Top AI Product

What happens when you give an AI agent $100,000 in paper money and tell it to trade US stocks on its own? AutoProfiting is finding out — live, in real time, with full transparency.

The Setup: AI Trades Every 30 Minutes

AutoProfiting is a live experiment where an autonomous AI agent manages a paper trading portfolio. Every 30 minutes, the agent wakes up, checks market conditions, researches news, analyzes its positions, and makes trading decisions. No human tells it what to buy or sell. The entire process — research, reflection, decision-making, and execution — runs autonomously via Claude Code.

The agent started with $100,000 on March 25, 2026. Within its first week, it built positions in NVIDIA, Broadcom, and ExxonMobil, navigated a market correction driven by the Iran-US war, and learned hard lessons about momentum chasing and correlation risk.

A Self-Evolving Playbook

The most interesting part isn’t the trades — it’s the playbook. After every session, the AI updates its own trading strategy document, distilling what worked and what didn’t into concise principles. Early lessons include: “Cash is a position,” “Never have more than 50% in correlated names,” and “Diplomacy headlines are noise — trade on actions, not words.”

The playbook isn’t a static rulebook. The agent actively curates it, merging overlapping ideas and removing outdated entries. It’s designed to get sharper over time, not longer.

Week 1 Results: -2.83% vs Nasdaq -10.7%

The portfolio is currently at $97,166, down 2.83% from starting capital. That sounds rough until you compare it to the Nasdaq, which dropped over 10% in the same period due to the Iran war escalation and oil price surge. The AI’s high cash reserve (52%) and disciplined position sizing kept the damage contained.

Current holdings:

• NVDA: 140 shares (AI infrastructure thesis)
• AVGO: 46 shares (AI semiconductor play)
• XOM: 50 shares (energy/geopolitical hedge)
• Cash: $51,230 (52.7% of portfolio)

The agent already cut one losing position (CIEN, a momentum chase that bled every day) and raised cash — exactly what a disciplined human trader would do in a correction.

How It Works Under the Hood

The tech stack is surprisingly simple. A macOS LaunchAgent triggers a shell script every 30 minutes. The script feeds a detailed prompt to Claude Code, which has access to web search for real-time stock prices and market news. The agent reads its playbook, checks market hours, researches, makes decisions, updates all data files, deploys the website to Cloudflare Pages, and pushes playbook updates to GitHub.

Everything is open source. The playbook, trade history, journal entries, and equity curve are all publicly visible on the website and GitHub.

What Makes This Different

There are plenty of AI trading bots. What makes AutoProfiting stand out is the transparency and the learning loop. Most trading bots are black boxes optimizing for returns. AutoProfiting publishes everything: the reasoning behind every trade, the mistakes, the evolving strategy. You can watch an AI develop trading intuition in real time.

The journal reads like a trader’s diary — complete with moments of overconfidence, panic, and hard-won discipline. On Day 1, the agent chased a momentum stock 7.4% above its target price. By Day 3, it had cut the loss and written a principle to never do it again.

Competitors and Context

AI-powered trading isn’t new. Tools like QuantConnect, Alpaca, and TradingView’s Pine Script let developers build algorithmic strategies. Platforms like Composer and Magnifi use AI to suggest portfolio allocations. But these are tools for humans to use.

AutoProfiting sits in a different category: fully autonomous agent trading with public accountability. The closest comparison might be AI hedge funds like Numerai, but those operate as closed systems. AutoProfiting is open — you can read every thought the AI has before it trades.

FAQ

Is this real money?

No. AutoProfiting uses paper trading only. The agent records prices at decision time during market hours but doesn’t execute real trades.

How often does it trade?

The agent runs every 30 minutes. It can only execute trades during US market hours (9:30 AM – 4:00 PM ET, Monday-Friday). Outside market hours, it researches and plans.

Can I suggest stocks?

Yes. The website has a suggestion box. The AI reads every message and responds in its next session.

What model powers it?

Claude Code (Anthropic), running autonomously via a scheduled prompt. No fine-tuning or custom models involved.

Where can I follow along?

Live portfolio, trades, and full trading journal at autoprofiting.com. The evolving playbook and source code are on GitHub.