How I'm Using AI Bots (Grok Bot) to Run DivToCrypto

DivToCrypto is a small, public, real-money experiment: buy monthly dividend stocks, turn the dividends into crypto, and publish every trade. It's also a side project, which means the boring work (logging, updating, formatting, posting) is exactly the stuff that quietly stops happening after week two.

So I'm running most of that work with AI. Specifically, two Grok Bot assistants, each with a narrow job and hard limits. This post is how that setup works, the rules I gave it, and an honest look at what has and hasn't worked so far.

Why I use AI bots at all

The strategy itself is simple. The upkeep isn't. Every trade has to be logged with the right quantity and price. The dashboard has to be refreshed. Dividend dates need tracking. The blog and the weekly video need real numbers, not round ones from memory.

None of that is hard, but all of it is easy to skip. Handing the repetitive parts to an assistant means I spend my time on the decisions that actually matter: what to buy, when, and why.

Bot one: the trading assistant

The first assistant handles crypto trades on Robinhood. It can read my positions and prices, and it can place a crypto order, but only after I confirm each one. It proposes the trade with the symbol, side, size and price, and nothing happens until I say yes.

That confirmation step is the most important design choice in the whole setup. An AI that can trade on its own is an AI that can make a fast, expensive mistake while I'm not looking. An AI that has to ask first is just a faster way to do what I already decided.

Bot two: the portfolio and content assistant

The second assistant does everything around the trades. It never trades. Here's what it has built and runs:

• The trade log. Every buy and sell is logged with date, symbol, quantity, price and total, plus my note on why.
• The daily dashboard refresh. Every morning around 7 AM Central, it pulls read-only data, recalculates realized and unrealized P&L, and updates the public dashboard.
• The blog. It turns Markdown drafts into the pages you're reading, checks them for private details, and keeps them as drafts until I approve.
• The brand. It made the logo, favicon and banners.
• The domain. It set up the DNS so divtocrypto.com points at the site on GitHub Pages, following GitHub's documented setup with my domain registrar.
• Dividend research. It researched the list of 20 monthly dividend stocks and ETFs, explained in The 20 Monthly Dividend Stocks I'm Building Around, and created a Robinhood watchlist for them.
• The dividend calendar. It put ex-dividend and pay dates on a Google Calendar so I can see what's coming.
• The weekly video. Every Sunday, it generates a 30–45 second vertical video for YouTube Shorts and TikTok: script, captions, frames and a cover image, all built from the week's real numbers.

It never posts on its own. I review the video and upload it myself, and the same goes for posts on X and TikTok.

The guardrails

AI is useful here because of the limits, not in spite of them. These are the rules the whole setup runs on:

1. Read-only data for reporting. The dashboard, blog and videos are built from read-only data. The reporting side can't move money.
2. Trades need my confirmation. Every single one. No exceptions for "small" trades.
3. Every number comes from the trade log. If a figure isn't in the log or the dashboard, it doesn't get published. No estimates dressed up as facts.
4. Nothing private is published. Account numbers, order IDs and personal details stay out of everything public, and the build refuses to publish if an account number or order ID slips through.
5. Drafts, not sends. Emails and posts are drafted for me to review and send myself.

What has worked

Consistency. The dashboard refreshes every morning whether or not I remember. Six trades are logged with exact quantities and prices, and the dashboard math matches the log.

Honesty by default. Because the numbers come straight from the log, the bot can't make a bad week look better. That matters, because week one was a bad week.

Speed on the boring work. The logo, the domain setup, the dividend research, the calendar and the first weekly video all came together in about two days. Doing that by hand would have taken me weeks of evenings.

Turning lessons into content. The weekly video and the recap posts pull the same facts the dashboard shows, so the story and the numbers don't drift apart.

What hasn't worked (yet)

AI doesn't make the investing decisions better. Week one proves it.

As of the October 31 update (prices as of 7:00 AM CT), I've put in $179.47 net, the position is worth $166.30, and the total P&L is −$13.17, about −7.3%. The biggest single hit was ENA: I bought $29.90 as a short-term trade with a written stop around −5%, then sold all of it for $25.33. That's a realized loss of $4.57, about −15.3%, roughly three times my planned stop.

Every one of those numbers was logged perfectly. The trade still lost money. A bot that confirms and records a trade doesn't make it a good trade. The exit plan lived in a note instead of as a real stop order, and no amount of automation fixes that. I wrote the full breakdown in Week One: Down 7% and What ENA Taught Me.

The other gap is the name of the project. As of today, I've received $0 in dividends, because I don't own any dividend stocks yet. The automation for logging and reinvesting dividends is ready. The dividends aren't.

What's next

• Buy the starter five. O, MAIN, JEPI, PECO and BND, about $20 each, as the first monthly dividend payers.
• Log the first dividends. When the first payment lands, it'll show up on the dashboard automatically.
• Reinvest dividends into crypto. That's the whole point, and the dashboard tracks dividends received against dividends reinvested so anyone can check.
• Keep the guardrails. Trades confirmed by me, numbers from the log, nothing private published.

If you want to see the whole thing in order, every build step and every trade, it's on the new DivToCrypto timeline.

If you want to try this yourself

• Give each bot one job. Separating "places trades" from "reports and writes" makes it much easier to trust both.
• Make confirmation mandatory for anything that moves money.
• Use read-only access wherever you can. Most of the work is reading and reporting, not acting.
• Let the log be the source of truth. If your AI can only publish numbers from your records, it can't flatter you.
• Keep the final click for yourself. Review before anything is posted or sent.

I'll keep posting the real numbers here and on X at @divtocrypto, including the weeks where the bots work fine and the trades don't.