This year’s “Xiao Ni’s Lobster Feast” during Token 2049 Week was likely one of the most vibrant events in the Chinese-speaking crypto community. Over 400 attendees showed up, consuming more than a hundred kilograms of chilled lobster at 88 SGD per serving—generating a massive wave of mentions for the event’s sponsor, @NeoSoulAI, on X. What is NeoSoul? It’s not just about building an AI trading bot. It aims to enable AI agents not only to analyze and execute trades, but also to accumulate experience, correct mistakes, and continuously improve based on real economic activity. For example: if an agent predicts BTC will rise on October 1st, and that prediction is validated on October 10th, a conventional system might simply record: “Prediction was correct.” But NeoSoul seeks to go further—recording: What information did the agent observe on October 1st? What was the liquidity and market context at the time? What was the predicted probability? What evidence was it based on? Under what conditions would the judgment fail? By saving timestamped snapshots of data and ranking supporting evidence, NeoSoul prevents the common research error of using hindsight—interpreting past correct predictions with information only known afterward. This is especially critical for financial AI. Currently, I’ve observed the following data for NeoSoul: Scale: Approximately 4.75 million on-chain agents have been created, covering over 4.56 million wallet addresses, with more than 24.28 million cumulative interactions. Activity: In the past 30 days, 227,400 active wallets generated 1.41 million interactions on BNB Chain—accounting for 35.6% of the chain’s total historical interactions. Commercialization: NeoTrade has been downloaded 493,300 times, with 32,600 paid wallets and $418,600 in total on-chain deposits. According to official disclosures, enterprise and subscription ARR stand at $4.78 million and $1.08 million respectively. The greatest future value of AI agents lies in continuous accumulation—making each subsequent judgment more reliable than the last. But how powerful this data flywheel truly is will ultimately be proven by long-term real-world performance. A lobster feast may make a project memorable—but what determines how far it will go is its product, its users, and its sustained ability to create value.