How Claude Max Lets a Non-Programmer Build an AI Investing Platform

Until recently, the idea of a public equities investor with limited programming knowledge building their own AI-powered investing platform would have been a punchline. You hire a developer. You buy a SaaS. You stay in your lane. The economics of software made it so.

That economics broke. I’m a stock investor, not an engineer. I read 10-Ks for fun, not Stack Overflow. And yet over the past year I’ve built MoatMap: a platform that screens roughly 19,000 stocks across 24 markets, runs AI deep dives on the most interesting ones, generates a paper portfolio with regime-adaptive rebalancing, and tracks every decision in an auditable trail.

The reason I could do this, and the reason a lot of investors will follow, is Claude Max.

The strange premise

Claude Max is Anthropic’s top consumer subscription for Claude, paired with an agentic coding tool called Claude Code. The combination is more than the sum of its parts.

I describe what I want in plain English. Claude proposes an architecture, asks the questions an experienced engineer would ask, writes the code, runs the tests, fixes its own bugs, and ships. I review the diff. I push back when the design feels wrong. I run it in production and watch what happens. I am, in effect, the product manager and the QA tester for an engineering team of one: an engineer that never sleeps and has read most of the open-source code on the internet.

This is not no-code. The output is real code, in real repositories, with real tests. But the cognitive load on me sits squarely where I’m comfortable: thinking like an investor about what an investor needs.

“The most important investment you can make is in yourself.”

What I’ve built so far

MoatMap breaks down into a small number of opinionated tools, each built to solve a specific problem I had as an investor. None of them are exotic. Most of them have institutional analogues that have been quietly outperforming for decades. The novelty is that they now exist in a single, cohesive platform that a retail investor can actually use.

Quantitative stock scoring (Quality, Value, Momentum)

The scoring engine pulls fundamentals and price history for the entire investable universe and ranks every stock on three classic factors: Quality (margins, returns on capital, balance sheet strength), Value (earnings yield, free cash flow yield, relative multiples), and Momentum (multi-timeframe price strength with volatility scaling). The composite StockRank is the first filter on every idea.

I wrote about the academic foundations of this approach in Factor Investing: How Tilting the Odds in Your Favour Actually Works. It is not glamorous. It is just a disciplined refusal to ignore decades of evidence. And Claude built the entire pipeline in the space of a few weekends.

AI deep dives

The Deep Dive feature is the one I am proudest of, and the one this article is really about. For any stock in the universe, MoatMap can generate a 5,000–10,000 word research report covering the business model, competitive position, financial history, recent news, and the bull and bear cases. The AI is given the company’s filings, five years of financial history, and a live news feed, and asked to think hard before writing.

A year ago I would not have given an AI-written report 30 seconds. Today I read them in full.

The 20–30 minute test

Here is the moment I want to flag, because it is the moment that changed my mind about how seriously to take all of this.

For most of the last few years, AI-generated text has been recognisable. The structure was tidy but bloodless. The arguments were balanced to the point of saying nothing. The language hedged where a human writer would commit. Reading AI output felt like the literary equivalent of eating airplane food: technically nutritious, structurally identical to the real thing, missing every interesting molecule.

The current generation of deep dives has crossed a line. When I read a Deep Dive on a company I know well, I find myself learning things: a perspective I had not considered, a connection between two parts of the business I had not made, a supply-chain risk I had filed away as background noise. I disagree with parts of every report. That disagreement is itself the proof that the report is worth reading: a generic, formulaic piece of writing produces no friction.

The honest test is this: would I spend 20–30 minutes of my finite reading time on this report, knowing it was written by AI?

For the deep dives MoatMap generates today, the answer is yes. That is new. It would not have been true twelve months ago. It is the single biggest reason I think the next leg of this journey is worth attempting.

“What I cannot create, I do not understand.”

The next step: an AI investor that makes the calls

The tools I have today are decision-support. They surface the universe down to a manageable list, flag the dangerous names, and produce reports good enough to read. But the decisions (what to buy, how much, when to sell) are still made by me, with all the human limitations that implies. I am slow. I am biased toward names I have held before. I am bad at selling winners. I forget to look at sectors I do not personally understand.

The next thing I want to build is an AI investor: an agent that takes the full output of the platform (scores, news verdicts, deep dives, macro regime, current positions) and makes the actual decisions. Buy this. Sell that. Hold the rest. Reasoning logged. Every call auditable. A paper-trading runway long enough to either earn my trust or expose its flaws.

I am not naïve about the difficulty. AI hallucinations in finance can cost real money. Markets are adversarial in a way that very few of an LLM’s training environments are. The historical record is full of clever systems that worked on backtests and broke on contact with the real market.

But the components are now good enough to try. The factor models are real and well-evidenced. The deep dives are good enough to read. The news filter catches the events I would have missed. The macro regime layer is the right kind of slow-moving signal. Stitch them together with a careful decision policy, run it on paper for long enough to see the edge cases, and you have something worth taking seriously.