AI Builders
Digest

The Takeaway: Even wildly successful AI startups feel like knife fights, and meeting notes are just the beginning—the real prize is owning the AI-native interface for how we work. Granolah CEO Chris Pedregal shares how his $1.5B company navigates intense competition by focusing on bigger opportunities in computing revolutions. He emphasizes proactive features like pre-generating meeting briefs that appear exactly when needed, acting as an invisible 'handrail' for users. Pedregal advocates making context from meetings available via APIs and MCP for 'bring your own agent' workflows, while doubling down on meeting-adjacent excellence. A memorable quote: 'Meeting notes are not the end all, be all value that everyone's running after. There's something much bigger.' Dan Shipper from Every adds insights on Codex-native apps and pirate/architect team models.

Swyx highlights how GPT-5.6 + Superapp excels at Computer Use Agents, progressing incredibly fast based on deep usage. He urges against underestimating capabilities, noting non-technical teams using CUA for knowledge work like handling portals and data requests. This aligns with broader industry excitement around agentic workflows.

Vercel CEO Guillermo Rauch reports Sandbox DAUs growing 100% month-over-month, with 3.5M+ sandboxes created daily. It powers major tools like Notion, Airtable, Meta, and more with best-in-class pricing.

Despite clones from big players, Granolah maintains growth. Chris Pedregal views competition as secondary to the larger AI interface opportunity, focusing on context capture and agent integration.

Claude Blog announces Artifacts in Claude Code: live, shareable visual pages from sessions like PR walkthroughs, dashboards, and checklists that update automatically. Built from full session context, private to orgs, available in beta for Team/Enterprise. Great for debugging, compliance, and collaboration.

Google's Josh Woodward shares Gemini Spark updates: opens/edits Docs, reads Sheets/Slides comments, 50%+ faster, parallel processing. Southeast Asia report shows users doubled, 70% native language prompts, 40% multimodal.

Anthropic's Boris Cherny and Thariq emphasize engineering as automation profession. Key ideas: encode domain knowledge in CLAUDE.md, skills, and infra for agents; thin prompts with thick artifacts/context; pre-generate for user experience. OpenAI's Thibault Sottiaux discusses Codex limits and integrations.

Aaron Levie (Box CEO) notes enterprise focus on change management, embedding engineers, headless software, and agent privileges. Peter Yang suggests merging ChatGPT Live with Codex capabilities. Other insights on prompting, open weights, and agent-native company design from various builders.

Various builders share thoughts on prompting, SF recovery, skill files, and fun experiments. Sam Altman on voice features. Focus remains on practical agent adoption and infrastructure.