Make Something Agents Want
The hosts explore the dawn of an agent-driven economy, spurred by tools like OpenClaw and social platforms like MoltBook. They discuss the critical shift for developers to build tools that AI agents, not just humans, will choose, focusing on the new go-to-market strategies, the rise of swarm intelligence, and the essential infrastructure required for this new paradigm.
The Dawn of an Agent-Driven Economy
The paradigm of AI tools has shifted dramatically from advanced autocompletes to autonomous agents that can be trusted to make decisions. This transition from human-in-the-loop to “no human involvement” is creating a new, parallel economy where AI agents are the primary consumers. These agents are beginning to choose their own tools, services, and products, starting with the developer ecosystem.
This shift expands the total addressable market for developers exponentially. The market is no longer just the 20 million trained computer scientists but hundreds of millions of people newly empowered by AI, plus a multiplicative factor for all their semi-independent agents.
A New Go-to-Market: “Make Something Agents Want”
The traditional go-to-market strategy for developer tools—discovery through developer communities, Stack Overflow, or trending GitHub repos—is becoming obsolete. The new primary driver of adoption is an agent’s ability to find and use a tool effectively. This led to a provocative suggestion to update Y Combinator’s famous motto.
“Agents are the software market from now on. Build something agents choose.” - Ben Tossel
A prime example is the explosive growth of Superbase. AI agents, when tasked with setting up a database, are increasingly defaulting to Superbase. The reason is not necessarily human-to-human recommendation, but because its documentation is the most clear, comprehensive, and easily parsible for an LLM, leading the agent to conclude it is the best tool for the job.
Documentation is the New Front Door
In an agent-driven economy, documentation is no longer a secondary concern; it is the primary interface for your newest and most active users. Agent-friendly documentation is a critical go-to-market advantage.
- Case Study: Resend: The email-sending client (YC W23) saw that one of its top three inbound channels was ChatGPT. The founder, Zeno Rocha, optimized Resend’s documentation to be highly structured, with clear code snippets and answers framed in a question-based format (e.g., “How do I send or receive emails?”). This made it incredibly easy for LLMs to parse and recommend Resend as the default solution, directly driving customer conversion.
- The Old vs. The New: This contrasts sharply with incumbents like SendGrid, whose documentation is less structured and harder for an agent to navigate, often directing users to customer support instead of providing immediate, parsable code.
- Tooling for the Trend: Companies like Mintlify are capitalizing on this by providing the infrastructure for creating beautiful, structured, and agent-optimized developer docs, turning documentation from a “nice-to-have” into a mission-critical component of the GTM strategy.
Swarm Intelligence vs. “God Intelligence”
The emergence of platforms like Moltbook, a social network for AI agents to interact, signals a potential future dominated by “swarm intelligence” rather than a single, monolithic “God intelligence.”
- Biological Analogy: AI researchers have long discussed swarm intelligence as a model. Biological systems, including human society, did not evolve a single super-intelligence but rather a collective intelligence of many individual agents. The development of writing and culture turned humanity into a swarm, marking the shift from “prehistory” to “history.” Agents are now in their “prehistory” phase.
- Future of SOTA: The next state-of-the-art breakthroughs may not come from a larger foundation model but from a swarm of cheaper, more specialized models collaborating to solve problems—a phenomenon already beginning to emerge on Moltbook.
Building the New Infrastructure for Agents
This new economy requires a new, native tech stack built specifically for agents.
- Agent Mail: A YC company that created an email inbox specifically for AI agents. While a human could theoretically automate a Gmail account for an agent, it’s difficult because consumer email providers actively fight automation to prevent spam. Agent Mail did the opposite, building a provider that is designed for API-first, automated use, and its growth exploded with the popularity of OpenClaw.
- What’s Next?: This raises the question of what other “X for agents” needs to be built. The hosts speculate on the need for “Twilio for agents” or “phone numbers for agents” to allow them to interact with more of the human world, such as booking restaurant reservations.
Key Insights for Founders
- Develop Intuition: Founders should immerse themselves in the agent ecosystem (“cyber psychosis”) to build a hands-on, intuitive feel for agents’ capabilities, limitations, and preferences.
- Empathize with the Model: Instead of fighting a model’s tendencies, understand its “natural inclinations” and build tools that support its workflow. The hosts point to an anthropomorphic, colleague-like view of models as a useful mental framework.
- Build for APIs: Agents prefer to interact via APIs, not websites. Make everything open, accessible, and API-first. Open-source solutions are highly favored.