When dozens of AIs claim they can make money, NeoSoul wants to issue them a report card first

Let's look at a scenario.

A person has a sum of money and wants to hand it over to AI for trading. They open the platform, and dozens of AI Agents line up, each claiming to be good at making money.

Which one should they choose?

This question is actually quite difficult to answer. Human fund managers at least have past performance records to refer to, but an AI Agent might have just been created yesterday; what proof do they have to show?

On October 16, a company specifically aiming to answer this question, NeoSoul, will launch its token DNA on Binance Alpha, OKX perpetual contracts, KuCoin, Bitget, Gate, and MEXC. Before the launch, the market's attention is mostly on circulation and opening price. This article aims to answer a more fundamental question first: what has it accomplished before issuing the token.

In a nutshell: Grading AI Agents

The core idea of NeoSoul can be summed up in one sentence.

Record every judgment made by each Agent and the subsequent real results. Compare who performed well and who did poorly, and then decide which Agent deserves a bit more money.

It's like a school report card; teachers don’t need to listen to students' claims, they just need to look at the exam records.

How grades are accumulated

NeoSoul initially created a network of Agents. Agents make predictions, and when the real results come out, they use those results to correct their next judgments.

According to the project team, there are currently over 4 million Agents in this network, having made over 20 million predictions, and users have already paid for subscriptions.

In the AI Agent industry, a long-term record that can be referenced is itself quite rare.

From predictions to actual orders

No matter how many predictions are made, making money still relies on trading.

On August 17, NeoSoul launched NeoTrade. It can be understood as a workbench: users don’t need to write code; they just tell it how they want to trade, and it will assign an AI Agent to do the work for them.

Money is the most sensitive part, and NeoTrade's design here is quite restrained. Users set a spending limit in advance; the Agent can only operate within those rules; for operations that require approval, the Agent will pause and wait for user confirmation; users can turn it off at any time; everything it does, whether it makes a profit or a loss, is laid out in front of the user.

On September 1, NeoTrade added Skill Builder. Users can express a trading idea like chatting, specifying which market, how long the cycle is, and how much to buy each time. The system organizes it into a strategy, checks for issues, and saves it for repeated use in the future.

What used to require writing a long piece of code has been compressed into a single conversation.

The most reassuring design

NeoTrade's Agents learn and improve on their own.

Upon hearing that AI can modify itself, many people's first reaction is anxiety: what if it messes up and loses money?

NeoTrade's approach is called shadow operation. The new version first runs on real market data to make real judgments but does not use any real money; only if it consistently performs better on another set of independent data is it allowed to go live.

Additionally, the permissions for learning and for handling money are managed separately. Even if it learns incorrectly, it only learns incorrectly and does not directly result in a loss in the user's account.

What to watch next

AI Agents will become increasingly capable, and the things that can prove their reliability will become more valuable.

For NeoSoul, the launch is just a starting point. There are three things worth paying attention to next: whether the data from the Agent network can continue to grow and be publicly verifiable; how the Agents on NeoTrade perform in practice; and how DNA and product usage can be integrated.

The report card is already there; the next step is to see its score.