I know these kinda questions get asked a lot these days, but hear me out.

I’m a junior-going-on-medior software engineer, graduated last year and have been (more or less) teaching programming for four years.

In my search to find meaning in life, I stumbled into programming and found that there was a lot to learn. I loved it. That’s why I eventually chose to switch careers once I musterred enoigh resources to do so. I chose this job because I like learning and I want to keep learning for the rest of my life. I love the way how learning is an integral part of the programming community and how everyone encourages and stimulates it.

AI has changed the job of an swe, that much I understand. That is not a deal breaker for me, as long as there is still a lot of meaningful, worthwhile, marketable things to learn. I am using AI to learn things faster now, but there is an easy pittfall to actually learn nothing from AI. Some say we don’t have to learn anything anymore, I don’t think that is true, but then what is true?

How is AI affecting the learning aspect of this job? Is learning still an important part of the job? Is there still enough worthwhile things to learn? Is it still valuable to learn more?

  • Ftumch@lemmy.today
    link
    fedilink
    arrow-up
    4
    ·
    5 days ago

    It’s easy to pump out lots of shitty, insecure software using LLMs. My hope, however, is that in the future good programmers will use AI to create better, more secure software.

    First of all, LLMs still suck at software architecture. It’s up to a human to define the overal architecture and structure of a project.

    It will also require a human to define the requirements, constraints and tests a module needs to conform to. With sufficiently clear and restrictive requirements and conatraints, the LLM can then generate and modifie the code until it meets all the requirements and produce something acceptable.

    I also expect AI will make formal verification, where you mathematically prove your code meets certain criteria, feasible for more than a handful of projects. It doesn’t really matter if a vibe coded formal proof is unreadable and convoluted. If the proof passes verification your code is correct (assuming your assumptions weren’t wrong).

    So IMO useful skills for the future include software architecture, compile-time guarantees and advanced type systems, test-driven development (TDD) and possibly formal verification.

    Personally I’m trying to learn Lean 4 and formal verification right now, to see how well AI can assist with proofs.

    • tea_n_typewriters@lemmy.world
      link
      fedilink
      arrow-up
      4
      ·
      5 days ago

      It’s easy to pump out lots of shitty, insecure software using LLMs.

      That’s just not true. I told Claude to fix any security holes.