I Built a Daily Investment Briefing with AI — This Week's Biggest Lesson Was Discipline
I spent a good chunk of this week on my daily investment briefing. It's not a new product, just a pipeline I built for myself: every day, AI reads through what's happening in the AI industry and the markets, writes a briefing based on my investment thesis, and publishes it to my personal site automatically.
This week it got an upgrade. It also broke twice. So I want to talk about what AI is actually good for when you use it to follow an industry and think about long-term value.
Quick note up front: this post is just my personal view, shared to swap ideas. It's not investment advice.
1. First, it broke: my data sources dried up
Midweek, the scraping tool I rely on shipped a major version bump. The response format changed, and my Reddit and X collectors went down all at once. Right after that, a finance data site I use a lot got redesigned, and the briefing's prefetch step broke too.
So the AI still had a brain, but nothing to feed it. Claude and I fixed it in two rounds: first we adapted to the new response format, then we got the prefetch working again.
The lesson is a simple one: automation isn't set-and-forget. Your data sources are the most fragile part of the whole system, and one upstream redesign is enough to leave your judgment built on thin air.
2. New: a ten-bagger radar
Once things were working again, I added a module that screens for companies within my thesis that could plausibly grow 10x, and then researches them.
I put some limits on it on purpose. First, it only looks inside the scope of my thesis. No chasing whatever is hot. Second, every candidate has to spell out what needs to be true for the thesis to hold, and what the biggest risk is. Honestly, the point of this radar isn't finding the next ten-bagger. It's forcing me to explain the long-term case clearly. If I can't explain it, it doesn't go on the list.
3. Letting the prompts review themselves, without touching the thesis
The second change is regular reviews: one at the end of every week, one at month-end, and one at year-end. The AI compares what actually happened during that period against the briefing and fine-tunes the briefing prompts.
I gave it one hard rule: don't change the thesis unless there's clear evidence that it's wrong.
The reason is that I noticed how easily people (and AI) get dragged around by short-term price action. A sector rips today and you want to write it into your core thesis. It drops tomorrow and you want to delete it. Long-term value needs an anchor that doesn't swing with intraday moves. Reviews are for sharpening the wording and the details, not for chasing rallies and panic-selling dips.
4. Post-buy discipline cards: write down when you'll admit you're wrong, in advance
I also run a model portfolio for research. This week, for every position I just opened in it, I had the AI play a buy-side analyst and write a post-buy discipline card: what my cost basis is, where the mechanical stop-loss sits, and what would mean the thesis is broken. It could only use analyst reports and public information it had actually found. No making up numbers.
On top of that, there's a weekly thesis check: go through the past 7 days of news, earnings, and regulatory events, and judge for each assumption whether it still holds or whether we've hit a red line.
My biggest takeaway after writing these cards: the reasons to buy are usually easy to write. The hard part is writing down when you'll admit you were wrong. Getting that on paper ahead of time matters more than nailing any single entry point.
Wrapping up
After this week, I see AI for following an industry a bit differently. It can't predict the future for me. But it's really good at three things: reading information without ever getting tired, organizing it into a structure based on rules, and reminding me every week to go back and look at what I said in the first place.
For someone like me, building products solo while also trying to take long-term investing seriously, that's already worth a lot.
One more time: everything above is just my personal view, shared to swap ideas. It's not investment advice, and the model portfolio is only for keeping research notes.
Do you write down your own conditions for admitting you're wrong when you invest? Drop a comment, I'd love to hear. I'll keep posting updates on how this system evolves.