John Burn-Murdoch

almost second nature to think, well, you know, do we really need that many people who are just going to be pinging me documents over Slack?

So I think there's certainly something there.

But the way to solve this is probably to go back to more firms where you have three or four days in the office, rather than saying five, which is where you then go over the peak and start getting into other problems.

It's tricky.

It's about trying to find the best data with the highest signal-to-noise ratio for every time we ask this question.

A few months ago, the best data we had was

The amount of activity that software engineers were performing, those using agentic AI versus those that weren't.

And that gave us a sense that, well, this does seem to be boosting productivity.

People are shipping more code.

But a few months later, we're now able to look at that further.

full pipeline, as it were, from lines of code written, pieces of software released, to how much those new pieces of AI-assisted software are actually getting used.

And it turns out that when you follow that pipeline to its end, where you really are capturing value, there's been very little uplift at all.

So it's constantly challenging.

And I see this in my own work as well.

I'm a fairly heavy user of the agentic AI tools.

And I definitely feel more productive.

Between me and Claude or Codex, we're producing loads of stuff.

But at the end of the week, I look back and I think, well, I still did two articles.

And there are so many jobs, I think, where that is true, where...

activity sort of low level activity does not always translate or frequently doesn't translate into high value that your, in my case, readers or in someone else's case, customers, clients will consume.