This x 10 . I don’t understand how people are saying you can’t use LLMs to get crazy productivity gains. If you can’t write quality code with LLMs at ludicrous speed, you’re holding it wrong. You will have occasional bad days and regressions. But overall you’re still going to be able to 4x your progress.
That’s kinda what I mean. Maybe it only works well in some languages, but with the harness I built for C and C++ does a fantastic job of adhering to very strict architecture and style guides. Way cleaner, more readable, better factored, and more interpretable than human generated code, except maybe one or two devs I have worked with. YMMV I guess?
TBF I do burn 200k tokens just preloading the context with onboarding, not including any code, just document trees of development policy documents, style and architectural standards, code and documentation review processes, company ethos and culture, etc. it’s a token fire, but it really works for us.
Also, documentation driven development all the way down.
If you're an enterprise (including startups), you worry about customers, not code quality. There are famously many startups that gained traction despite shit code and then eventually got around to fixing it, to whatever extent was possible, like Facebook HHVM, Stripe's Sorbet, etc.
Ok, and? You can live with that if there are more important things to deal with.
I've stared at ugly LLM code, that I had just had generated, and worked well enough for my purposes. (generally, some quick recursion into a nested python dictionary in order to dig out some property -- especially for linting or quick data analysis).
And I wanted something better, sure, something a bit more readable ...but I just needed it to work well enough to recurse through a yaml file for config file linting, not be battle-hardened against every test case.
So to deal with the mess, I shoved it in a pure function, threw a few basic sanity unit tests around it, put a comment with a disclaimer of "#this is LLM generated code, it is lightly tested, do not use it for anything truly load-bearing without a lot more tests" and I moved on to something else.
I measured an ~8x increase in my project's commit count after AI, and I'm painstakingly reading, reviewing, understanding and editing everything the models write. It's gotten to the point I'm trying to slow down in order to let the new knowledge crystallize. I'm manually writing articles about what I'm doing as I go.
I can only imagine what people are doing at their jobs with unlimited token budgets.
I wonder if the people getting 10x productivity gains are spending less time on HN and more time tending to their agents. Personally I now spend so much time productively arguing with agents that it feels like an utter waste of effort arguing with humans, if people can't see the value in LLMs by now I'm not sure what I could say to change their minds.
Definitely enjoying the lack of eye-rolling, being asked to explain obvious things multiple times, and stopping things being done for resume-stuffing reasons.
There's a small minority of people who are adamantly refusing to change, such as there are in every technological revolution. Ego prevents them from even wholeheartedly trying the tool, because it would be admission they were wrong.
The opportunities available for these people are rapidly, rapidly shrinking. I believe it's possible to be a developer today who's EXCEPTIONAL and never uses AI. Most opponents are not exceptional, though, and even these opportunities are shrinking.
Most exceptional developers in my org adopted AI in their workflows and went from 10x developers to 20x developers.
If you refuse to adapt, you're going to be out of a job complaining about the kids and their newfangled technology REAL quick. You have a few years remaining, maybe less.
I can’t turn 10x work into 20x work because I have to ensure the two juniors in my team who are now creating 50x work won’t merge complete garbage, reviewed by another engineer that has already given up on caring.
I can’t turn 10x work into 20x work because my Product Manager thinks changing fundamental premises of tasks I already spent two weeks on (mostly removing human blockers) is very simple. After all, when he asked Claude to update his prototype, it only took it 10 minutes.
I can’t turn 10x work into 20x work because the company dedicated entire teams to write company-wide skills for everything. They suck, but if I don’t use them, I’m not following the new “golden path for engineering”, and I lose points in my performance review.
I can, however, turn 10x work into 20x work, or even much more than that, if AI actually did what it’s promising and eliminated most of my team, the product manager, and the middle managers. Or me. I could use a break.
What about the 6x developers? Was there just a doubling multiplier across the board, resulting in them becoming 12x developers, or did they too become 20x developers?
>> What concrete business advantage are you getting from LLMs?
> Speed.
Speed of what?
Speed of understanding what needs to be done? I highly doubt it.
Speed of LoC checked into git? Sure, I'll give you that.
But one can use any number of tools to generate hundreds of thousands of lines of code. See any build tools which support specifications such as RAML, OpenAPI, CORBA, etc.
fixing minor bugs takes one slack message for us now. bugs go down, goodness go up.
fixing more serious regression also easier. connect honeycomb mcp, ask agent to debug while i walk to coffee and get some pistachio rose dates. by time im back with my oat latte ive got a full report on what happened and can send the next slack message to fix.
I needed to deeply understand a code base I had no experience with in a language I don't normally use with what I would describe as haphazard documentation at best. You can't argue with the speed at which I gained the required understanding of the project.
In the time it took you to type that, your hourly market comp went down another basis point.
I am appalled none of this is clicking with you anti-AI folks. This is all so exciting -- alarming even! --, and software careers are never going to be the same.
I don't know how you just metaphorically stand there and act like nothing at all is happening. We've never seen anything like this in our entire lives.
Some of you are standing right in front of the steam roller, yelling to all of us that steam rollers aren't real.
Did you use an LLM to write this for you? How odd.
For all of you people who think these LLM models are “earth shattering” how the hell do you reconcile that it’s a net positive for anyone but those who want to consolidate knowledge and power.
From my brief window of Fable usage, speed wasn't its strong point at all.
For actually building software, I'm starting to suspect a human with a dumber (but faster) model is going to get the job done quicker than Fable (and possibly even cheaper). Bug-finding and vulnerability detection is a different story.
My conclusion was the exact opposite. Maybe each individual response was slower, but it took so many fewer round trips to get what I wanted wanted. I had a project fable was progressing steadily and correctly on. Opus on the same project keeps handing me garbage it insists is working and meets the stated requirements, but isn’t and doesn’t.
This x 10 . I don’t understand how people are saying you can’t use LLMs to get crazy productivity gains. If you can’t write quality code with LLMs at ludicrous speed, you’re holding it wrong. You will have occasional bad days and regressions. But overall you’re still going to be able to 4x your progress.
I have plenty of experience with LLMs and use them daily but definitely wouldn't call generated code "quality code." Often looks like complete vomit.
That’s kinda what I mean. Maybe it only works well in some languages, but with the harness I built for C and C++ does a fantastic job of adhering to very strict architecture and style guides. Way cleaner, more readable, better factored, and more interpretable than human generated code, except maybe one or two devs I have worked with. YMMV I guess?
TBF I do burn 200k tokens just preloading the context with onboarding, not including any code, just document trees of development policy documents, style and architectural standards, code and documentation review processes, company ethos and culture, etc. it’s a token fire, but it really works for us.
Also, documentation driven development all the way down.
If you're an enterprise (including startups), you worry about customers, not code quality. There are famously many startups that gained traction despite shit code and then eventually got around to fixing it, to whatever extent was possible, like Facebook HHVM, Stripe's Sorbet, etc.
3 replies →
Ok, and? You can live with that if there are more important things to deal with.
I've stared at ugly LLM code, that I had just had generated, and worked well enough for my purposes. (generally, some quick recursion into a nested python dictionary in order to dig out some property -- especially for linting or quick data analysis).
And I wanted something better, sure, something a bit more readable ...but I just needed it to work well enough to recurse through a yaml file for config file linting, not be battle-hardened against every test case.
So to deal with the mess, I shoved it in a pure function, threw a few basic sanity unit tests around it, put a comment with a disclaimer of "#this is LLM generated code, it is lightly tested, do not use it for anything truly load-bearing without a lot more tests" and I moved on to something else.
Not everything has to be bulletproof.
13 replies →
I measured an ~8x increase in my project's commit count after AI, and I'm painstakingly reading, reviewing, understanding and editing everything the models write. It's gotten to the point I'm trying to slow down in order to let the new knowledge crystallize. I'm manually writing articles about what I'm doing as I go.
I can only imagine what people are doing at their jobs with unlimited token budgets.
Kind of weird how LoC has become a metric for people to chase again.
You don’t have to imagine, listen to Boris’ publicly saying how he works with these things and it’s safe to assume others do it similarly or better
I wonder if the people getting 10x productivity gains are spending less time on HN and more time tending to their agents. Personally I now spend so much time productively arguing with agents that it feels like an utter waste of effort arguing with humans, if people can't see the value in LLMs by now I'm not sure what I could say to change their minds.
We must then assume you're not getting those 10x gains
1 reply →
Definitely enjoying the lack of eye-rolling, being asked to explain obvious things multiple times, and stopping things being done for resume-stuffing reasons.
There's a small minority of people who are adamantly refusing to change, such as there are in every technological revolution. Ego prevents them from even wholeheartedly trying the tool, because it would be admission they were wrong.
The opportunities available for these people are rapidly, rapidly shrinking. I believe it's possible to be a developer today who's EXCEPTIONAL and never uses AI. Most opponents are not exceptional, though, and even these opportunities are shrinking.
Most exceptional developers in my org adopted AI in their workflows and went from 10x developers to 20x developers.
If you refuse to adapt, you're going to be out of a job complaining about the kids and their newfangled technology REAL quick. You have a few years remaining, maybe less.
I can’t turn 10x work into 20x work because I have to ensure the two juniors in my team who are now creating 50x work won’t merge complete garbage, reviewed by another engineer that has already given up on caring.
I can’t turn 10x work into 20x work because my Product Manager thinks changing fundamental premises of tasks I already spent two weeks on (mostly removing human blockers) is very simple. After all, when he asked Claude to update his prototype, it only took it 10 minutes.
I can’t turn 10x work into 20x work because the company dedicated entire teams to write company-wide skills for everything. They suck, but if I don’t use them, I’m not following the new “golden path for engineering”, and I lose points in my performance review.
I can, however, turn 10x work into 20x work, or even much more than that, if AI actually did what it’s promising and eliminated most of my team, the product manager, and the middle managers. Or me. I could use a break.
2 replies →
What about the 6x developers? Was there just a doubling multiplier across the board, resulting in them becoming 12x developers, or did they too become 20x developers?
>> What concrete business advantage are you getting from LLMs?
> Speed.
Speed of what?
Speed of understanding what needs to be done? I highly doubt it.
Speed of LoC checked into git? Sure, I'll give you that.
But one can use any number of tools to generate hundreds of thousands of lines of code. See any build tools which support specifications such as RAML, OpenAPI, CORBA, etc.
So I ask again; speed of what?
fixing minor bugs takes one slack message for us now. bugs go down, goodness go up.
fixing more serious regression also easier. connect honeycomb mcp, ask agent to debug while i walk to coffee and get some pistachio rose dates. by time im back with my oat latte ive got a full report on what happened and can send the next slack message to fix.
life is good
I needed to deeply understand a code base I had no experience with in a language I don't normally use with what I would describe as haphazard documentation at best. You can't argue with the speed at which I gained the required understanding of the project.
In the time it took you to type that, your hourly market comp went down another basis point.
I am appalled none of this is clicking with you anti-AI folks. This is all so exciting -- alarming even! --, and software careers are never going to be the same.
I don't know how you just metaphorically stand there and act like nothing at all is happening. We've never seen anything like this in our entire lives.
Some of you are standing right in front of the steam roller, yelling to all of us that steam rollers aren't real.
Did you use an LLM to write this for you? How odd.
For all of you people who think these LLM models are “earth shattering” how the hell do you reconcile that it’s a net positive for anyone but those who want to consolidate knowledge and power.
We are really looking at idiocracy in the making.
Very very fast steam rollers.
Nice strawman[0], but you avoided answering my core question:
With ad hominems and a non sequitur. How about I narrow the question with the hope it engenders a relevant response:
0 - https://en.wikipedia.org/wiki/Straw_man
5 replies →
From my brief window of Fable usage, speed wasn't its strong point at all.
For actually building software, I'm starting to suspect a human with a dumber (but faster) model is going to get the job done quicker than Fable (and possibly even cheaper). Bug-finding and vulnerability detection is a different story.
My conclusion was the exact opposite. Maybe each individual response was slower, but it took so many fewer round trips to get what I wanted wanted. I had a project fable was progressing steadily and correctly on. Opus on the same project keeps handing me garbage it insists is working and meets the stated requirements, but isn’t and doesn’t.
I’d say you tried on an insufficiently complex codebase. I’ve tried on a MLOC+ and the results were excellent compared to anything else.
Not saying the results were bad - quite the opposite. But it was very slow (and if I was paying API rates, hideously expensive).
Drawing debt
We'll just rebuild stuff when we get new requirements. The models will be even faster and better for the next version, anyway.