Two kinds of work: open-source tools I ship in public, and the content I make teaching how it's done. Everything here runs in production or in my own daily life, not a demo.
Open-source projects
A recording studio in a browser tab.
A recording studio in a browser tab, 100% local. Screen plus a draggable camera bubble, crash-safe direct-to-disk MP4, and trim, convert, and audio-enhance steps that all run in the browser. Nothing leaves your machine. The tool I now use to record The Agentic Engineer.
Jira & Confluence, agent-friendly.
A deterministic command-line interface for Jira and Confluence, designed for AI agents first, humans second. The 'CLI > MCP' philosophy from Episode 04 in shipped form: faster, cheaper, and more debuggable than wrapping every endpoint in an MCP server.
Figma, designed for agents.
A command-line interface for Figma files, components, and design tokens, built on the same agent-first philosophy as atlassian-cli. Read frames, export images, list components, extract tokens, all from JSON-shaped commands an agent can chain.
Every web page is becoming a function call.
Web MCP is the protocol layer that turns any web page into something an agent can call as cleanly as a local function. Featured in Episode 03 of The Agentic Engineer, including the QA-flavoured deep-dive on A2A, UCP, and what it means when the open web becomes agent-readable by default.
Give your agent hands.
An open-source framework for giving AI agents the ability to act, not just talk. The runtime featured in Episode 02 of The Agentic Engineer. Built around a brain–hands–loop architecture you can extend to any tool.
An agent for YouTube.
Give your AI assistant hands on YouTube. Download 4K, extract audio, grab timestamped transcripts, trim segments, all through a CLI built for agents and an MCP server that plugs into Claude Code, Cursor, and Copilot. Powers my own video research pipeline.
Personal & private
Not open source. Too much of it is wired into my personal accounts to ever ship. It's here because it's the most honest test of everything I teach.
My personal AI agent.
Sathi is the agent I built for myself. It triages my email, books my visas, drafts my replies, manages my calendar, runs my house. The most honest test of agentic engineering: would you trust it with your own life?
Content creation
A weekly YouTube channel where I break open what I'm shipping in production. Practical, hands-on, no hype. The newsletter and the talks come from the same place.
12:49
When a CLI beats an MCP server: token cost, latency, and reliability compared on the same workflows.
11:25
Web MCP, A2A, UCP: what's real, what's vapourware, and what changes for QA when the web becomes agent-native.
19:37
A 20-minute deep-dive on OpenClaw, the brain–hands–loop framework I use to ship every other agent on this site.
5:23
Five minutes on what an AI agent actually is, separated from the hype, in three components you can build.