cd /news/large-language-models/brett-harrison-critiques-llms-abilit… · home topics large-language-models article
[ARTICLE · art-65709] src=cryptobriefing.com ↗ pub= topic=large-language-models verified=true sentiment=↓ negative

Brett Harrison critiques LLMs’ ability to build trading systems, says human expertise remains essential

Brett Harrison, former FTX US president and Jane Street veteran, argues that large language models are fundamentally unsuited for building high-frequency trading systems because market data is stochastic, not linguistic. He says LLMs can assist with code generation and feature selection but should not be used for core quantitative models or real-time trading operations. Harrison's firm Architect Financial Technologies raised $35 million in December 2025 to build trading infrastructure blending human expertise with crypto-native innovation.

read3 min views1 publishedJul 20, 2026
Brett Harrison critiques LLMs’ ability to build trading systems, says human expertise remains essential
Image: Cryptobriefing (auto-discovered)

The former FTX US president and Jane Street veteran argues that large language models fundamentally misunderstand financial markets because market data isn't language

Brett Harrison wants to pump the brakes on the AI-will-replace-traders narrative. The founder and CEO of Architect Financial Technologies laid out a detailed case in a recent Medium post arguing that large language models, the technology behind ChatGPT and its competitors, are fundamentally ill-suited for building the kinds of trading systems that actually make money in high-frequency environments.

Coming from someone who spent 11 years at Jane Street leading algorithmic trading system development, the critique carries more weight than your average LinkedIn hot take about AI.

The core problem: markets aren’t language #

Harrison’s central argument is elegantly simple. LLMs are built to process and generate language. Financial market data is, by its very nature, not linguistic. It’s stochastic, meaning it involves randomness and probability distributions that behave nothing like the patterns found in human text.

He’s not dismissing AI entirely, though. Harrison acknowledges that LLMs can be genuinely useful for specific supporting tasks. Code generation, for instance. Feature selection, the process of identifying which variables matter most in a model, is another area where LLMs can add value. Harrison argues that LLMs should never be relied upon for building core quantitative trading models or for real-time operations like continuous market monitoring. High-frequency trading operates on timescales of microseconds to nanoseconds. LLMs, which require meaningful computation time to generate responses, simply cannot operate at those speeds.

Why Harrison’s background matters here #

He studied computer science at Harvard with a focus on AI, then spent over a decade at Jane Street, one of the most respected quantitative trading firms in the world. After that, he served as president of FTX US before departing in 2022, prior to the exchange’s spectacular collapse under Sam Bankman-Fried.

In early 2023, Harrison launched Architect Financial Technologies with backing from notable crypto-native investors including Coinbase and Circle. The firm raised a $5 million seed round in January 2023, then followed up with a $35 million funding round in December 2025. The company’s ambition is to build trading infrastructure that brings crypto-style market design, things like perpetual futures, into the traditional finance world.

Rather than pursuing fully autonomous AI trading, the firm is focused on building robust infrastructure, including a regulated perpetual futures exchange, that blends human expertise with technological innovation.

What this means for investors and the broader market #

For investors evaluating AI-focused trading platforms, Harrison’s argument serves as a useful filter. Companies claiming that LLMs alone can generate alpha, the excess returns above a benchmark, deserve extra scrutiny. The technology has genuine applications in finance, but those applications are narrower and more specialized than the marketing materials suggest. The funding trajectory of Architect itself tells a parallel story. The jump from a $5 million seed to a $35 million round suggests growing investor confidence not in AI-does-everything approaches, but in infrastructure plays that thoughtfully combine traditional finance expertise with crypto-native innovation. Perpetual futures, a staple of crypto exchanges, represent a massive derivatives market that traditional finance has barely begun to tap.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our

Editorial Policy.

── more in #large-language-models 4 stories · sorted by recency
── more on @brett harrison 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/brett-harrison-criti…] indexed:0 read:3min 2026-07-20 ·