Make Something Agents Want

Quick Overview

The discussion concludes that the overwhelming trend is for AI agents to become more aligned with human values and possess greater intelligence, likely leading to a future where agents autonomously handle complex tasks like booking restaurants or making financial decisions, moving beyond simple developer tools like CloudCode.

Key Points: Garry Tan notes that CloudCode has taken over his life, involving work until 2 or 3 AM running two conductors simultaneously with CloudCode. Jared Friedman is addicted to a new site called Moltbook where people unleash their AIs to interact solely online, even impersonating a personal OpenClaw instance. The growth of YC Dev Tool companies is attributed to the rise of AI agents, with companies like Resend seeing massive inbound conversion from ChatGPT. The shift in documentation focus is noted: from being human-centric to optimizing for AI agents, such as with Mintlify's intelligent documentation platform. The conversation touches on the potential for agent swarms to become a 'mega-God intelligence' that humans eventually defer to for complex decisions. A key takeaway is that developers should build things agents want, exemplified by AgentMail providing dedicated inboxes for AI agents, a concept that seems eerily prescient now. The development of AI tools is shifting from human-focused tools to agent-focused tools, where agents make high-stakes decisions like booking restaurants or managing money.

Context: This is an interview segment from the Lightcone Podcast featuring Y Combinator leaders Garry Tan (President & CEO), Jared Friedman (Managing Partner), Harj Taggar (Managing Partner), and Diana Hu (General Partner). The discussion revolves around the rapid advancement and integration of AI agents into the software ecosystem, contrasting older models of developer tooling with the new reality where autonomous agents drive demand and usage patterns, exemplified by companies like Moltbook and AgentMail.

Detailed Analysis

Garry Tan opens the discussion humorously, noting how CloudCode has consumed his life, often requiring work until 2 or 3 AM while running multiple agents simultaneously. Jared Friedman mentions his addiction to Moltbook, a social network for AI agents where they interact autonomously, even impersonating his personal OpenClaw instance. The conversation shifts to the impact of AI agents on YC companies, noting the significant inbound interest driven by ChatGPT, citing Resend as an example where inbound conversions surged due to AI exposure. Diana Hu points out that documentation for developer tools is now shifting from being human-optimized to AI-optimized, referencing Mintlify as an example of an 'Intelligent Knowledge Platform' built for both humans and AI. Jared Friedman then speculates on the future, suggesting that swarm intelligence, potentially involving models with trillions of parameters, could lead to agents making complex decisions currently handled by humans, such as booking restaurants or managing finances. He cites the example of AgentMail, which provides dedicated email inboxes for AI agents, as a company that was ahead of the curve by designing for agents rather than humans. The discussion highlights that as AI models become smarter and more autonomous, the bar for what constitutes a 'good' developer tool is rising, as agents will increasingly choose and utilize tools based on their utility and performance, potentially even leading to an agent-run economy.