📰 AI News: Is AI Really “Bigger Than COVID”? The Graph That Has Economists Freaking Out

🔥

Nov '25 • AI News

📝 TL;DR

A new analysis suggests AI capabilities are doubling roughly every seven months, following the same kind of exponential curve that made COVID explode out of nowhere. The expert behind it says if this keeps going for just five more years, AI could reshape the economy far more deeply than the pandemic did.

🧠 Overview

A UK tech founder and former government adviser has shared a graph comparing how fast AI is improving with how COVID spread in early 2020. The pattern looks strikingly similar: slow, slow, slow... then suddenly everything at once. His core point is simple. AI might feel a bit messy and toy-like now, but the curve underneath is moving very fast, and we are not planning for what that actually means.

Graphs Mentions in this post: https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/

📜 The Announcement

Using data from AI research on software development tasks, the analysis shows AI systems are already matching or beating humans on more and more coding problems.

The key statistic: AI capability in this area appears to be doubling roughly every seven months, and could keep doing that for at least another five years. The expert argues that, unlike COVID which was a temporary shock, AI looks set to cause a permanent reshaping of how economies and workplaces operate.

He also warns that progress will only slow when we literally hit physical limits such as how much energy is available to train these models.

⚙️ How It Works

👉 The graph tracks how often AI models successfully complete real software development tasks compared to humans.
👉 Each new generation of models solves a bigger share of these tasks, and the improvement forms an exponential curve rather than a straight line.
👉 Exponential here means “doubles over and over”: at first it looks boring, then suddenly it looks like it went vertical.
👉 The same pattern tripped up leaders during COVID because the real danger shows up only at the end of the curve, when it is already huge.
👉 For AI, the “curve” is about capability, not infections. It is the range of things AI can do well enough to be genuinely useful.
👉 The expert estimates several years more of this doubling are plausible before things hit energy and infrastructure limits.

💡 Why This Matters

🎯 We are bad at intuitively understanding exponential change. Our brains expect linear progress. That is why AI still feels like a toy to many people even while the underlying curve is quietly getting steeper. If you wait until it feels obviously huge, you are already late.
🧱 AI still looks flawed, but that is not the full story. Hallucinations and mistakes make it easy to dismiss AI as unreliable. The graph says “yes, today is messy, but the direction and speed of improvement are the real headline.”
🧭 The comparison to COVID is about speed and disruption, not doom. COVID was a fast shock. AI is a fast transformation. One shut things down, the other will rewire how things run. That is a different kind of disruption, but just as important to plan for.
🔍 Energy and hardware will become real constraints. At some point, training ever-bigger models runs into physical limits like electricity and compute. That means governments and companies will have to think hard about where to invest and what capabilities actually matter.
🧓 Your skills will age faster without active adaptation. If AI keeps doubling at this pace, a “good enough” skillset today could feel outdated much sooner than you expect. The good news is that learning to use AI gives you leverage rather than replacing you.
🛟 Planning beats panic. The lesson from COVID was that ignoring exponentials is expensive. With AI, we have a clearer line of sight. The smart move is calm preparation, not fear or denial.

🏢 What This Means for Businesses

📈 Assume AI capability in your niche will double, then ask “what breaks?” If AI in your industry got 2x better every seven months, which parts of your workflow would be obviously automatable in 2 to 3 years? That is where you should experiment now, not later.
🧰 Treat AI as a junior team you are training, not a threat you are dodging. The companies that win will put AI into daily operations as a co-pilot for writing, research, coding, analysis, and customer support. Waiting until it is “perfect” means gifting your advantage to competitors.
🧪 Run small, low-risk experiments instead of big, vague “AI strategies.” Pick one process. Document it. Then explore how AI can draft, summarise, or monitor that process. Measure time saved or quality improved, then scale what works.
🎓 Upskilling your people is the real moat. Tools will keep changing fast, so betting on a single platform is risky. Training your team to think in prompts, workflows, and checks creates an adaptable organisation that can ride the curve instead of being crushed by it.
💶 Budget for AI like you did for the cloud. AI is moving from “nice extra” to core infrastructure. Even small businesses will benefit from setting aside a modest, recurring budget for AI tools, experiments, and training.
🤝 Engage with regulation and policy, do not wait for it to happen to you. Governments are starting to invest and build institutes around AI safety and security. If your work touches regulated areas, get curious now so you are not blindsided later.

🔚 The Bottom Line

AI today can still feel clumsy, but the underlying curve is moving in a way that could reshape work and the economy far more deeply than COVID did. The big risk is not that AI suddenly “takes over” tomorrow, it is that people and businesses ignore the exponential until the change feels uncomfortably fast.

You do not need to predict the whole future to win, you just need to start using AI intentionally as your co-pilot and adapt a bit faster than the average person.

💬 Your Take

If AI really does keep doubling in capability every seven months for the next few years, what is one part of your work or business you would most want to future proof first?

2 comments