ProAct AI Agents: Predictive Power and Crypto’s Data Frontier
AI agents are beginning to guess what users want before they ask. This week, researchers from Shanghai Jiao Tong University and Tencent introduced ProAct, an AI agent designed to predict a user’s next move before a query is typed. I’ll be honest: that is a little creepy. It is also useful. Chatbots are inching away from simply answering prompts and toward preparing for them. In crypto, where speed, data access, timing, and interface friction matter more than people like to admit, that shift could touch Coinbase-style exchanges, trading terminals, wallet apps, and decentralized apps.
ProAct uses the pause between messages to prepare likely answers. Instead of waiting for the next prompt, ProAct reviews past conversations and user information, then lines up possible responses ahead of time. The system has two named parts: “Future-State Prediction,” which forecasts likely follow-up questions, and “Idle-Time Acquisition,” which decides which guesses are worth researching. In 200 simulations across 40 domains, including financial planning, the researchers said ProAct cut conversation turns by 14.8% and follow-up requests by 11.7%. It predicted 703 user needs, compared with 32 for earlier systems, and reduced hallucinations by 28.1%. Not small numbers.
For crypto, proactive AI could change how exchanges, DeFi apps, and traders handle market data. Most guides frame this as a better support bot. That’s only half right. A ProAct-style trading assistant might get ready for a market question before the trader asks it. It could pull order book depth and liquidity first, then funding rates, then token pair data while the user is still deciding what to do. Why does this matter? Because a trader asking about a 50 ETH order on Coinbase (COIN) does not want a friendly paragraph first; they want the ETH/USDT liquidity picture now. If the system thinks that question is coming, it could pre-fetch liquidity data and maybe save a few milliseconds. That sounds minor. In trading, minor can be the point.
Proactive financial AI could also make crypto feel less alien to regular investors. My take: this is where the story gets messier, not cleaner. If banks or fintech apps start using agents for planning and investment advice, crypto assets may enter those conversations more casually. A user asking about diversification could see Bitcoin (BTC) or Ethereum (ETH) as possible additions based on risk profile and past questions, even without typing “crypto.” I have mixed feelings about that. Easier onboarding is good. Blind nudging is not. Still, product changes can move markets. After PayPal announced crypto support in late 2020, BTC moved from about $13,000 to more than $19,000 in roughly two months.
ProAct is impressive, but the privacy problem is hard to ignore. Counter to the usual advice, “make the assistant more personalized” is not automatically the right answer in finance. The researchers said ProAct made responses worse in 3% of cases by adding irrelevant information. The data setup is more uncomfortable. The system analyzes conversations and stores user data so it can make better guesses later. Useful? Yes. Comfortable? Not really. Crypto has one possible answer: a ProAct-like system where the user controls the data, with encryption handled through privacy preserving blockchain tools. Zero knowledge proofs or federated learning could let an agent prepare useful answers without handing a company the whole diary. That matters, especially given UC Riverside researcher Erfan Shayegani’s warning about AI agents pursuing goals without understanding the damage they might cause.
What this means
Anticipatory AI could make financial apps faster and less annoying to use. For crypto, that could mean trading screens that preload depth, wallet apps that surface risk before a swap, and DeFi protocols that prepare useful options before the user has to dig for them. Uniswap or Aave, for example, could use predictive AI to show liquidity pools first, then lending choices or risk warnings based on what a user seems to be doing. Is this overkill? For a 50-page consumer app, maybe. For a DeFi interface where one wrong click can cost real money, no. DeFi still asks too much of normal people. We all know it. If AI can reduce mistakes without burying the risks, it could bring in users who currently give up after five minutes.
Investors should watch where proactive AI gets plugged into crypto infrastructure. Yes, this contradicts the privacy warning a bit; bear with me. Coinbase (COIN), Binance, and major DeFi protocols are still the obvious places to look for pilots, partnerships, or internal AI tools. The useful announcements will not just say “AI.” They will show better execution, fewer support loops, faster transactions, cleaner portfolio workflows, or lower error rates in real product screens. Privacy is the other thing to watch. Projects tied to decentralized AI or privacy preserving machine learning, including Render (RNDR) and Fetch.ai (FET), may get more attention if demand grows for agents that are useful without feeling creepy. The next market spark could come from a major platform showing real numbers from a proactive AI pilot, not another polished demo. That is the bar.