What you will have at the end: Claude Code installed and working, a project folder for your engine, a profile file that tells Claude Code who you are and how you invest, your two secret keys stored safely, and proof that your API key returns live market data (that is, real data pulled over the internet using your real key, as opposed to made-up test data).

You will also have pulled the same company’s numbers from SEC EDGAR, so you have a second source from day one.

Read more

This “Lesson 0” delivers the map and vocabulary for the entire Crash Course: what you are about to build, how it decides what to send you, where its numbers come from and where they may be weak, what the whole thing costs, and every technical word the later lessons use.

Twenty minutes here saves an hour of confusion in Lesson 1, and it is where you learn the full cost of building the engine before you are asked to spend a dollar. Read it once, at your own pace. No need to memorize any of it; come back to this lesson whenever you like.

Subscribe now

What you will be able to do

By the end of this course you will own a small program that emails you an investment idea on most working days. It runs every weekday and sends when it has something new. Each idea arrives as a one-page memo: what the company does, why it scored well against your criteria, where it looks weak, and the raw valuation numbers so you can judge for yourself.

The engine is an idea funnel for your own research, not investment advice, and not a buy list. Each email surfaces at most one candidate a day from a large global pool of ideas. Verify the engine’s assertions against the company’s own filings, as data issues are unavoidable in a limited-scope project such as this one. The data the engine draws on can be wrong or stale, and a machine that ranks numbers cannot know what you know. Treat each memo as a lead worth a bit of your time, never as a verdict.

The mental model

In this project, you are not going to write code. You are going to talk to an assistant called Claude Code, in plain English, and it writes and runs the program for you.

Two qualifications: You will read plain-text settings files, which hold your country list, your size limits, and your scoring weights, and you will either edit them yourself or ask Claude Code to edit them for you. And in Lesson 2 you will have an opportunity to hand Claude Code a list of the exact names your data provider uses for revenue, operating cash flow and the rest, because guessing those is the one thing an assistant cannot do reliably. That is closer to handing a new analyst the list of field names your data provider uses than to programming, and the lesson gives you the list. Nothing in this course asks you to write or debug a line of Python.

Picture a very capable analyst who is excellent at the computer parts. You type a request in ordinary sentences, like “only include companies above $300 million in size, and skip banks.” Claude Code does the technical work, shows you the result, and waits for your next instruction. If it is not what you wanted, you say so in plain words, and it adjusts.

You are in control at every step, and the controls are worth knowing before you need them. Claude Code has a setting called “Manual” in which it shows you each change it wants to make and waits for you to accept or reject it; Lesson 1 starts you there, and that is where you stay for this course. (Newer accounts otherwise start in a mode called “Auto,” where a second AI checks each action in the background instead of asking you. Fine later; not where to learn.) Each time you send it a request it first saves a snapshot of your files, so you can put things back the way they were before that request. You can stop it mid-task. And there is a mode, called plan mode, in which it can read and think but cannot alter anything, which is where you should start any step you are unsure about.

So the honest version of “you cannot break anything” is this: you can break things, and then put them back. On a bad day a result looks wrong, you say “that is not what I meant,” you roll back if you need to, and you try again. That is the rhythm of the whole course: plan, ask, look, check, adjust.

Two ways to use Claude Code

You have a choice to make in Lesson 1, and it is worth understanding now.

The desktop app is an ordinary application window. You type your request in a box, and in the setting this course uses, when Claude Code wants to change a file it shows you the old version and the new version side by side, with Accept and Reject buttons. It has a menu for choosing plan mode. Nothing is typed as a command. This course leads with the desktop app in every lesson, because the course is written for people who have never used a terminal.

The terminal version does exactly the same work in a plain text window. It is what you will use for a few steps in Lesson 5, when you set up the always-on computer that runs the engine each morning. Where the two differ, the lesson gives you the terminal steps in a marked box.

They share everything that matters: the same assistant, the same files, the same settings. You can switch between them, or use both. The desktop app needs a Claude Pro or Max subscription. The terminal version also runs on an API key alone (a paid account where you are billed for what you use, rather than a flat monthly subscription), which is the cheaper route if you do not want a subscription. It does mean working in a text window for every lesson, in a course written for people who have never opened one.

Read more

That is what you build in our Claude Code Crash Course, which opens on Tuesday, August 18. It is the follow-up to our popular AI Bootcamp for Non-Technical Investors, and it stands on its own. The bootcamp leaned toward concepts, the new course gets you a working analyst.

An idea in your email inbox, daily

By the end of Lesson 5 you own a small program that runs itself each weekday morning and emails you a one-page memo on a single company: what it does, why it scored well against your criteria, where it looks weak, and the raw valuation numbers so you can judge for yourself.

The screen is a hard filter and nothing more. Countries, size floor and ceiling, industries you never want to see. The scoring model is what picks the day’s idea, ranking the survivors on six weighted factor groups: valuation 35, returns on capital 25, margin quality and stability 15, capital discipline and consistency 15, balance-sheet strength 5, growth and reinvestment 5. Every ratio is computed from the raw income statement, balance sheet and cash flow statement, never from a vendor’s precomputed ratio fields, because you cannot see how those were built or what they did with an odd accounting item. The engine will not send you the same company twice until that company reports a new period.

Before any memo goes out, the engine checks the data against the company’s own SEC filings and puts one of three outcomes in the memo: the two sources agree within 1 percent; they disagree, with both figures and the size of the gap; or the company does not file with the SEC, so no independent check was possible.

Subscribe now

Your feedback trains your analyst

When a memo arrives, you read it the way you would read a junior analyst’s note. Some ideas fit your process. Others miss. Here is what no subscription can do for you: you hit reply. You write, in ordinary words, “too capital intensive,” or “raise the minimum size to $2 billion,” or “more quiet compounders like this one.” The engine reads your reply before it builds the next morning’s idea, converts the judgment into a concrete filter or weight change with the chosen number stated, tells you at the top of the next memo exactly what it changed, and reverses it if you reply “undo.”

Over a few weeks the stream bends toward how you actually invest. You are not editing settings in a menu. You are training an analyst, in the same plain words you would use with a colleague. A newsletter sends the same names to everyone; your engine is private, it is yours, and it runs on your definition of a good business at a fair price.

The discipline is the real prize. The engine enforces a repeatable, criteria-driven process and sets one pre-screened candidate on your desk every weekday, whether or not you feel like screening that morning.

The six lessons

Each lesson is written and self-paced, so you can work through it the morning it lands or a month later, and revisit any of it whenever you like. Each one ends with something that works.

• Lesson 0, Tuesday, August 18. Orientation in plain words. Nothing to install and nothing to set up: the mental model, the vocabulary, and a tour of what you are about to build. A 20-minute read.

• Lesson 1, Wednesday, August 19. Setup and your first data pull. Claude Code installed, your project created, your market-data key returning real numbers, and SEC EDGAR ready for checking them.

• Lesson 2, Thursday, August 20. The screen and the scoring model. One instruction you can run that ranks your filtered universe, with a clear breakdown of how each company earned its score.

• Lesson 3, Tuesday, August 25. Selection, no-repeat, and the memo. A full one-page memo on the single best company, generated end to end, with the filing check built in.

• Lesson 4, Wednesday, August 26. Email and the reply-to-refine loop. The memo lands in your inbox, and replying to it in plain English changes the engine’s criteria.

• Lesson 5, Thursday, August 27. Cloud automation and maintenance. The engine runs itself every weekday on an always-on server, whether or not your own computer is awake.

No-code approach

You describe what you want in ordinary sentences, and Claude Code writes and runs every line. The course leads with the Claude desktop app in every lesson, an ordinary application window with a text box, a side-by-side view of any change it proposes, and Accept and Reject buttons, because it is written for people who have never opened a terminal. The terminal steps are there too, in clearly marked boxes, for anyone who prefers them.

Two qualifications: You will read plain-text settings files that hold your country list, your size limits and your scoring weights, and you will edit them or ask for them to be edited. And in Lesson 2 you hand Claude Code the exact field names your data provider uses for revenue, operating cash flow and the rest, because guessing those is the one thing an assistant cannot do reliably. The lesson gives you the list. That is closer to handing an analyst a data dictionary than to programming, but it is not nothing, and you should know it is coming.

A complete reference build of the engine comes with the course, rebuilt on the current data provider and live-tested on August 10, 2026. If your own build gets stuck you can compare it against a version that works, or ask Claude Code to set that version up for you. It is a worked example rather than a substitute for the lessons; the lessons are the path I would take.

What you need

An hour or two per lesson, a computer running macOS, Windows or Linux, and a normal level of comfort with email and a web browser. No programming experience is assumed. The engine relies on your own market-data subscription and a modest amount of AI usage; Lesson 0 lists every account and every cost in full, including the cheaper route at each decision, so nothing arrives as a surprise in Lesson 1.

A caveat

The engine is an idea funnel for your own research. It is not investment advice and it is not a buy list. Every memo is a lead worth an hour of your time, never a verdict.

The score is relative, not absolute. Each company is ranked against the others that pass your filter on that day, so in an expensive market the cheapest survivor can still be expensive. That is exactly why every memo carries the absolute valuation numbers alongside the score.

Data quality varies by country, and the course is candid about where the numbers come from. Companies that report only twice a year, common in the United Kingdom and much of continental Europe, and some non-US cash flow statements, are the weak spots. Worse, SEC EDGAR holds only companies registered with the SEC, so the thinnest part of your data is also the part with no automatic second source. Verify every idea against the primary filings before it means anything.

How to get it

Paid Latticework subscribers and MOI Global members have full access to the course, alongside everything else we produce.

Lesson 0 goes out on Tuesday. If you are not a subscriber, consider becoming one now.