|##### FROM THE FRONTIER
Anthropic researcher dismisses talk of an AI slowdown, shows AI is fast approaching human ability|
|
Made with Midjourney|
|Talk of an AI plateau might be premature. Anthropic’s Julian Schrittwieser, who previously also worked at DeepMind, just wrote a blog post that shows AI capabilities are actually doubling at a steady clip — and the implications are significant.
Here’s what’s happening: A METR study shows AI models can now handle software engineering work for over two hours straight with a 50% success rate — double what they achieved just seven months ago. The trend is remarkably consistent, suggesting this isn’t a fluke but part of an exponential growth curve.
AI is no longer limited to coding. OpenAI’s GDPval benchmark tested AI across 44 occupations in nine industries, using 1,320 real-world tasks drawn from professionals averaging 14 years of experience. Claude Opus 4.1 is already “almost matching industry expert performance” across multiple domains, proving that AI is approaching cross-industry proficiency, not just software engineering.
Full-day AI work is on the horizon. If the current pace continues, models could operate autonomously for eight-hour workdays by mid-2026. By year-end, at least one AI may match human experts across many industries. By late 2027, outperforming specialists on complex tasks could become routine, claims Schrittwieser.|