Notes from the foundry
Engineering essays on generative AI, retrieval systems, and what it takes to ship intelligent software to production.
Creating an Autonomous Coding Agent using Vercel AI SDK and MCP
Learn how to build your own AI software engineer. We use the new Model Context Protocol (MCP) to give the Vercel AI SDK agent access to read your local file system, write code, and execute bash commands safely.
Letting an Agent Run Shell Commands Safely with Vercel AI SDK v7 Sandboxes
I built a dependency-audit agent that runs npm commands on real projects. The new experimental sandbox support in Vercel AI SDK v7 is what made me comfortable shipping it.
Building a Real-Time Voice Agent with Vercel AI SDK and LiveKit
Voice is eating text. Learn how to wire the Vercel AI SDK's streaming capabilities with a WebRTC backend to create a low-latency conversational voice agent that handles natural interruptions.
Building a Browser-Controlling "Computer Use" Agent with Vercel AI SDK
Chatbots are limited. Learn how to connect the Vercel AI SDK to Playwright to build an agent that navigates the web, clicks buttons, and scrapes data autonomously.
Building a Sub-Second Voice Agent with Gemini Omni Flash
Voice agents die if they have latency. I used Gemini Omni Flash to build a real-time conversational agent that responds in under 500 milliseconds.
Designing a Multi-Agent Swarm with Vercel AI SDK for Complex Workflows
Single-prompt chatbots are dead. Learn how to orchestrate a 'Swarm' of specialized agents that seamlessly hand off tasks and state to each other using the AI SDK's tool-calling capabilities.
Extracting Structured JSON from Video Feeds with Gemini Omni Flash
Processing video usually requires massive compute overhead. I used Gemini Omni Flash to extract structured JSON data directly from raw video feeds in real time.
Building a Persistent Data Analyst Agent with Vercel AI SDK and E2B
Giving an LLM the ability to write code is dangerous. Learn how to securely connect the Vercel AI SDK to an E2B cloud sandbox so your agent can write Python, execute it on CSVs, and generate charts safely.
Building a Resilient Support Bot with Vercel AI SDK v7 Durable Workflows
Process crashes used to mean lost context for AI agents. I tested the new WorkflowAgent in Vercel AI SDK v7 to see if it actually survives mid-thought restarts.
Deploying Gemini Omni Flash at the Edge for Zero-Latency RAG
RAG pipelines are slow because of server round-trips. I moved a Gemini Omni Flash workflow entirely to Edge Functions to kill the cold start.
Debugging Agents Locally with the Vercel AI SDK v7 Terminal UI
Console logging agent tool calls is a nightmare. I set up the new Terminal UI in Vercel AI SDK v7 to visualize a local coding assistant.
Securing Database Writes with Vercel AI SDK v7 Tool Approvals
Giving an LLM permission to delete database rows is terrifying. I used the new human-in-the-loop tool approvals in Vercel AI SDK v7 to build a safer admin agent.
Processing Video Files with Vercel AI SDK v7 Multimodal Support
Language models can finally see more than just text. I tested the new multimodal capabilities in Vercel AI SDK v7 to analyze raw video files.
Finding Slow Tool Calls in Vercel AI SDK v7 Telemetry
When an agent takes ten seconds to respond, you need to know why. I used the new OpenTelemetry support in Vercel AI SDK v7 to track down a slow API call.
The ROI of AI in 2026: Stop Building Demos
Founders are burning cash on flashy AI features. I'm seeing the actual returns come from boring backend automation.
Integrating Claude Code with Vercel AI SDK v7
You do not have to abandon your existing agent frameworks to use the new SDK. Here is how I brought Vercel AI SDK v7 into a custom Claude Code harness.
Controlling Model Think Time with Vercel AI SDK v7 Standardized Reasoning
Different models handle 'reasoning' in entirely different ways. Vercel AI SDK v7 normalizes this, letting you explicitly dial up the think time for complex agents.
Why Your Internal AI Project is Stalled
Smart engineering teams struggle to ship AI because traditional software practices don't map to probabilistic models. Here is what breaks.
When to Buy AI and When to Build It
Founders waste money building custom AI for generic problems. Here is how to decide when to pay for SaaS and when to hire engineers.
Connecting Local Databases with Vercel AI SDK v7 MCP Apps
Model Context Protocol is finally starting to feel accessible. Vercel AI SDK v7 integrates MCP out of the box, so I used it to connect an agent to a local SQLite file.