Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI productivity gains are closer to 10% than 10x

leaddev.com

Post reply on HN
Print view View original

31–38 of 38 posts

Re: AI productivity gains are closer to 10% than 10x

#31

Post by mytailorisrich »

goatlover wrote: post #4
mytailorisrich wrote: post #2

AI has the potential to be extremely powerful but it is also very new. People and organisations are just at the beginning of the learning process.

Sure but the hype has been about the 10x productivity increases and how less employees are needed. Also all the AGI/singularity stuff being just around the corner, not simply AI potentially making employees a lot more productive.

> Sure but the hype has been about the 10x productivity increases and how less employees are needed

Which will probably be true at some point when we have become good at using AI, and maybe when AI has improved further as well.

Re: AI productivity gains are closer to 10% than 10x

#32

Post by watwut »

satvikpendem wrote: post #24
watwut wrote: post #19

Earlier quoted context omitted.

You dont need to focus if you dont care about quality. Overworked teams do it all the time.

You need to make sure the text you're typing actually compiles into a working program, so there is some degree of focus required. Compare that with vibe coding, just type a sentence of what you want and it does everything itself with no continued focus from you. The degree is completely different.

I did worked in teams that produced crap. They were initially all able to proceed much faster then teams that did not produced crap. They felt like being super productive the whole time, cranking out issues one after another. The initial speed slowed down after 2-3 months, but they did not noticed it. They ended up being actually slower, customers were unhappy because they were not getting functional software they wanted and the team believed themselves to be victims.

The overall results, if you joined them later was "wow this is slow, you are producing crap and customer is entirely correct when they complain".

My point here is that "don't care about quality" does a lot in that claim. It leads to a lot of useless and contra-productive work - you would be better off if they have done nothing. And I think what people are reproducing that state here, just faster.

Re: AI productivity gains are closer to 10% than 10x

#33

Post by joshstrange »

discreteevent wrote: post #11
mycentstoo wrote: post #8

My unsolicited thoughts: * AI will drive additional features as development accelerates * There will be a window where fast adopting AI companies can deliver features faster than peers * Those fast adopting AI companies will be able to garner additional revenue * Gap closes as AI benefits slow in their impact to feature development and slower adopting companies catch up * Companies that were faster adopters can no lo…

> There will be a window where fast adopting AI companies can deliver features faster than peers >Those fast adopting AI companies will be able to garner additional revenue The companies going 50% slower on reliable features that customers want to pay for will continue to wipe the floor with the hyper productive companies producing fluff.

I reject the premise that you can’t use LLM’s to build reliable features.

Re: AI productivity gains are closer to 10% than 10x

#34

Post by jijijijij »

horticulturist wrote: post #13

The variation in individual developer throughput was as much as 20x pre-AI. 10% is basically nothing. You’d be better off firing your worst performer and replacing them with someone good (or great).

Good luck getting those devs. Between the fallout of covid and AI in education, there is likely half an upcoming generation poorly educated and comparatively less qualified.

Right now the market favors employers, but I think when the sobriety sets in this will rapidly change. Senior devs are gonna get more expensive. Companies should massively invest in education programs.

Re: AI productivity gains are closer to 10% than 10x

#35

Post by satvikpendem »

watwut wrote: post #32
satvikpendem wrote: post #24

Earlier quoted context omitted.

You need to make sure the text you're typing actually compiles into a working program, so there is some degree of focus required. Compare that with vibe coding, just type a sentence of what you want and it does everything itself with no continued focus from you. The degree is completely different.

I did worked in teams that produced crap. They were initially all able to proceed much faster then teams that did not produced crap. They felt like being super productive the whole time, cranking out issues one after another. The initial speed slowed down after 2-3 months, but they did not noticed it. They ended up being actually slower, customers were unhappy because they were not getting functional software they wa…

Great but none of that was my claim at all, I just said that AI vibe coding takes less physical effort, meaning literal effort to press keys, than hand coding. Either way, AI models these days are much better than those low tier coders as well as average coders, and the best models are even better than most top tier coders, all at a much higher speed, so you can't compare bad human coders and AI anymore.

Re: AI productivity gains are closer to 10% than 10x

#36

Post by andrekandre »

matrix87 wrote: post #27
mycentstoo wrote: post #8

My unsolicited thoughts: * AI will drive additional features as development accelerates * There will be a window where fast adopting AI companies can deliver features faster than peers * Those fast adopting AI companies will be able to garner additional revenue * Gap closes as AI benefits slow in their impact to feature development and slower adopting companies catch up * Companies that were faster adopters can no lo…

> There will be a window where fast adopting AI companies can deliver features faster than peers The fundamental problem is having people doing the right things. Lines of code does not equal real economic value. I think AI will have a much different effect: raw headcount will no longer give large companies the same moat that they once had. If AI makes a 10x engineer into an 100x engineer and unlocks this whole world…

  > If AI makes a 10x engineer into an 100x engineer and unlocks this whole world of productivity, getting them to actually deliver 100x requires the proper culture where they aren't blocked by environmental factors, management, and incentive structure.
yea, from what i've seen, even "ai-first" companies have too many barriers in place to allow anyone to reach that hypothetical 100x... to many reviews, daily meetings, incidents, goals set by management that you cant ignore, etc etc

just a guess but i suppose that hypothetical 100x (if its even possible) is only going to be reachable by people in better suited orgs (my cynical side says very few) or just indie devs...

Re: AI productivity gains are closer to 10% than 10x

#37

Post by budsniffer952 »

joshstrange wrote: post #33
discreteevent wrote: post #11

Earlier quoted context omitted.

> There will be a window where fast adopting AI companies can deliver features faster than peers >Those fast adopting AI companies will be able to garner additional revenue The companies going 50% slower on reliable features that customers want to pay for will continue to wipe the floor with the hyper productive companies producing fluff.

I reject the premise that you can’t use LLM’s to build reliable features.

And you get down voted for saying something truthful and mundane because, apparently, HackerNews is now filled with morons.

Re: AI productivity gains are closer to 10% than 10x

#38

Post by mstaoru »

My rather simplistic take:

Everybody have access to the same LLMs. Eventually if everybody is a "Nx engineer" (see PS below), then nobody is. Just a new baseline.

Once you are at this point, you can either invest further in LLMs, marginally increasing "productivity", or invest in people and culture, increasing potential for innovation by bringing in deep domain expertise that is not in any LLM and never will be.

PS: many measure "Nx engineer" by lines of code or number of "features" someone churns out; a better metric would be real world impact. Can I 10x my "made the world a better place" metric with LLMs? Hard to say. Probably not.

Post reply on HN