According to @MRRydon, frontier language models now sustain about five hours of independent task execution at roughly 50% success, shifting AI from short prompts to full work sessions and making duration a more trading-relevant metric than marginal accuracy gains, source: @MRRydon. He adds that the doubling time for task duration has compressed from about seven months historically to nearer four months recently, signaling a steepening exponential curve, source: @MRRydon. He argues that if this trajectory persists, agents will soon handle full-day problem ownership with continuous research loops and end-to-end software build-test-deploy cycles that many market participants are not yet pricing, source: @MRRydon. For crypto and AI-equity traders, the core signal to monitor is time-to-autonomy moving from minutes to hours to full-day runs as a leading indicator of automation-driven productivity shocks and narrative momentum, derived from the timeline compression he highlights, source: @MRRydon. He frames 2026 not as a discrete breakthrough but as the curve continuing its current pace, an inflection that becomes obvious only in hindsight, reinforcing urgency in positioning before the final doubling, source: @MRRydon.