The Doubling Schedule
The METR benchmark is unforgiving: AI task completion capability doubles roughly every 7 months. Not years. Months. And on some benchmarks, we're seeing acceleration—doubling in under 3 months.
Here's what that timeline looks like:
- Early 2026 (now): 1-3 hour tasks run reliably. 4-8 hour tasks possible with recovery loops. 8+ hour tasks emerging but inconsistent.
- Late 2026: 8-hour autonomous workstreams baseline. Chaining multi-step processes standard.
- 2027-2028: 20+ hour workstreams. Multi-day autonomous operation. Minimal human intervention.
What This Means for Your Exposure
A task that's currently 40% automatable isn't "partially safe." It's a 18-month countdown. In 18 months, that task will be roughly 85% automatable. Not 40%.
Your AIGMI score has an expiration date. It's not a permanent assessment—it's a snapshot of the current moment. The exposure measured in February 2026 is not the same as February 2027.
The Real Risk: Complacency
If you score 45/100 today and think you're "safe," you're not thinking exponentially. You're thinking linearly. At 7-month doubling, the gap between 45 and 75 is just over a year. The boundary between "affected" and "deeply exposed" moves constantly.
The safe move isn't to find jobs with low exposure. It's to build skills and complementarity fast enough to outpace the doubling curve.
Your Adaptive Response
Check your AIGMI score quarterly, not annually. When benchmarks improve, your score changes. Build redundancy into your value proposition—develop skills that aren't on the AI roadmap yet. Focus on areas where AI will need human judgment for decades, not just months.
Capability doubles every 7 months. Are you evolving every 6?