[alpha_squared]

> - Significantly increased my productivity as a software engineer.

This is exactly the point that keeps coming up that folks are struggling to grasp, myself included. How are you measuring this? It certainly makes me feel productive, but I'm not sure I can confidently say it has actually made me more productive. It's made the easy stuff a no-brainer (e.g. boilerplate, simple logic) and the moderate stuff really hard. Never mind the hard stuff. Vetting the code has become a whole other job on its own. The only folks I've found who confidently claim it increases productivity appear to be online (and without evidence), because no one in person is willing to claim that and show it.

[capnjngl]

I'll share my experience.

I've never been a developer. Dabbled in frontend web for a bit (HTML/CSS/JS, no large frameworks) and felt like if I really dedicated some time to learning how to code, I'd be pretty decent at it. It's always intrigued me, and I've always had an itch to build things, but just never found the time. I'm in marketing now - I own an agency.

Over the last 6 months since the coding models really began to step up and get good, I've built several dedicated apps to support my business:

-Profitability optimizer and forecaster based on unit economics and current ad efficiency.

-Creative strategy tool that ingests brand and product data and helps explore primary and secondary personas and emotional motivators.

-Reporting tool that processes natural language queries and connects to multiple data sources to fetch results. Can schedule reports to post directly to Slack or email.

All robust and hosted on Railway. Team members can use them. Clients can use them. OAuth via Google.

Would any of this have been possible for me before the rise of frontier LLMs? Absolutely not. Learning the frameworks alone would have taken me longer than it's taken to just... build. Rapidly build and deploy. Total game changer for me.

Oh - and I'm building a game on the side. LLMs know Godot.

[dsharlet]

I can agree with the skeptics that LLM generated code is usually crap. I rarely accept its output without significant edits unless it's truly boilerplate, and I want to avoid the need for that kind of code in the first place.

For me, the killer use case is debugging. I hate wasting time debugging something that should work except for mistakes, and now I do that probably 75% less than I used to because AI does it for me.

I don't know if it makes me that much more productive, but I certainly enjoy my work more not having to do as much tedious debugging, and it feels like I waste a lot less time doing it.

[vunderba]

I have an example I share a lot.

I have about a dozen programs that are now close to ten years old, all running quietly in the background on my computer doing various tasks. I originally wrote all of them as JavaScript/Node programs. That was fine at the time, but the memory usage really adds up, since each one needs its own V8 runtime (anywhere from 50 to 100 MB per process).

I have basically zero familiarity with Golang, but I was able to use an agentic harness to run in the background and convert every single one of those scripts into Go equivalents. Because I've accumulated lots of before-and-after data, it was easy to confirm that the new versions matched the behavior and accuracy of the originals.

End Result: memory usage dropped from tens of megabytes per program to just a couple of megabytes each.

[peab]

I've been able to build things that I otherwise would not have been able to build, in the free time that I have:

- a VST audio plugin

- a wedding website with RSVP functionality

- a relaxing game for my wife

At work, I've been able to build much more than I would have been capable of in the past. I'm a backend eng, and it allows me to build much much nicer frontends than I've ever been able to do in the past.

And before you tell me that the code is crap - it doesn't matter! It may or may not be good code, but it works and serves it's purpose very well. Anyways, I'm I'm not launching a rocket, or putting software into cars.

[olalonde]

> How are you measuring this?

I attempt a programming task with and without LLM assistance. The attempt with LLM assistance is pretty much always completed faster and cleaner.

Another example: https://news.ycombinator.com/item?id=43991777

[evdubs]

How much faster? How much cleaner? What tasks are you accomplishing?

[olalonde]

I linked to an example in the comment. In that particular case, I'd say probably 10-20x faster. I do embedded, backend, web and mobile app development.

[nevertoolate]

I also notice these things. Otoh i spend definitely less than 50% of my time typing in code so it is impossible that it gives more than 2x speedup. And sometimes i lose time babysitting and rewriting stuff so all in all it is kinda no productivity gain.

[baobabKoodaa]

Can someone bring this man a cup of Kool-Aid, stat?

[goosejuice]

> It's made the easy stuff a no-brainer (e.g. boilerplate, simple logic) and the moderate stuff really hard. Never mind the hard stuff. Vetting the code has become a whole other job on its own.

Not everyone has the same requirements, skills, usage patterns, and outcomes. It's that simple.