In unfamiliar market conditions, historical data-driven AI trading bots will falter

Today’s AI trading bots are based on a limited amount of historical data which means totally unfamiliar market events like the 10/10 liquidations or even last week’s severe selloffs will leave agentic trading models short of the mark.

These historical data-driven AI models have never seen huge liquidations on a single day and would find this “very unfamiliar” said Bitget CEO Gracy Chen on a panel about agentic trading bots at Consensus Hong Kong 2026. As such human intervention is needed.

“As an exchange, we don’t plan to build our own LLM [large language model]. But trading bots are a big thing,” Chen said. “Current AI bots are a bit like an intern: faster, cheaper but needs a little supervision.”

However, further down the line this will be more like a “full employee,” and in 3-5 years AI can replace a lot of us, Chen said.

These are sentiments heard regularly in the algorithmic trading world when it comes to AI.

While complex LLM and machine learning trading technology is improving at a rapid clip, there are still plenty of people who think a human overlay is an essential part of the process – particularly in situations like the severe volatility that recently gripped crypto markets.

Joining Chen on the panel, Saad Naj, founder and CEO of agentic trading startup PiP World agreed the tech is in its infancy and that comes with risk. But he pointed out that 90% of day traders and retail players lose money.

“As humans we are too emotional. We can’t compete with AI solutions,” Naj said.