ASUS and crypto firm Poesis completed a week-long proof-of-concept in which autonomous AI agents conducted market research, managed risk and executed real trades on an ASUS ExpertCenter Pro ET900N G3 laptop equipped with NVIDIA technology. The trial was not a long-term performance evaluation and was not designed specifically for cryptocurrency trading, and the companies have not disclosed product names, financial terms or a commercial timeline. The experiment illustrates the shift from AI-assisted analysis toward agentic finance, in which systems continuously collect data, develop strategies and act within defined permissions. ASUS is also advocating local deployment, arguing that constant background API calls can make cloud-based agents difficult to price. It estimated that an ASUS Ascent GX10 would require about $4,000 upfront and $384 in annual electricity costs, versus $6,000 to $8,000 in annual API fees for a comparable cloud workload. The GX10 was not used in the Poesis test. Financial institutions and crypto exchanges are building similar infrastructure, but adoption remains bounded by account separation, spending limits and human accountability. Binance’s Agent OS, launched on August 20, blocks withdrawals and applies limits including $50,000 a day for swaps, $100,000 by default for DeFi transactions and $20 for x402 payments. Kraken and OKX rolled out related infrastructure in March, Bitget and Gemini in April, and Coinbase in June. A March 2026 paper described agentic finance as the next stage after rule-based trading and machine-learning prediction, spanning data perception, reasoning, strategy generation and controlled execution. A Cambridge Judge Business School report found that 52% of surveyed firms had experimented with or deployed agentic AI. Risks include crowded trading responses, unclear accountability and limited blockchain settlement capacity if machine-driven activity scales.