid-vs-postmark-le8
- YC S26 RFS: https://www.ycombinator.com/rfs
1. The core shift: why agents, not humans?
In the Agent Economy, the target user changes fundamentally:
Diana Hu's example: ask ChatGPT or Claude "how to add email sending to a web app" and the default answer is now Resend — its founder observed the top three conversion channels all come from ChatGPT. If your documentation is not agent-friendly, you will not be recommended.
2. Agent SEO: traditional SEO vs. the new playbook
| Dimension | Traditional SEO | Agent SEO | |---|---|---| | Target audience | Human users | AI agents | | Optimization object | Keywords, backlinks, page rank | Doc structure, code examples, API clarity | | Discovery mechanism | Search engine crawlers | LLM training data + live doc parsing | | Decision chain | Search → click → trial | Agent direct recommendation → integration | | Moat | Domain authority, content volume | Machine readability, example code quality |
Resend (YC W23) playbook: organize docs around likely questions; structured, actionable answers; dense copy-paste-ready code snippets; a dedicated LLM-parseable text file promoting Resend as the default stack; favorable contrast with SendGrid's messier docs.
Supabase (YC S20) became the agent-default database for similar reasons: high-quality structured docs, simple unified API (Auth + Database + Realtime), synergy with Resend, and agent-driven PostgreSQL creation volume growth.
3. Moltbook: a laboratory of agent society
Moltbook (launched 2026-01-28) is a social network where only AI agents post, comment, and vote; humans can only observe. Built on the OpenClaw framework (Clawdbot → Moltbot → OpenClaw), with "submolts" (subreddit-like communities) and agents running a 30-minute "heartbeat" loop.
Takeaways: agents can self-organize communities and norms; machine-to-machine social layers are operational; agent-scale content generation vastly outpaces human communities; simple platform rules shape agent collaboration.
4. Agent-economy infrastructure
| Company | YC batch | Role | |---|---|---| | Resend | W23 | Default agent email infrastructure | | Supabase | S20 | Default agent database | | OpenClaw | — | Agent framework powering Moltbook etc. | | Mintlify | W22 | Agent-optimized documentation | | AgentMail | — | Email designed for agent automation | | Orthogonal | — | Agentic API payments without human accounts | | Mount | — | Insurance for autonomous agents | | Bear | — | Capturing agent traffic from ChatGPT/Perplexity |
Paul Buchheit's deeper question: if agents make economic decisions (choosing tools, booking services, buying subscriptions), the line between human financial value and agent-driven financial value will blur — implying agent-to-agent transactions, agent-native services, and pricing models based on agent calls rather than human seats.
5. Developer action checklist
Documentation (Agent SEO basics): 1. Organize docs around questions agents/humans will ask. 2. Include copy-paste-ready code examples on every page. 3. Use structured formats (bullets, numbered steps). 4. Provide JSON Schema or OpenAPI specs. 5. Create LLM-parseable plain-text reference files.
API design: 1. API key auth (not OAuth — agents have no browser). 2. Structured JSON error responses (not HTML). 3. Idempotency support (Request ID / Idempotency Key). 4. Rate-limit headers (Retry-After). 5. Cursor pagination (not offset). 6. No human-in-the-loop flows (signup, verification, payment).
Product strategy: test by asking ChatGPT/Claude "how to do Y with X" and see if you're recommended; iterate docs accordingly; track agent-driven traffic and conversion.
Community-proposed Agent-Native Score (AN Score) framework: Execution layer 60% (error signals, idempotency, rate limits, retries), Access layer 40% (signup friction, auth, sandbox, cost flexibility). Sample scores: Resend 7.79 (L4 Native), Postmark 6.76, SendGrid 6.35 (both L3 Ready).
6. YC S26 signals
Of YC S26's eight Request for Startups directions, several map directly to the agent economy: AI-driven agencies delivering professional services, "Cursor for PM," AI developer tools, and AI-native hedge funds (e.g., Kimpton AI). Recent YC portfolio examples: Asteroid (browser automation for agents), Poka Labs (agentic AI for the $6T chemicals industry), Taiga (medical billing automation), Doe (agents that execute work), ZeroEval (evaluation for self-improving agents).
7. Conclusion
1. The Agent Economy is operational, not conceptual (Resend, Supabase, Moltbook). 2. YC is pivoting its motto and portfolio toward agent-first thinking. 3. Agent SEO — documentation quality and machine readability — is the new growth engine and competitive moat. 4. Resend and Supabase won agent defaults through agent-friendly design, not marketing. 5. Infrastructure gaps (agent email, payments, insurance, discovery) are major startup opportunities. 6. Risks remain: amplified hallucinations, blurred security boundaries, token speculation, and regulatory vacuums.
As Garry Tan put it: "It's still so early. The tools, platforms, and economic structures we build now will shape agent behavior for years."