This week's issue is a little different.

I usually walk you through one specific workflow I've been running.

Today I'm zooming out and giving you my full AI stack instead.

I get the same email a dozen times a month.

Some version of "I'm trying to figure out which AI tools to actually use for investing, what should I use?"

I understand the feeling.

There are now hundreds of AI products that claim to be useful for investment professionals, and most of them aren't.

I have a luxury most working investors don't.

I get to spend all day testing these tools :)

So here's my current AI stack, the version I'm actually running right now.

The Framework

I organize the stack into four tiers:

Daily Driver. The thing I open the moment I sit down. The agentic environment that everything else plugs into.
Constant Companions. Tools running in the background of every workday. I’m not always actively in them, but they’re always there.
Specialists. Pulled out for a specific job, then put back. Worth paying for, not worth living in.
Experimenting. Things I’m testing but haven’t formed a view on yet. The “might be the future” bucket.

Daily Driver

My Daily Drivers are Claude Code and Codex.

These are the two agentic environments I have open the moment I sit down, and most of what I do for the rest of the day routes through one or the other.

They sit at the foundation of the stack, with everything else plugging in on top.

A few of the workflows that live in the "daily driver bucket" include:

Multi-agent orchestration for thematic research
Portfolio stress testing
Alternative data scraping
Majority of due diligence tasks
Custom skills for repeated tasks (e.g., initiation memos, earnings reviews, email work)

Claude Code and Codex are both very solid as the frontier daily driver.

This is where you should spend most of your time in AI.

Constant Companions

Four tools sit here that act as "enhancers" for Claude Code / Codex.

Granola for call notes. The MCP server is what actually makes it sticky. So basically every call I take is now information AI can read ... super valuable for private information on the job.

Notion and Obsidian for information storage, split by purpose. I connect these as MCPs for context ingestion for any SOPs, notes I have, or information I want to systematically capture for AI to know.

Google Workspace through the MCP suite for Gmail, Calendar, and Drive. With this wired in, AI understands the context you had over email. This turns generalized outputs into personalized outputs (plus enhances productivity via scheduling and email work!).

ChatGPT in the consumer app, currently on GPT-5.5. I keep this open separately for second-opinion reads, fast brainstorms, and whatever doesn’t need a full agentic session.

Specialists

I use these when the right job shows up.

AlphaSense is where I go for expert call transcripts data especially.

Perplexity Computer is what I reach for when I need a quick and dirty agentic session. Especially when I don't need to touch local files (e.g., preparing a data room, putting together a quick document, etc).

Manus sits in the same lane as Perplexity Computer.

Excel plugins for both Claude and ChatGPT. Both are good and roughly comparable in quality. For any sort of excel work, these plugins are a must.

Experimenting

I'm always experimenting with everything. Two things are catching my eye at the moment:

OpenAI’s new Realtime API, which shipped last week. Use cases here include real-time translation for non-US equities + real-time modeling for "analyzing earnings calls" as the CEO speaks ... you can imagine how you can scale earnings season with this.

Anthropic’s new finance agent plugins, which shipped May 5. I have generally found Anthropic's plugins lacking depth for reliable investing use cases (because they try to solve for all of finance including stuff like accounting). But I'm keeping an eye to see how this improves.

The Data Layer (MCPs)

To get the absolute best results with tools like Claude Code & Codex, you need data.

The solution is MCPs that pipe in private and public data. Here are some of my favorite ones I use:

Polygon for market data (real-time and historical equities, options, FX, crypto)
Finnhub for fundamentals, ownership, insider activity, and the earnings calendar
FRED for macro series (yields, CPI, employment)
Unusual Whales for options flow and congressional trades
X API for X and fintwit signals
Pitchbook MCP for private markets intel

For workflow optimization, I use the following MCPs:

Slack
Granola
Google Workspace
Notion
Obsidian
Exa (for expanded web research)
Firecrawl (for scraping)

Personal

If you’ve read this far and want to actually wire any of this up, that’s exactly what I’m teaching in my new Claude Code for Investors live bootcamp.

Reserve your seat!

I’m running a five-day live cohort June 1 to 5.

You'll learn hands on the setup from scratch, the MCP wiring, the agent stack that runs research overnight, and the diligence and memo workflows that come out of it.

This is the same playbook I’ve been delivering in private workshops at some of the largest hedge funds in the world.

As a thank you for being a part of my newsletter, I'm giving a 50% discount to you.

The first 100 early bird seats are at $497 (~50% discount) and they close May 25. After that, the standard tier is $999.

Hope to see you there!!

Talk soon,

-Dave

When you're ready, here's how we can work together:

Join Wall Street Prompt - On-demand AI learning
Live Corporate Training - Live AI for finance workshops

Get one high impact AI prompt every week