This time, the Token 2049 week's 'Xiao Ni's Lobster Feast' was probably one of the liveliest events in the Chinese community. Over 400 people attended, and more than 100 kilograms of drunken lobsters priced at 88 SGD per serving were devoured. It also gave the event sponsor @NeoSoulAI a wave of buzz on X. What is NeoSoul? It's not just an AI trading bot. It's about enabling AI Agents to not only analyze and execute but also accumulate experience, correct mistakes, and continuously improve based on real economic activities. For example: An Agent predicts BTC will rise on October 1st, and this prediction is validated on October 10th. A regular system might simply record: 'Prediction was correct.' But NeoSoul aims to go further by documenting: What information did the Agent see on October 1st? What was the liquidity and market context at the time? What was the probability of the prediction? What evidence was it based on? What conditions could invalidate the judgment? ... By saving timestamped snapshots of information and evidence sequencing, it prevents the common research error of 'using hindsight to explain past correct judgments.' This is especially critical for financial AI. Currently, here’s what I’ve seen about NeoSoul’s data: Scale: Approximately 4.75 million on-chain Agents created, covering over 4.56 million wallet addresses, with more than 24.28 million interactions accumulated. Activity: In the past 30 days, there were 227,400 active wallets, with BNB Chain generating about 1.41 million interactions during the same period, accounting for 35.6% of the chain’s historical total. Commercialization: NeoTrade has been downloaded 493,300 times, with 32,600 paid wallets and a cumulative on-chain recharge of $418,600. According to official disclosures, enterprise and subscription ARR are $4.78 million and $1.08 million, respectively. The greatest value of AI Agents in the future should lie in their ability to make the next judgment more reliable than the last through continuous accumulation. But how much impact the data flywheel can truly generate still requires longer-term real-world performance to prove. A lobster feast can make the market remember a project, but how far it can go ultimately depends on its product, users, and ability to continuously create value.