A Number That Should Change How You Think About Your Career
7 months.
That's how long it takes for AI's ability to autonomously complete tasks to double. Not get 10% better. Double.
This finding comes from METR, an organization that rigorously measures AI performance on real-world tasks. They've been tracking how long AI can work autonomously — measured in how much time a human expert would need to complete the same task — and the trend line is remarkably consistent.
In 2023, AI could reliably complete tasks that took a human a few minutes. By 2024, it could handle tasks taking 30-60 minutes. By early 2026, it's handling tasks that take humans several hours.
Extrapolate the trend: by late 2026, AI will be tackling 8-hour workstreams. By 2027-2028, multi-day projects.
This isn't science fiction. It's the trendline. And while trendlines can break, this one has been remarkably stable across multiple generations of AI models.
Why 7 Months Changes Everything
Most people think about AI disruption on a 5-10 year timeline. "Sure, AI will be important eventually, but I have time to prepare."
The 7-month doubling smashes that assumption.
Let me make it concrete. Think about a task in your job that AI currently does okay at. Not great, not terrible — maybe 40% as good as you. You look at it and think, "Eh, AI's not ready for this yet."
In 7 months, AI will be at ~80% capability on that task.
In 14 months, it'll be at ~95%.
In 21 months, it'll be better than most humans.
Three cycles of doubling turn "AI can't really do this yet" into "AI does this better than most people." And three cycles is less than two years.
Now think about a task AI does at 70% capability today — something it's already decent at but still needs human oversight. In 7 months, that's at ~95%. In 14 months, it's at near-perfect quality with minimal oversight.
The window between "AI is not a threat to this task" and "AI handles this task routinely" is not a decade. It's 18-24 months. For many task categories, it's happening right now.
The Categories Moving Fastest
Not all capabilities improve at the same rate. Here's a rough ranking of where AI improvement is fastest to slowest, based on benchmark trajectories:
Moving fastest: Code generation and software engineering. AI went from "helpful autocomplete" in 2024 to "builds and ships production-quality software" in 2026. Frontier models write 25-30% of new code at Google and Microsoft. This is the canary in the coal mine for knowledge work broadly.
Moving fast: Structured data processing, mathematical computation, and information retrieval. AI is already at 85-90% capability in these areas and improving. Financial modeling, data analysis, research synthesis — all approaching human-level reliability.
Moving steadily: Text generation, routine communication, and teaching/tutoring. These are already strong (85% capability for writing) and continuing to refine, especially in nuance, tone, and audience awareness.
Moving slowly: Strategic reasoning, creative ideation, and complex judgment. These are at 35-55% and improving, but the improvement curve is gentler. The gap between "AI can generate plausible strategic options" and "AI can make the right strategic choice" is a hard one to close.
Barely moving: Physical dexterity and in-person presence. At ~5% and unlikely to impact labor markets before 2030. If your job requires physically being somewhere and doing something with your hands, AI is not your near-term concern.
How to Read the Clock for Your Career
Here's a framework I use with professionals in my workshops:
Step 1: List the 5-7 major task categories that make up your work week. Be honest about time allocation. (For most knowledge workers, this looks something like: communication 25%, analysis 20%, meeting/coordination 20%, writing 15%, decision-making 10%, other 10%.)
Step 2: For each task category, estimate where AI currently is on a 0-100% capability scale. Use the benchmarks above as a guide.
Step 3: Apply the 7-month doubling. Where will AI be on each task in 14 months? 21 months?
Step 4: Identify the "crossover tasks" — the ones where AI will reach near-human capability within your planning horizon. These are the ones to focus on.
Step 5: For each crossover task, decide: Am I going to orchestrate AI on this task (staying involved but directing the AI)? Or am I going to shift my time to the tasks where AI can't match me?
This exercise usually produces a moment of clarity — and sometimes a moment of mild panic. Both are healthy.
The Common Mistake: Waiting for "Ready"
The biggest career mistake I see right now is waiting for AI to be "ready" before adapting.
"AI isn't good enough yet for my specific work" is a statement I hear every week. And it's usually true — today. But remember the doubling. The task AI does at 50% quality today will be at 100% in 14 months.
If you wait until AI is good enough to threaten your work before you start adapting, you've waited too long. The time to adapt is when AI is at 40-50% — good enough to be useful as a collaborator but not yet good enough to replace you. This is the sweet spot for developing orchestration skills, building AI-augmented workflows, and positioning yourself as someone who works with AI rather than someone who will be working against it.
The Optimistic Reading
The 7-month clock isn't only a threat. It's also an opportunity, if you move quickly.
Because here's the thing: the organizations and individuals who learn to work with AI during the doubling period have a compounding advantage. Each improvement in AI makes their existing workflows more powerful. Each new capability opens new possibilities they can exploit because they've already built the muscle.
The person who started orchestrating AI when it was at 40% capability has 14 months of practice and workflow refinement by the time it reaches 90%. The person who waited until 90% to start is now scrambling to learn what their competitor mastered months ago.
In a world of exponential improvement, the early adopters don't just have a head start. They have a compounding head start.
The clock is ticking. But it's ticking for you if you start now — and against you if you wait.
Seven months. That's one doubling. What are you going to do with it?