AI for Investing
The trust problem: why you can't just ask a chatbot to invest
You have heard that you can just use ChatGPT, Claude or Gemini to manage your finances, as long as you know the right prompts. Try it. Paste in: "I have $10,000. Help me invest in stocks. I am risk-averse and want at least 2% annual return." You will get a response. Maybe 40% in dividend stocks, 30% in bonds, 30% in a broad market ETF. It will mention low volatility and stable returns. It will sound reasonable, and confident.
Now the question: can you trust it? Did the model actually check whether those stocks have low volatility? Did it calculate whether that allocation can deliver 2% in today's market? Did it check that the holdings are genuinely uncorrelated? Maybe. There is a good chance it did not. When it did not, it produced recommendations that sound like they fit your criteria, without checking whether any of it is true, and delivered all of it with complete confidence. Now imagine that powering your portfolio, your risk assessments, your money.
This is how LLMs work, not a bug
It is not a flaw in one chatbot. An LLM is good at two things: taking in language and producing fluent language. Natively it is not a database and not a calculator. It cannot see today's prices and it cannot run a computation. A simple lookup, such as Apple's current price, a model with web search can handle. But real finance questions are not lookups. "Build me a low-volatility portfolio" needs historical prices, a volatility calculation for each candidate, the correlations between them, and an optimisation. When the model cannot do those steps, it fills the gap with text that sounds right, and it never says it is guessing. So you cannot tell which parts are grounded and which are fiction. A portfolio that sounds low-volatility might be three correlated tech stocks. Advice that appears personalised might be generic patterns from training data, dressed in confident language.
The fix: give the LLM tools
There is a fix. Not a perfect one, since you are still using an LLM with all its faults, but a step change from hoping it gets things right. The problem is not what the model can say. It is what it cannot do natively: fetch data, run calculations, take actions. So give it tools that can.
A tool is a function the model can call: get this stock's volatility, get the correlation between these two, fetch the price history. Hand the model those functions and it becomes an AI agent, an LLM that can act rather than only respond. It receives your request, decides which tools to call, reads the results, and goes round again until it has what it needs. That is the difference between asking someone a question and asking someone to go and find out. Ask for a low-volatility portfolio now, and the model calls the volatility tool on candidate stocks, calls the correlation tool, runs an optimisation on real data, and then explains why the result fits your risk profile. Same question, real numbers. The model still does the language. It is no longer guessing the numbers.
LLMs work best as buns, not meat.
One analogy carries the whole idea. The buns are what you interact with: they parse your question and turn the output into something readable. The buns are not doing the real work. The meat is, and it is ordinary computation: returns from an API, volatility from a library, a correlation matrix, an optimiser. Consistent, explainable, checkable. The vegetables are the boring but necessary layer: data validation, error handling, rate limits, audit logs. The repeated mistake is treating the model as the meat, asking it to know facts, run calculations, and make decisions. That is when it invents things, because you are asking the bun to do the meat's job. Not smarter models, not more training data. Just the right architecture.
Tool calling is the foundation. Real finance work needs more: fetching data, multi-step work across several candidates, custom calculations, and memory so a follow-up such as "what if I increased the bond allocation?" makes sense. Those are the four agent patterns, and before them it helps to know when you need an agent at all.
Keep going
This is from the primer. Start at AI for investing for the whole picture, or get the full book, AI Agents for Investing, which builds it in code. I am publishing more in the coming weeks. Subscribe to get it.