[Benjammer]

Are you paying for the higher end models? Do you have proper system prompts and guidance in place for proper prompt engineering? Have you started to practice any auxiliary forms of context engineering?

This isn't a magic code genie, it's a very complicated and very powerful new tool that you need to practice using over time in order to get good results from.

[tempodox]

That's the beauty of the hype: Anyone who cannot replicate it, is “holding it wrong”.

[orangecat]

Or maybe it works well in some cases and not others?

[goalieca]

It ain’t a magic code genie. And developers don’t spend most of their day typing lines of code. Lots of it is designing, figuring out what to build, understanding the code, maintenance considerations, and adhering to the style of whatever file you’re in. All these agents needing local context and still spit junk.

[QuantumGood]

   > it's a very complicated and very powerful new tool that you need to practice using over time in order to get good results from.

Of course this is and would be expected to be true. Yet adoption of this mindset has been orders of magnitude slower than the increase in AI features and capabilities.

[dingnuts]

guy 1: I put money in the slot machine everyone says wins all the time and I lose

you: HAVE YOU PUT MORE TOKENS IN???? ARE YOU PUTTING THEM IN THE EXPENSIVE MACHINES???

super compelling argument /s

if you want to provide working examples of "prompt engineering" or "context engineering" please do but "just keep paying until the behavior is impressive" isn't winning me as a customer

it's like putting out a demo program that absolutely sucks and promising that if I pay, it'll get good. why put out the shit demo and give me this impression, then, if it sucks?

[lordnacho]

The way I ended up paying for Claude max was that I started on the cheap plan, it went well, then it wanted more money, and I paid because things were going well.

Then it ran out of money again, and I gave it even more money.

I'm in the low 4 figures a year now, and it's worth it. For a day's pay each year, I've got a junior dev who is super fast, makes good suggestions, and makes working code.

[Avicebron]

> For a day's pay each year

For anyone trying to back of the napkin at $1000 as 4-figures per year, averaged as a day salary, the baseline salary where this makes sense is about ~$260,000/yr? Is that about right lordnacho?

[lordnacho]

Yeah I thought that was a reasonable number in the ballpark. I mean, it probably makes sense to pay a lot more for it. A grand is probably well within the range where you shouldn't care about it, even if you only get a basic salary and it's a terrible year with no bonus.

And that's not saying AI tools are the real deal, either. It can be a lot less than a fully self driving dev and still be worth a significant fraction of an entry level dev.

[antihipocrat]

I assume it's after tax too..