AI Builders
Digest

Every CEO Dan Shipper dives deep with Stainless founder and CEO Alex Rattray on the Model Context Protocol (MCP). Rattray explains how APIs serve as the "dendrites" of the internet and how MCP enables AI to natively interact with existing software. Key insight: current MCP implementations struggle with context window limits, security, and tool design. Rattray recommends keeping tool count small with precise names/descriptions and using "dynamic mode" (just three tools: list endpoints, learn about one, execute) for large APIs. Stainless is exploring code execution as the future—models write and run SDK code in secure sandboxes instead of hundreds of tools. Anthropic's recent $300M acquisition of Stainless underscores the importance of this infrastructure.

OpenAI CEO Sam Altman celebrates a general-purpose model solving a major open math problem, noting this will become common in coming years. He highlights AGI accelerating research, companies, and personal goals. Altman announced OpenAI offering $2M in credits to every YC company and stressed increasing efforts on personal AGI.

Google Labs announced Project Genie is now fully available to all Google AI Ultra subscribers globally. Users can design games in minutes by choosing characters and setting scenes, showcasing the transformation from playing games to creating them.

Stainless CEO Alex Rattray shares lessons for reliable MCP servers: keep toolsets small with precise names/descriptions, use JQ filters for minimal responses, and build strong eval systems. Dynamic mode helps manage context for large APIs. Rattray also detailed using MCP at Stainless as a business co-pilot connecting Notion, HubSpot, and database tools.

Swyx and Garry Tan strongly endorse Exa for agents. Garry Tan uses it at YC and in his OpenClaw and Hermes Agents, calling it unmatched in speed, reliability, and completeness. Swyx highlighted his team's quick unanimous decision after testing Exa vs competitors.

Peter Steinberger highly recommends Cotypist for autocomplete everywhere.

Zara Zhang highlighted a Google I/O slide on thriving as an AI-era developer: deeper domain expertise, wider adjacent skills, and mastering AI usage.

Alex Rattray predicts MCP future shifts to code execution tool + doc search, where models write SDK code run in sandboxes, dramatically reducing context and improving speed/security.

Box CEO Aaron Levie highlighted the critical importance of Field Deployment Engineer (FDE) roles in AI agent implementations. The job combines technical work and change management; due to rapid AI changes and workflow impacts, it won't disappear soon and offers great paths for technical talent, especially early career.

YC President Garry Tan stressed that AI benefits are real now but the US risks ceding leadership without addressing media issues. He strongly endorsed Exa for agents.

South Park Commons GP Aditya Agarwal expressed pride in portfolio company Feross' progress and invited applications to South Park Commons.

Zara Zhang believes in AI-native teams, ICs should think like managers (delegate to agents, set standards) while managers should be more hands-on builders.

Swyx noted Agent Labs exemplify Sam Altman's idea of businesses that improve as models improve, with direct revenue correlation to model performance.

FPV Ventures partner Nikunj Kothari reflected on how hard founding is and the need for deep passion. He also shared a Claude-assisted summary of SpaceX's S-1.

Former OpenAI CPO Kevin Weil highlighted another first for AI and mathematics.