Vercel — searchable YouTube transcripts & AI Q&A

Fast AI panel: A conversation with Cerebras, TypeSafe AI, and Vercel

The panel argues that fast inference is crucial for making AI genuinely useful, as speed enables real-time interaction and opens up new markets. They highlight that sub-100ms response times allow for dynamic UI generation and on-the-fly program creation, though reliability remains a key challenge. The speakers also emphasize that while hardware improvements like Cerebras's chips are important, software innovation and exploring diverse AI paths are equally vital for future progress.

The Third Customer: A Fireside with Tobi Lütke and Guillermo Rauch

Tobi Lütke argues that AI has reached a point of maximum optimism, where models like Opus 4.5 have made previously impossible technical tasks achievable, and he now sees himself as a verification engineer rather than a hand-coder. He reports that Shopify's internal AI agent, River, writes over 50% of code and that the company celebrates code deletion, with a leaderboard where River dominates. He also predicts agentic commerce will replace 20-50% of traditional commerce while growing the overall pie, and emphasizes that undifferentiated commerce will struggle while unique, thoughtful experiences thrive.

ShipSF Workshops - Embed Generative UI Agents in Your App with Codex and Vercel

The speakers demonstrate how to embed a generative UI agent into an existing application using OpenAI's Codex and Vercel's sandbox and drive services. They show that the agent can dynamically generate UI elements, personalize recommendations based on stored user preferences, and be deployed to Vercel. The process can be generalized into a reusable skill for agentifying other applications.

ShipSF Workshops - Build an agent that plays a game for you

Vercel now supports long-running back-end workloads like Python servers through Fluid Compute and Demons, enabling complex applications such as a collaborative Jupyter notebook environment or a game server. The speaker demonstrates how to build an agent that plays a game by deploying a Python backend on Vercel, using services for atomic deployments and private networking. He argues that serverless architecture is superior for scalability and recommends deploying back ends on Vercel.

Building the Future of Data Infra on the Back of Open Source

ClickHouse's co-founder argues that a fully featured open-source product is the best driver of demand for a commercial cloud offering, not a liability. By separating compute and storage in the cloud, ClickHouse Cloud can be cheaper than self-managed open source, making migration a no-brainer. He also outlines the company's evolution into a universal data platform, emphasizing the importance of a platform approach for integrating acquisitions and partnerships efficiently.