Ann
The creator of the AI bot gave the robot a clear condition: “Pay yourself or die”, so if the balance reaches zero, That agent’s system has been shut down.
According to a publication dated February 10 of this year, the agent Autonomous traders in prediction markets De Polimarket, In other words, it is a platform where participants buy and sell contracts associated with future events, such as sports results or weather data.
However, CriptoNoticias could not confirm whether it was an environment using real money or a simulation.
Bots not only survive; to cover one’s own operating costsThis includes the use of artificial intelligence models and the infrastructure on which they run.
Argona messages maintain that every 10 minutes the system has analyzed: « 500 and 1,000 active marketswe created a fair value estimate and performed the operation if a deviation of more than 8% was detected.
The following image is a screenshot taken from a video published by Argona that shows the gains achieved.
How did the AI agent work?
According to the users themselves, the agent Claude Application Programming Interface (API) (language model developed by Anthropic) for inference. That is, the system referred to an external AI model to evaluate probabilities and paid for that service with a portion of the profits.
Additionally, the algorithm applied the Kelly Criterion, a mathematical risk management formula that calculates what percentage of capital should be staked based on the estimated advantage. In practice, this limits each position to 6% of the available capital. reduce the likelihood of bankruptcy In the face of adversity.
The agents used by Argona acted as periodic scanners to detect specific inefficiencies. In the weather market, for example, the bot analyzed data from the National Weather Service, he explained. before it is fully reflected in the price. Polymarket.
Similarly, in the sports market, it investigated injury reports to predict movement, and in markets related to cryptocurrencies, it combined on-chain metrics with sentiment analysis, Algona said.
The usefulness of this method is that if the information arrives before the market average, You can capture the price difference.
In that sense, CriptoNoticias reported at the end of January that the implementation of the ERC-8004 standard on the Ethereum network has enabled the operation and interaction of AI agents, allowing them to perform transaction tasks similar to those mentioned by Argona, among other things.
Experimental infrastructure and limitations
The user said he developed the agent in Rust, a performance-oriented programming language, and ran it on a “virtual private server (VPS) for $4.50 per month.” This means that the technical barrier is not the cost of infrastructure, but the quality of predictive models and risk management.
To tune the system, we used Openclaw. Openclaw is a platform that enables the deployment of autonomous agents. Interact with artificial intelligence models and perform external tasks.how to work with the financial API.
In reality, Openclaw serves as an environment where agents make decisions and execute market orders.
However, in fluid and competitive markets, persistent inefficiencies tend to be corrected quickly, making it difficult to sustain these gains on an ongoing basis.
Therefore, while this experiment shows how agents can automate analysis, risk management, and execution, the real challenge is to verify whether these results are repeatable over time or responsive to specific market conditions.