The AI Productivity Illusion | Hacker News Reader

The AI Productivity Illusion

hardresetmedia.com

This particular phrase captured my attention:

    > Productivity Is About (Much) More than Finishing Tasks Quickly

I read something similar in the book The Goal from Eliyahu Goldratt more than 20 year ago.

You can be highly productive and yet produce nothing of value.

yes!

as an aside: two books that made it almost impossible for me to deal with management "goals" at large corporations anymore:

1. the goal - by goldratt

2. out of the crisis - by deming

highly recommended!

> You can be highly productive and yet produce nothing of value.

I'm sorry, but this is nonsense.

Yes, you can produce nothing of 'traditional' value whilst still being productive in some way , but you 100% can not be 'highly productive' when you don't produce anything of value.

Sitting around and jacking off all day is not as productive as working on some unmonetizable project. The value can be in many aspects of that project.

In a mind boggling number of companies, “productive” measures output, not value.

No offense, but it sounds like you just skipped reading the blog and jumped right to the comment section.

The blog post is literally about various measures that seem "productive" but don't seem to be creating any value.

Economic productivity is an interesting perspective. I think you are right. Current AI spends too many resources but gets too few values (real values).

Too much hype in the process of AI development.

If analyzing the basic theory of LLM or AI (second hand data, data conversion, cognitive bias, etc.), this will make it more persuasive.

Over the last year I have seen an astonishing spike in poor-quality output within multiple employers. The problem with AI coding, ticket filing, etc is that you must review it afterward, but review would take nearly as long as just doing it yourself. The 10x productivity gain becomes 1.05x.

The adopted solution to this, to meet internal targets, is to accept reduced quality. Flood the ticketing system with bogus reports, each with 5 paragraphs of technobabble. In response, you get similarly faulty AI-driven triage tools to deal with the slop. But the triage tools have the same problem: they just don't work that well. Plenty of garbage gets through, and plenty of actual issues get filtered out.

The first movers get some eye-popping numbers to show management, but the aggregate productivity of the corporation is unchanged at best.

So is the productivity increase an illusion? Depends how you look at it. Is the productivity boost for the corporation an illusion? Yes. But, most ICs are measured by tickets closed, commits pushed, lines of code written, and tokens used. Like it or not, that is the definition of "productivity" that has been given to ICs. They're just doing what they're told.

Insightful point, but we should also consider we are in a very transitory period where were are adjusting every level of these tools (how verbose? How sycophantic? What should it ignore, which details matter, etc.) And more importantly, we as users are becoming attuned to reading and gaining understanding with AI.

When everyone first uses AI they are swayed by its intelligent sounding sentence construction. It took time to be able to look past that and understand basic assuptions still needed to be questioned.

> Insightful point, but we should also consider we are in a very transitory period where were are adjusting every level of these tools

Invalid excuse, in my mind. This isn't a matter of isolated mistakes. Mistakes are understandable, systemic failure is unacceptable.

> But, most ICs are measured by tickets closed, commits pushed, lines of code written, and tokens used.

Tokens used? That's a cost. One might as well include coffees drunk.

Tell that to management.

I think it actually ends up a negative net gain. The slop ends up taking more time and energy than it brings in value. Everyone is working harder, but the result is an even bigger pile of cards waiting for the wind to shift.

I wrote this back in 2023, and I think it holds

“After compiling all these, it still seems neutral to net negative. Maybe a net positive on the customer service side.“

https://garrickvanburen.com/media-tetrad-generative-ai/

My guess is AI mostly increases the throughput of low-quality output, and the cleanup burden lands on the people with better judgment

bike shedding with 28 agents lets you do nothing at scale

…inefficiently.

I can also do nothing at scale just with a hammock :)

I think this calculation is bad? I mean, they take sand and produce AI chips, and then the AI chips turn electricity into entertainment. So highly productive use of resources. Global entertainment industry is $3 trillion, easily justifies investment when combined with coding usage. So restructuring the economy etc. - those are moonshots that are like the cherry on top.

AI companies just need to capture more than half of the entire global entertainment market to avoid insolvency.

Seems unlikely, especially before the investment money runs out.

I can do a nice task with AI but can't easily schedule it.

Enterprise controls make what could be a simple cron job into a nearly impossible scheduled task.

So AI has to be one shot for me where I use it.

So productivity is dramatically lower if I have to manually track and execute tasks as some kind of human scheduler system.

Claude has a scheduler; have you tried it?

> the mere fact that AI allows people to work faster

It doesn't. It just lets them work less.

Those mean the same thing.

Strange then that we have separate words for speed and quantity.

If I can do something faster or I can do something with less effort, those are both the same productivity increases. It’s just the how the human involved wants to spend the increase.

coding wasn't ever the bottleneck or even a large contributor to outcomes. same for lots of this other stuff LLMs do.

so even if you 10x your coding, you aren't gonna have an appreciable effect on total output

If you work twice as fast but don't add a single $ to your income then you haven't increased productivity at all.

AI productivity only works if you were the only person in the world using it and no-one else was aware of it. You will look like a wizard, passing off slop to others who are too amazed and ignorant to know any better.

The fact that that is exactly what is happening (everyone passing off slop to everyone else and pretending either they didn't use AI or it isn't slop actually) is telling.

The mistake was giving it to everyone too early too fast. "If everyone is super, no one will be"

I prefer working alone, and even more so now with AI. When I work alone and own the whole project, I make the judgment call when and where slop is fine and where it's not. When working on a team there is always this battle to get code through the tests and review, often for code that is temporary or simply does not matter that much. Reviews rarely result in any meaningful changes anyway, unless the changes are < 50 LoC, in which case it's nitpicked to death instead of the "LGTM" you get for larger PRs. It all reminds me of airport security theater. Similarly, a 2 hour flight is now a 4 hour time waste of time, and a simple change that should take 10 minutes now takes up most of your day.

Reply on news.ycombinator.com