Anthropic just made an admission the AI industry has been dreading: their own AI systems are now making AI research happen roughly twice as fast as it did a year ago.
This isn’t a theoretical milestone. METR, an independent AI safety research organization, spent months conducting a deep audit inside Anthropic to measure exactly this. What they found is striking. Between October 2025 and October 2026, Anthropic’s rate of AI capabilities progress doubled compared to the year before, and the primary driver is AI automating the research itself, not bigger teams or more computing power.
Researchers at Anthropic are reporting around a 12x personal productivity boost from using the company’s own internal AI models. Think about what that number means. One engineer is now getting done what twelve engineers would have accomplished a year ago. Not because the team expanded. The AI handles so much of the thinking, writing, testing, and iterating that individual researchers now operate at a fundamentally different speed.
That gap will keep widening.
The technical term for this is “recursive self-improvement,” meaning AI systems are getting better at making AI systems better, and each new generation helps develop the next one more quickly than the last. METR researchers found no evidence the trend is slowing. If anything, as of late 2026, the AI contribution to research speed appears to still be growing.
For a business that has nothing to do with building AI models, this still matters quite a bit. The pace at which new tools become available on the market is set to accelerate, compressed by the same feedback loop now driving research at frontier labs. Six months from now, capabilities that seemed two years away could realistically arrive. That’s a planning reality, not a sales pitch.
There is also a harder truth embedded in this news. If AI development is accelerating this quickly, the window to thoughtfully build an AI strategy inside your organization is narrowing faster than most executives realize. Waiting for the right time carries real cost. Organizations that have been experimenting with AI-assisted workflows for six months will have hard-won operational experience that late movers simply cannot shortcut.
Start small. Start now. Pick one workflow your team repeats every week. Find an AI tool that handles at least part of it, then watch closely how your people interact with the assistance, what they trust, what they question, and where the remaining gaps are. That knowledge compounds. The companies positioned well in 18 months are the ones systematically building it today.
An independent audit of one of the world’s leading AI labs just confirmed that AI progress is no longer running on a years-long clock. That changes when you need to act.
Want to explore how AI automation could benefit your business? Let’s talk.