Piloting Agentic Engineering - What Software Engineers Can Learn From The Aviation Industry • Maximilian Walterskirchen
mwalterskirchen.dev
AI coding agents are amazing, but lean on them too hard and your engineering skills atrophy. Aviation already lived through this. Here's what we can steal from how they fixed it.
  • sem@piefed.blahaj.zoneEnglish
    2·
    1 hour ago

    I’ve seen 2 of the 3 crashes on Mentour Pilot, and if memory serves, it was more about the systems being designed in a confusing way and having poorly understood failure modes, not “the pilots forgot how to fly the plane”.

    So I can’t take the rest of the article seriously.

  • moseschrute@piefed.socialEnglish
    6·
    7 hours ago

    Their findings are uncomfortable: code churn (lines reverted or rewritten within two weeks of being committed) has roughly doubled since the pre-AI baseline.

    Whether you are hand crafting code or using AI tools, you goal should be to solve the problem in front of you so well, that you don’t have to touch that part of the code for years. It’s not always possible, but that’s a good goal to strive for.

    A better way to leverage agentic coding imo is:

    • Pretend you’re a senior engineer and you hate this implementation. What would you do better
    • Find the edge cases in this PR and write test cases to prove them
    • Quiz me on my understanding of this part of the codebase
    • Here are very detailed descriptions of real user flows. How can we design tests to cover this behavior
    • Etc

    There are a lot of idiots that just put Claude on autopilot and merge everything without reading the code. Thats like if the airlines got rid of the pilot and only had autopilot in an empty cockpit. Of course things are gonna go wrong

    • pageflight@piefed.socialEnglish
      2·
      2 hours ago

      But subject matter experts provide a clean, well optimized abstraction — the programming language — so it’s OK not to understand assembly unless you’re solving a very specialized problem.

      LLMs are not experts, do not provide any consistent abstraction, and do not indicate you from the details of what they produce.

  • matsdis@piefed.socialEnglish
    5·
    8 hours ago

    New engineers shouldn’t just learn how to prompt. […] The signal that an agent is bullshitting you is a learnable skill, and right now we’re mostly learning it by accident.

    No. Education should focus on basics that are likely to remain relevant. The biggest signal that something will remain relevant is that it has been relevant for more than a decade already. Laws of physics, the PID control loop, what is a register, what is a LRU cache, asymmetric cryptography. Failure mode effect analysis, stuff like that. LLM prompting is very new. Better learn about big-O notation first, or you’ll never realize that the LLM went off rails. They didn’t teach you the latest Javascript framework at University either.

    A simulator for engineers. This is the one I haven’t seen anyone build, and I think there’s a real gap.

    We are having big fun with those.

    A simulator for engineers.

    You haven’t played Factorio, have you? ;-)

    [A simulator for] debugging unfamiliar production-like code, reasoning about state in a real system, recovering from a nasty incident without help. Someone should build that. (Hit me up if you already are. I would be very eager to try this.)

    You probably have been building mostly new software, and not yet had the pleasure to maintain something that was built two decades ago by a team that isn’t around anymore to maintain it. There is a big market for the skill to work on high-value legacy systems without breaking them. This kind of work that you don’t see in the hyped blog posts. (Or if you do, it will have “post-mortem” in the title. In fact, you have succeeded if your work on those system never makes it to the news.)

    (Edit: The problem is not building this simulator. The problem is finding both the budget and the cruelty to beat an engineer into analyzing a legacy system that is currently working as it should. At the end of the day they are frustrated not having done anything, and the company has spent money with no tangible result. I guess we really could learn something from aviation - this kind of “getting intimate with the system” for its own sake just isn’t valued.)

  • Asetru@feddit.org
    9·
    10 hours ago

    Like most of my colleagues and friends, I’m a heavy Claude Code (or insert hyped agent harness of the week) user. These tools are genuinely amazing.

    Lost me there in the first paragraph.