4 ms·

I understand the pitch here ("it finds bugs! it's basically all upside because worst case there's no output anyways"), but I'm finding some of these agents to b

• Wowfunhappy 11mo ago
  Sometimes you hit a wall where something is simply outside of the LLM's ability to handle, and it's best to give up and do it yourself. Knowing when to give up may be the hardest part of coding with LLMs. Notably, these walls are never where I expect them to be—despite my best efforts, I can't find any sort of pattern. LLMs can find really tricky bugs and get completely stuck on relatively simple ones.
  • ori_b 11mo ago
    Doing it yourself is how you build and maintain the muscles to do it yourself. If you only do it yourself when the LLM fails, how will you maintain those muscles?
    • Klathmon 11mo ago
      If the LLM is able to handle it why do you need to maintain those specific skills?
      • ribosometronome 11mo ago
        Should we not teach kids math because calculators can handle it? Practically, though, how would someone become good at just the skills LLMs don't do well? Much of this discussion is about how that's difficult to predict, but even if you were a reliable judge of what sort of coding tasks LLMs would fail at, I'm not sure it's possible to only be good at that without being competent at it all.
        • Wowfunhappy 11mo ago
          > I'm not sure it's possible to only be good at that without being competent at it all. This is, in fact, why we teach kids math that calculators could handle!
        • jjmarr 11mo ago
          > Should we not teach kids math because calculators can handle it? We don't teach kids how to use an abacus or a slide rule. But we teach positional representations and logarithms. The goal is theoretical concepts so you can learn the required skills if necessary. The same will occur with code. You don't need to memorize the syntax to write a for loop or for each loop, but you should understand when you might use either and be able to look up how to write one in a given language.
          • ori_b 11mo ago
            Huh. I was taught how to use both an abacus and a slide rule as a kid, in the 90s.
        • Klathmon 11mo ago
          Should you never use a calculator because you want to keep your math skills high? There are a growing set of problems which feel like using a calculator for basic math to me. But also school is a whole other thing which I'm much more worried about with LLMs. Because there's no doubt in my mind I would have abused AI every chance I got if it were around when I was a kid, and I wouldn't have learned a damn thing.
    • Wowfunhappy 11mo ago
      I agree, and I can actively feel myself slipping (and perhaps more critically, not learning new skills I would otherwise have been forced to learn). It's a big problem, but somewhat orthogonal to "what is the quickest way to solve the task currently in front of me."
      • ori_b 11mo ago
        Which needs to be balanced with "How do I maintain my ability to keep solving tasks quickly?"
      • kryogen1c 11mo ago
        > but somewhat orthogonal to "what is the quickest way to solve the task currently in front of me." That depends on if you ignore the future. You are never just solving the problem in front of you; you should always act in a way that propagates positivity forward in time.
        • dcow 11mo ago
          Some jobs require investment in the future. Some do not. That’s just reality. Not white how I feel about it personally, but I think there is a fair amount of the developer trade that is operational.
      • AbstractH24 11mo ago
        The thing i struggle with is I feel like it’s hard to lock into which skill to learn properly. Which so much changing so quickly and it becoming easy to learn things superficially.
    • RA_Fisher 11mo ago
      By moving up a level in the abstraction layer similar to moving from Assembly to C++ to Python (to LLM). There’s speed in delegation (and checking as beneficial).
      • ThrowawayR2 11mo ago
        Moving up abstraction layers really only succeeds with a solid working knowledge of the lower layers. Otherwise, you're just flying blind, operating on faith. A common source of bugs is precisely a result of developers failing to understand the limits of the abstractions they are using.
        • AbstractH24 11mo ago
          So we can all only succeed if we know how CPUs handle individual instructions?
          • Wowfunhappy 11mo ago
            I'm not sure whether I agree with GP, but I think you may be misinterpreting their point. I can have an understanding of CPUs in general without knowing individual instructions, and I do think knowing about things like CPU cache is useful even when writing e.g. Python.
            • RA_Fisher 11mo ago
              Yes, for sure! And being able to orchestrate AI to use that knowledge provides leverage for fulfilling tasks. Eventually, yes, I think we'll delegate to AI in more and more complete ways, but it's a process that takes some time.
            • AbstractH24 11mo ago
              I see what you’re getting at and it makes sense. Goes to the larger idea that strategic and logic is important for scalability and long term success. Not just execution. Something LLMs miss often (mostly because people fail to communicate it to them).
  • rtpg 11mo ago
    Sure, I agree with the "levels of automation" thought process. But I'm basically experiencing this from the start. If at the first step I'm already dealing with a robot in the weeds, I will have to spend time getting it out of the weeds, all for uncertain results afterwards. Now sometimes things are hard and tricky, and you might still save time... but just on an emotional level, it's unsatisfying
• adastra22 11mo ago
  So you feed the output into another LLM call to re-evaluate and assess, until the number of actual reports is small enough to be manageable. Will this result in false negatives? Almost certainly. But what does come out the end of it has a higher prior for being relevant, and you just review what you can. Again, worst case all you wasted was your time, and now you've bounded that.
• joshvm 11mo ago
  When models start to forage around in the weeds, it's a good idea to restart the session and add more information to the prompt for what it should ignore or assume. For example in ML projects, Claude gets very worried that datasets aren't available or are perhaps responsible. Usually if you tell it outright where you suspect the bug to be (or straight up tell it, even if you're unsure) it will focus on that. Or, make it give you a list of concerns and ask you which are valid. I've found that having local clones of large library repos (or telling it to look in the environment for packages) is far more effective than relying on built-in knowledge or lousy web search. It can also use ast-grep on those. For some reason the agent frameworks are still terrible about looking up references in a sane way (where in an IDE you would simply go to declaration).
  • shadyKeystrokes 11mo ago
    [dead]
  • theshrike79 11mo ago
    Context7 MCP is the one I keep enabled for all sessions. Then there are MCPs that give LSP access to the models as well as tools like Crush[0] that have LSPs built in. [0] https://github.com/charmbracelet/crush https://github.com/charmbracelet/crush
  • embedding-shape 11mo ago
    Yeah, I do the same too, cloning reference repos into known paths, tell it to look there if unsure. Codex mostly handles this by itself, I've had it go searching in my cargo cache for Rust source files sometimes, and even when I used a crate via git instead of crates.io, it went ahead and cloned the repo to /tmp to inspect it properly. Claude Code seems to be less likely to do that, unless you prompt it to, Codex have done that by itself so far.
• MattGaiser 11mo ago
  Just ask it to prioritize the top ones for your review. Yes, they can bikeshed, but because they don’t have egos, they don’t stick to it. Alternatively, if it is in an area with good test coverage, let it go fix the minor stuff.
  • rtpg 11mo ago
    I don't like their fixes, so now I'm dealing with imperfect fixes to problems I don't care about. Tedium
• SV_BubbleTime 11mo ago
  Ok, fair critique. EXCEPT… What did you have for AI three years ago? Jack fucking shit is what. Why is “wow that’s cool, I wonder what it’ll turn into” a forbidden phrase, but “there are clearly no experts on this topic but let me take a crack at it!!” important for everyone to comment on? One word: Standby. Maybe that’s two words.
  • advael 11mo ago
    With all due respect, "wow this is cool, I wonder what it'll turn into" is basically the mandatory baseline stance to take. I'm lucky that's where I'm still basically at, because anyone in a technical position who shows even mild reticence beyond that is likely to be unable to hold a job in the face of their bosses' frothing enthusiastic optimism about these technologies
    • dns_snek 11mo ago
      Is it that bad out there? Yeah, I don't think I could last in a job that tries to force these tools into my workflow.
      • bravetraveler 11mo ago
        Drive-by comment: it's not so bad, here. I work with a few peers who've proven to be evangelists with much stronger social skills. When the proposition comes up, I ask how my ass should be cleaned, too. Thankfully: the bosses haven't heard/don't care. Varying degrees of 'force' at play; I'm lucky that nobody significant is minding my [absence of] LLM usage. Just some peers excited to do more for the same or, arguably, less reward. Remember: we're now all in an arms race. Some of us had a head start. How crass I respond to the suggestion depends on their delivery/relevance in my process/product, of course. May be placated like a child with a new toy... or the gross question to, hopefully, express the suggestion isn't wanted, needed, or welcome. Faced with a real mandate, I'd feed it garbage while looking for new work. Willing to make the bet I can beat enough machines while people are still involved at all.
      • advael 11mo ago
• j2kun 11mo ago
  > except at least the junior is a person. +1 Juniors can learn over time.
• solumunus 11mo ago
  Communication with a person is more difficult and the feedback loop is much, much longer. I can almost instantly tell whether Claude has understood the mission or digested context correctly.
• embedding-shape 11mo ago
  > I understand the pitch here ("it finds bugs! it's basically all upside because worst case there's no output anyways"), but I'm finding some of these agents to be ... uhhh... kind of agressive at trying to find the solution and end up missing the forest for the trees. And there's some "oh you should fix this" stuff which, while sometimes isn't _wrong_, is completely besides the point. How long/big do your system/developer/user prompts end up being typically? The times people seem to be getting "less than ideal" responses from LLMs tend to be when they're not spending enough time setting up a general prompt they can reuse, describing exactly what they want and do not want. So in your case, you need to steer it to do less outside of what you've told it. Adding things like "Don't do anything outside of what I've just told you" or "Focus only on the things inside <step>" for example, would fix those particular problems, as long as you're not using models that are less good at following instructions (some of Google's models are borderline impossible to prevent adding comments all over the place, as one example). So prompt it to not care about solutions, and only care about finding the root cause, and you'll find that you can mostly avoid the annoying parts by either prescribing what you'd want instead, or just straight up tell it not to do those things. Then you iterate on this reusable prompt across projects, and it builds up so eventually 99% of the times the models do exactly what you expect.
• bontaq 11mo ago
  I would say a lot of people are only posting their positive experiences. Stating negative things about AI is mildly career-dangerous at the moment where as the opposite looks good. I found the results from using it on a complicated code base are similar to yours, but it is very good at slapping things on until it works. If you're not watching it like a hawk it will solve a problem in a way that is inconsistent and, importantly, not integrated into the system. Which makes sense, it's been trained to generate code, and it will.