Anthropic Claims AI Increases Productivity By 80%, But I Have a Question…
I read Anthropic’s 25-page research so you can read this 2-minute summary instead.
I like Anthropic. Their research is always fun to read… although, by page 5 of their PDF, you’re already wondering if this is what you want to be doing with your life. But I guess that’s why they call it research.
On the bright side, it gives people like me a solid excuse to rewrite the whole thing into a compact, relatable summary, which is what we have here. You’re welcome.
So… Anthropic recently dropped a study, titled “Estimating AI productivity gains from Claude conversations”, they analysed 100,000 actual conversations with Claude to answer a simple question: how much time is AI really saving people?.
What they found…
Without AI, the average task in their dataset would take around 90 minutes.
With Claude, people got the same work done 80% faster.
Meaning, many tasks that took 90 minutes could be done in roughly 18 minutes with AI (on average).
When they broke this down by job type, they found some variations:
Some tasks (e.g. preparing school curriculum), which might take humans hours, could be done by AI in minutes.
Other, simpler tasks (food prep, basic maintenance, etc.) showed smaller gains or didn’t make a big difference.
After all their fancy modelling and economic cross-checking, they concluded that if AI like Claude were widely adopted, it could raise annual productivity by 1.8%. For context, that’s double the usual productivity growth rate in the U.S. since 2019.
Important caveats…
The authors (Alex Tamkin and Peter McCrory) were honest enough to say, “Relax, this is not gospel truth”.
Here are some of the limitations they pointed out:
Their method assumes that all time saved by using AI gets turned into productive work. But in reality, sometimes the time saved equals “let me think about my life for 44 minutes”, which we can’t exactly allocate to productivity.
Claude’s time estimates are based on AI’s opinion of how long humans spend on tasks. And AI isn't exactly known for being street-smart.
The analysis doesn’t cover all types of work. It only uses tasks actually done with Claude (so by people choosing to use the AI), which may bias toward tasks that are already good fits for AI, i.e, tasks AI suck at would have taken longer.
And of course… no research is 100% accurate.
My deductions…
Yes, AI can significantly speed up work. We already knew this, but it’s nice to see numbers attached instead of vibes.
But faster ≠ better. If you save 60 minutes only to spend 45 minutes correcting what the AI wrote, the “gain” is just an anger management session.
My big question…
If AI really makes the average person more productive…
I believe the natural consequence of that is that companies will need to hire fewer people.
Which means… As individuals, we celebrate the speed boost → “My work is faster now! AI is saving my life!”
But on the other side of that same coin… → “Ah, fewer people needed. AI is taking our jobs.”
Same technology. Different beneficiaries. One cup, two sips.
I know productivity and employment don’t necessarily move in a straight line.
But the question is: Would this be a fair deal?
