Prateek Karnal

Hi, I'm Prateek Karnal.

AI infrastructure · agentic orchestration · decentralized systems · building in public

I work on AI infrastructure at Composio, where I'm responsible for our integrations layer — making sure we cover every possible one and that they're actually complete and high quality. I also created Agent Orchestrator, an open-source system for managing 20+ parallel coding agents on your codebase. It sounds excessive until you try it. If you would like to contribute, join our community on Discord.

Outside of work, I run a system of AI agents that handles my actual life — finances, email, trading, memory, relationships. Not a demo. Real infrastructure. When my agent parses a bank statement at 3 AM and tracks my net worth, that's the part I find interesting.

AI is becoming a collaborator, not just a tool. The gap between "I use ChatGPT sometimes" and "I have agents running my life" will be the defining divide of the next decade. I write here because building in public is the only honest way to figure this out.

Meta note: this site is itself maintained by an agentic system. It wrote the blog posts, deployed the pages, set up the SEO, and is probably the reason you're reading this right now. I just talk to it over Telegram. I wrote about how it works.

Oh, and if any of this resonates — we recently raised a $25M Series A from Lightspeed and our customers range from YC startups to Zoom. We're hiring in SF and Bangalore if you want to come build with us.

You're welcome to hang around and read my writing — it's all public. Some parts of this site are private though — investment dashboards, travel plans, personal research. If you're curious, you can request access and I'll let you in.

Recent Posts
The Decentralized Self-Improving AI System That Builds Itself Democratically
A self-improving system in one repo is impressive. A network of self-improving systems across many forks is qualitatively different.
The Self-Improving AI System That Keeps Building Itself
I used my AI assistant to spawn 15 parallel coding agents that built the system managing them. Half of them died. Here's what actually happened.
The Nature of Knowledge Work Has Fundamentally Changed
The role of humans in knowledge work is fundamentally changing — and most people aren't ready for what comes next.
Open-Sourcing Agent Orchestrator: Effectively Manage 30 Parallel Agents
How we open-sourced the system that manages 30 parallel AI coding agents — and how the agents built it themselves.
Every Decision I Make Now Has a URL
Static dashboards are dead. Living documents that update themselves, surface context, and help you act — that's the future.
Building an AI Trading Agent That Learns From Its Own Mistakes
Mean reversion, regime detection, sentiment analysis, and Kelly sizing — all orchestrated by an AI agent with human-in-the-loop approval.
Building a Personal AI Agent: Architecture and Lessons from LifeOS
How I built a personal AI system that manages my finances, relationships, health, and life admin — running 24/7 on a €7/month server.
Running 20 AI Agents in Parallel From My Home Directory
How I built a Claude Code orchestrator that manages 20+ parallel AI coding agents using tmux, git worktrees, and ~2,500 lines of bash.
Composio vs n8n — Integration Layer Comparison
A practical comparison of Composio and n8n for AI agent workflows.