Robo-advisors and automation

AI trading explained: how AI trading bots work, what they can’t do, and how to start

Written by

Reviewed by

Arpit AgarwalCharvi AgarwalThe Amaltash Team

Updated · Published

The short answer. AI trading means using artificial intelligence to research, plan or place trades. The AI usually does one of four jobs. It answers questions, predicts prices, turns your idea into rules, or trades as a bot. No AI can reliably predict markets or promise a profit. You can lose money.

How big should one position be?

$
%
%

A stop is not a guarantee: prices can gap past it.

Per position
$5,000
Full-size positions that fit
10
At risk per position
$5,000
At risk, share of portfolio
10.00%
Position$5,000
At risk$5,000

Hypothetical illustration based on your inputs. Not a prediction, a recommendation, or the performance of any Amaltash strategy.

Turn this into a rule

Build a trading bot for SPY. Buy when SPY closes above its 200-day moving average. Sell when it closes below it. Use 10% of the account. Add a stop-loss 8% below the buy price.
Turn this into a rule

Opens the Strategy Builder with this rule filled in. It starts building after a 3-second countdown; type to stop it. Nothing trades until you deploy.

What is AI trading?

AI trading means using artificial intelligence, or AI, to help you trade. Trading is buying and selling stocks, funds or other assets. Most traders hope to profit from price moves. The AI can help you research, decide or act.

Computers have traded by fixed rules for decades. That is called algorithmic trading. Two kinds of AI are newer. Machine learning finds patterns in past data. Large language models, the AI behind chat apps, let you talk to software in plain English.

Inside AI trading apps and platforms, the AI usually does one of four jobs. Each job can fail in its own way. Knowing the job tells you what to check. The table is a map, not a review of any app.

The four jobs AI does in AI trading, and how each one fails

Job What it does Who places the trade How it can fail
Research Answers questions and sums up news, filings and prices You Gets a fact wrong and sounds sure
Predict Guesses where prices will go, sometimes from the mood of the news (sentiment), and sends buy or sell signals You, or a bot Stops working when markets change
Write rules Turns your plain-English idea into exact trading rules The rules, once you turn them on Reads your idea in a way you did not mean
Trade A bot or AI agent places orders on its own The bot or agent Repeats a bad trade many times, fast

Source: Our own grouping. FINRA’s June 2020 report on AI sorts its uses into customer communications, investment processes and operational functions.

How do AI trading bots work?

A trading bot is software that places trades for you by set rules. You connect it to a brokerage account. It watches prices and sends orders when its rules say so.

At its core, every bot answers three questions. When do I buy? When do I sell? How much do I trade? Careful bots add a fourth: when do I stop? A stop-loss is one answer.

The AI can sit in one of two places. It can write the rules once, from your words, and then step back. Or a model can make each choice itself, from patterns it learned in past data. You can read the first kind. The second kind is much harder to check.

Here is what happens inside a bot each time it checks.

  1. Data comes in, like prices, trading volume and sometimes news.
  2. The rules or the model decide: buy, sell or wait.
  3. The bot sends an order to your broker.
  4. Limits check the order, like the most it may spend.
  5. The bot logs what it did, so you can review it later.

Does AI trading work?

AI trading works for some jobs, like following a rule without getting scared. But no AI reliably predicts the market. Prices react to news that no model has seen yet.

Regulators say the same. On January 25, 2024, the CFTC reminded the public that AI technology cannot predict the future. The SEC, NASAA and FINRA warned the same day that AI output “can be faulty, or even completely made up.”

A model that looked great on old data can fail on new data. This is called overfitting. Backtests can also flatter a bot. They may use hindsight, or leave out costs like fees and spreads. That is why a backtest is a test, not a promise.

So the useful question is not “does AI trading work?” It is “which job does this AI do, and how can I check it?”

Is AI trading legit, or a scam?

Using AI to trade is legal in the US. But “AI trading” is also a common hook for scams.

The CFTC’s advisory on trading bots tells of one case, Mirror Trading International. Its bot program “guaranteed at least a 10 percent monthly return,” the CFTC says. It took more than $1.7 billion in bitcoin from at least 23,000 people. Very little money was ever traded.

Scams like this keep coming. On September 29, 2026, the SEC charged the firms behind one offer to rent out AI trading bots. The SEC alleges there were no AI trading bots at all.

Real firms can overstate their AI, too. On March 18, 2024, the SEC charged two investment advisers over false claims about their use of AI. The SEC called this “AI washing.” The two firms agreed to pay $400,000 in penalties in total.

Watch for these red flags in any AI trading offer.

  1. It promises high or guaranteed returns, or a “100%” win rate.
  2. A stranger or an influencer found you online and pushes you to act fast.
  3. It asks you to send crypto or a wire to an account it controls.
  4. It asks for your brokerage login or password.
  5. It will not show its rules, or calls them a secret.
  6. The firm is not registered. Look it up on Investor.gov first.

AI trading vs algorithmic trading: what is the difference?

Algorithmic trading follows fixed rules that a person wrote. AI trading adds a model that learns from data, or AI that writes the rules for you. Many tools mix both. A robo-advisor is different again: it manages a long-term portfolio for you.

Each has its own guide: algorithmic trading, AI investing for money you plan to hold, and the robo-advisor.

AI trading, algorithmic trading and robo-advisors compared

Question AI trading Algorithmic trading Robo-advisor
Who writes the rules? A model that learned from data, or AI working from your words A person, usually in code The firm, from your answers to a quiz
Can you read the rules? Sometimes. A learned model is hard to read. Yes, if the maker shares them Partly. You see the mix of funds.
How often does it trade? From seconds to months From split seconds (high-frequency trading) to months A few times a year, to rebalance
Main way it fails The model is wrong, and you can’t tell why The rule fit the past, not the future The mix does not fit your goals

Source: Our own comparison of how each kind of tool usually works. Real products vary.

How to build an AI trading bot with no code (and test it first)

You do not need to code to try AI trading. In the video above, Charvi Agarwal builds a first bot from one plain-English sentence. Then she backtests it and looks at where it went wrong.

Charvi is Amaltash’s co-founder. She was one of the first data scientists at Waymo, back when it was Google’s self-driving car company.

Here are the same steps in Amaltash’s Strategy Builder. It is one example of the “write rules” job. Other tools work differently.

  1. Write your idea in one sentence. Say what to buy, when to buy, when to sell and how much.
  2. Let the AI turn it into exact rules. It adds guardrails, which are limits like a stop-loss.
  3. Read every rule. One sentence can become rules in more than one way. Change anything by chatting.
  4. Backtest the rules on past US stock and ETF prices. The results are hypothetical, not real trades.
  5. Find where it went wrong. Look at the worst drop, called the max drawdown, and when it came.
  6. Fix one thing at a time. Then test again on a different stretch of years.
  7. Paper trade it, which uses no real money.
  8. Turn it on only if you choose, with a small amount. It runs until you change or stop it.

Is AI day trading a good idea?

Day trading means buying and selling the same stock within one day. AI does not change the math. Costs, taxes and speed still count against you.

Research on day traders is grim. One study followed everyone who started day trading one futures market in Brazil from 2013 to 2015. Of those who kept at it for 300 days or more, 97% lost money.

That is one market, but FINRA warns the same way in a June 2026 guide. You should “never fund this type of trading with essential assets,” it says. That means money for rent, bills or savings.

Amaltash is built for rules-based investing over months and years, not for day trading.

Can an AI agent trade stocks for you?

As of 2026, some brokers and apps let an AI agent place trades in your account. An AI agent is AI that takes actions, not just answers questions. That makes it a bot you talk to.

Be careful what you hand over. In July 2025, FINRA warned about auto-trading services that ask for your login. That access puts your money and your personal data at risk.

If you try an agent, set it up like this.

  1. Give it its own small account, not your main savings.
  2. Make it ask you before each trade, if the tool allows it.
  3. Never share your password. Use access keys that cannot withdraw money.
  4. Set loss limits, and check what it did every day.

How to choose an AI trading app or platform

There is no single best AI trading app for everyone, and we do not rank them. Ask these questions of any app, including ours.

  1. Is the firm registered? Look it up on Investor.gov or adviserinfo.sec.gov.
  2. Which of the four jobs does its AI do? If it cannot say, be careful.
  3. Can you read every rule before any money moves?
  4. Can you backtest and paper trade before you use real money?
  5. Who holds your money? Look for a regulated custodian and SIPC membership.
  6. What does it cost in total? Add app fees, fund fees and trading costs.
  7. Can you stop it at any time, and does it have loss limits?

Risks and limits

  • The AI can be wrong. It can misread your idea or give a confident, false answer.
  • Models trained on past data can fail when markets change.
  • Backtests are hypothetical. They can be overfit, use hindsight and leave out real trading costs.
  • A bot acts fast. A bad rule or a glitch can repeat many trades before you notice.
  • Guardrails reduce risk but cannot remove it. A price can gap past a stop-loss.
  • Past results, real or simulated, do not predict future results. You can lose money.

Frequently asked questions

Can AI predict the stock market?

Not reliably. AI finds patterns in past prices, but markets change and patterns fade. The CFTC reminds investors that AI technology cannot predict the future.

Is AI trading legal?

Yes. Using AI to trade is legal in the US, and fraud and securities laws still apply. The firm behind an AI trading service may need to be registered, so look it up first.

Can you make money with AI trading?

Trading can make or lose money, with or without AI. No AI can promise a profit. The CFTC calls claims of high or guaranteed returns red flags of fraud. Test any idea before real money is at risk.

Do you need to code to build an AI trading bot?

No. Some tools, like Amaltash’s Strategy Builder, turn a plain-English sentence into rules. You still need to read the rules and test them, because the AI can misread your idea.

What is the best AI trading app for beginners?

We do not rank apps, and no app fits everyone. For a beginner, look for a registered firm, rules you can read, paper trading and loss limits. Then start small.

How much money do you need to start AI trading?

You can start with none by paper trading, which uses pretend money. With real money, use only an amount you can afford to lose. Your broker may set a minimum.

Sources

Share this article

Educational content published by Amaltash Advisors LLC. Nothing here is investment advice or a recommendation to buy or sell any security. All investing involves risk of loss, including loss of principal, and past performance does not guarantee future results.

Get started