371–380 of 656 posts
When it was Copilot tab-completing lines, people would say, "yea, but you still have to make sure you're the one writing the whole functions". Then when it was completing functions, people would say, "yeah, but you still have to make sure you're the one writing the logic around the functions" Then when it was completing the logic around the functions, people would say, "yeah, but you still have to make sure you're th…
Earlier quoted context omitted.
Sounds like you're just fine depending on an extremely imprecise abstraction (natural language) and an extremely opaque third party (anthropic).
I can also just do it myself though lmao its not like I dont look at what it is producing The recommended tool cant even produce mobile friendly, like why would I ever use it?
Earlier quoted context omitted.
So, basically you need to micro-manage it. Where are your 10x gains now? And is it fun to work like that?
> you need to micro-manage it. It is significantly easier to micro-manage an AI than a suite of junior developers. The AI doesn't replace a principal engineer, it's replacing junior and weaker senior developers who need stories broken down extremely concisely to be able to get anything done. The time it takes to break down a story such that a junior through weak senior developers can pick it up and execute it well wo…
When it was Copilot tab-completing lines, people would say, "yea, but you still have to make sure you're the one writing the whole functions". Then when it was completing functions, people would say, "yeah, but you still have to make sure you're the one writing the logic around the functions" Then when it was completing the logic around the functions, people would say, "yeah, but you still have to make sure you're th…
The "people" in your hypothetical story have been wrong the whole time. The correct attitude is: When AI can complete lines, you still have to read and understand the code. When AI can complete whole functions, you still have to read and understand the code. When AI can complete features and tickets, you still have to read and understand the code.
Earlier quoted context omitted.
The generated code is more than fine, it’s good in many cases. And I read it :) Indeed for the task of “jump into an unfamiliar codebase and make a requested change that aligns with existing styles and patterns, and uses existing functionality” I would say something like opus 4.7 exceeds the capabilities of most developers.
I agree with both statements, but that doesn't change the problem I stated. If an agent produces reasonable code 80-90% of the time, and 10-20% of the time it makes mistakes that could render the codebase irretrievably unevolvable once they accumulate, the only thing you can do is to carefully review the agent's output 100% of the time. That it gets things right 80% of the time as opposed to 40% of the time doesn't c…
As soon as requirements change the abstractions fall apart and everything gets shoehorned.
Earlier quoted context omitted.
I agree with what you're saying, but I think we do have a problem right now with definitions where there's a lot of people basically getting supercharged tab completions or running a chatbot or two in a parallel pane, but still clearly reviewing everything; and on the other side of things is freaking Steve Yegge pitching a whole new editor that lets you orchestrate a dozen or more agents all vibing away on code you'r…
That blog post is surreal. It's like cryptocurrencies and the whole web3 nonsense. Cryptocurrencies basically don't work, so there have been a hundred aimless attempts at fixing self inflicted problems caused by deficiencies of cryptocurrencies with no actual goal that has any impact on the real world. It's the same thing here. AI has dropped the cost of software development, so developers are now fooling themselves…
Earlier quoted context omitted.
And the solution is the same, as when it was outsourced- and the "patch" was fix it by writing spec. Thus i conclude my TED talk with the statement: LLMs are the new outsourcing and run into the same problems.
Not quite, because the architecture often needs to evolve when you learn more as the project evolves. People will complain when they feel the constraints drive them to unnatural workarounds, the agents don't. You can try telling the agent to stop and ask when a constraint proves problematic, except it doesn't have as good a judgment as humans to know when that's the case. I often find myself saying, "why did you writ…
Here’s what’s working for me right now:
1. The basics: use best model available, have skills and rules that specify project guidelines, etc.
2. Always use plan mode. It works much better to iterate on the concept of what we’re going to do, then do the implementation. The models will adhere to the plan at very high rates in my experience.
3. Don’t give chunks of work that are too large in scope. This is just art, and I’m constantly experimenting with how ambitious I can be.
4. I review all code to some extent, but I have a strong mental model of what areas of the app are more critical, where hidden bugs might accumulate, etc, and I review both tests and impl more strenuously in those areas. Whereas like a widget for my admin panel probably gets a 2 second glance.
5. Have the discipline to go through periodically and clean up tech debt, refactor things that you’d do differently now, etc. I find the AI a huge help here, because I can clean up cruft in an hour that would have once taken me days, and thus probably wouldn’t have gotten done.
6. I’m experimenting with shifting my architecture to make it easier to review AI code, make it less likely it’ll make mistakes, etc. Honestly mostly things I should have always been doing, but the level of formalism and abstraction on my solo projects is usually different than on a bigger team.
To each their own, but I’ve grown this from nothing to about $350k in ARR over the last ten months, and I’m very confident I never could have built this product without AI help in triple that time.
Earlier quoted context omitted.
Yeah, their statement just isn't true. With enough instruction, I've been able to get great output from models. I think that's the key: with detailed, pointed instructions, the output will match.
how do you know it matches? You did read it then?
After reading a bunch of other comments, it sounds like people are referring to letting agents go wild and code whatever off a limited prompt. I'm not using LLMs like that; I'm generally interacting only via conversations with pretty detailed initial prompts. My interactions with the chat after that are corrections/guiding prompts to keep it on point and edit the prompt output from time to time.
Earlier quoted context omitted.
The "people" in your hypothetical story have been wrong the whole time. The correct attitude is: When AI can complete lines, you still have to read and understand the code. When AI can complete whole functions, you still have to read and understand the code. When AI can complete features and tickets, you still have to read and understand the code.
I heard a talk from a VP at NVIDIA a couple of months ago and he echoed this. Essentially their policy is "you are still fully responsible for the code you ship, whether AI helps with it or not"
But that still doesn’t mean I review all the code. I tend to review defensively, based on the potential for harm if this piece of code is broken. And I rely a lot on tests, static analysis, canaries, analytics, health checks, etc. to reduce risk for when I’m wrong. So far it’s working.