AI trading bot: what it can and cannot do (2026 guide)

An AI trading bot is a program that decides when to buy and sell and places those orders on your brokerage account without waiting for you. The phrase covers everything from a fixed moving-average rule to a language model reading news, and most products blur the line on purpose. This guide separates the kinds, says what to expect from each, and gives you the questions that expose a bad one.

Three things people mean by "AI trading bot"

Rule-based (algorithmic) bots. Fixed conditions — price above its 50-day average on rising volume, buy; stop 4% below; exit at target or after 40 sessions. No learning after deployment. Transparent, testable, and where most retail automation actually is. Firefly's 1,000+ strategies are this kind.

Machine-learning bots. Models trained on history to pick among rules, size positions or forecast short-term direction. Can adapt; can also overfit. The honest ones publish out-of-sample results and re-train on a schedule you can see. Firefly's swarm layer belongs here: it does not predict prices, it re-weights the rule-based strategies by how they have done recently.

LLM "agents". A language model reads news, filings or your prompt and proposes trades. Good at explaining; unproven at executing. Treat any product that lets a chat model place orders directly as experimental.

US and India: how the plumbing differs

In the US, a retail bot connects to a broker such as Alpaca or Interactive Brokers with an API key, trades stocks, ETFs and options, and pays no exchange-level algo registration. Pattern-day-trader rules and margin requirements still apply. In India, the bot connects to a SEBI-registered broker's API (Angel One, Zerodha), the machine placing orders needs a static IP, orders above an exchange threshold require the algorithm to be registered through the broker, and third-party providers are empanelled with the exchange. SEBI's 2025 rules explained. Either way the money stays in your own account and the key cannot withdraw.

Seven questions that expose a bad bot

• Can I see every trade? A full log with entries, exits, stop hits and drawdown. A return figure without a drawdown next to it is marketing.
• Which months lost? If the answer is "none", leave.
• Where is the stop? Every rule should have one, and the portfolio should have a drawdown limit on top.
• Whose account are the orders in? Yours, through an API key with no withdrawal right. Anything pooled is a fund, and needs a licence.
• What does "AI" mean here, precisely? Which model, trained on what, re-trained when. Vague answers mean a rule set with a label.
• What happens if the provider goes down? Open positions should sit with their stops at the broker, not in the provider's memory.
• How is it billed? Subscription, profit share, or brokerage markup. Markups hide the real cost.

Where Firefly sits

Firefly is a rule-based engine with a performance-weighted allocation layer, trading on your own broker account in India, the US and crypto. Its daily output is public on Signals, so you can watch it for months before connecting anything. The AI products in development — Advisor, Coach and Allocator — explain, teach and analyse; they do not place trades, and we say so on every page.

Questions people ask

Fintrens Technologies Pvt Ltd is not a SEBI-registered investment adviser or research analyst. This guide is general information, not advice; trading involves risk of loss.