Bombshell Harvard Study Debunks The AI Productivity Narrative

It cost more to run than humans it replaced. @ 8:48 So now you don't hire the graduates and make senior engineers work harder to debug this stuff and create a shortage of mid level and senior engineers in the future.

$6 trillion is a wild market figure for a fancy autocomplete to achieve in just a few years.

In my experience, i ship way more code but a lot of them are just not tracked by AI. In production you have to tweak so many things to improve the process, most of them don't warrant a jira ticket.

Is having more comments on PRs a bad thing? People miss things in review all the time, or just let them go because they don't want to sweat the small stuff. If the comments are wrong, obviously that's bad. But if they're correct then this is what i want from static analysis. Otherwise it's just mistakes getting through human review, no?

AI is a juicero for capital to squeeze labour

Companies are extremely hamstrung by all their BS, office and “productivity tools”, complex processes, handoffs and too many chefs and every decision by committee… hence a one line change can often take a week to get into prod, no matter how it is done 😅 in these cases AI or any other tech improvement makes absolutely no difference to “productivity”

Well I am pretty sure this is AI generated and Story Blocks also

So you're publishing a video about how AI isn't all that it's cracked up to be and you're sponsor is a company that sells AI tools. What's wrong with this picture? Oh. And the English language narration is obviously AI generated. You can't have it both ways.

Harvard itself is not productive and not as prestigeous as it used to be.

Biggest question mark about this study: their sample is from April to December 2025. Given the pace of AI improvement, would they get the same result today?

I work with software, this is no surprising to me. As parts of development get easier, there are more steps we need to follow: more testing, more Q&A, more design, more meetings, etc. the time to do things in the real world is about the same.

I'm just a customer support rep, but they want us to use AI for my job. The AI cannot learn my job. There's so many tiny case by case things, it's difficult for it to understand everything people need from us. Worst yet, our system doesn't know how to even ask for more information, it just spits out confidently wrong information. Like the engineers commented here, I spend more time having to correct its answers and apologizing for thr wrong information than had I just been the one to answer the person to begin with.

They are doing it wrong