I work in IT, and surprisingly I use AI for about 5% of my work. Mainly because the suggestions are almost always useless and if I’m troubleshooting, I don’t exactly want to sit here reading irrelevant AI summaries of menu options that don’t even exist. I still use it for scripting, but I audit the output entirely and test it out before deploying it widely. Call / meeting transcript summaries are also very helpful and other menial tasks like powerpoint creation.
But my coworker is senior to me by a couple years and he’s technically my direct report. And I’ve stood behind him and watched him frantically type what users describe to him during a live call into Claude and follow the suggestions line by line and get frustrated when they don’t work, and then turn to me and ask me what I think without really trying anything himself. I feel like I’ve solved more of his problems than him solving mine. Every single troubleshooting call, claude is pulled up immediately.
A lot of the issues are either repeat or similar to each other, and even without detailed documentation I think one should be able to recognize some patterns and know what to look at first. But he’ll complicate simple issues by trying AI suggestions then when they don’t work, doubling down on what the AI think is wrong and try to find different ways to achieve its suggestions.
Elsewhere, I’ve had to delay retrieving old computers because the non-technical user decided to start a huge complicated application project in Claude on the old computer, and it’s not easy to transfer contexts and conversations over to a new computer.
I don’t know man. I’m just tired
I think the only people I can get along with on an intimate level are creatives and teachers because at least the probability of them not relying AI entirely in their lives is higher
I have experienced a programmer with decades of experience spend hours trying to get ChatGPT to fix an issue he was having with an external library. I solved the problem in five minutes by reading the God damn documentation.
I was also once given the thankless task of consulting for the CEO of a manufacturing company who had convinced himself that the magic wizard in the computer could turbocharge his business. He wanted to cast an AI spell that would summon hitherto unknown leads for his B2B sales and he insisted that it should be done in Microsoft's cursed Copilot Studio web interface which is really an application for building annoying chat bots to avoid doing customer service. I was charged with fixing the clusterfuck he had set up, seemingly using prompts generated by other LLM's
It was so extremely frustrating to work with, nothing was documented, and instead of being able to program your logic with coding like a civilised person you had to set it up with some bullshit drag and drop web interface. Testing anything took ages because you had to wait for the AI to hallucinate information on completely irrelevant topics before you could see if stuff worked. But you couldn't really see if stuff worked because LLMs are non-deterministic and the prompt that provided something resembling something close to useful in one run would output utter shite in the next. He wasted too much of my time and his own money trying to cast this spell until he gave up and ended the partnership. He would have gotten more value for money by hiring an intern to do Google searches and taking notes for a couple of days.