I tire of all of this discourse at this point because people who don't need to hear it are reading it and nodding, and the people who DO need to hear it are busy making everything worse, earning leadership accolades (or are the ones handing them out), and are completely closed to any criticism or input.
Anyone with intellectual curiosity and any sense of subtlety or balance is operating in a separate reality from people who are so AI-pilled and very proud of it and there's basically no possible cross talk (usually because the AI-pilled people are the ones who own all the shares).
I say this as someone who really likes LLMs and what roles they'll play in software engineering in the future (and already are, right now).
I can't wait until the AI craze is in the same spot that the crypto craze is currently. People are still using and building with crypto, but the sale of ape PNGs has mostly stopped, and now if I have a problem that a blockchain could solve, I could actually talk about it instead of accidentally picking up a battle standard in the middle of a trench war.
I have to disagree a bit here, people (including me sometimes) can take the lazy way out and just tell you what an agent said in response to a question which we shouldn't do. At that point it's little better than a let me google that for you link. However, while I may be passing along what I found using an agent, but usually I still had to do the work of prompting it, asking it to lookup information for me, and otherwise filtering the agent down to the relevant information. There is a good chance that I'm doing that because of my specific expertise in the subject area which is something outside of what you could have found by just asking an agent yourself.
Yes, a subject matter expert who uses their domain experience to word the prompt correctly, incorporate tacit knowledge and relevant sources, ask follow-up questions, and judge the LLM-generated responses for relevance and correctness is not taking a shortcut. They're using a tool to provide a better and faster answer than they would have been able to otherwise. It's a matter of how the tool is used and maybe the optics around how the response is communicated, but relaying Claude's final answer is not inherently the wrong thing to do.
The one thing I do find extremely frustrating is how there can be some ridiculously painful obstacle that hits you in the face all the time when trying to ship code, and everyone is seemingly sympathetic, but it can't be changed for dubious compliance reasons, or maybe because if the barriers were lowered it would threaten the quality control for the code. Fair enough.
Then, of course, all of those speedbumps are demolished in record time, only for the AI agents and not for the humans, regardless of what caveats remain unsolved, with little regard for whether or not it meets compliance standards.
The obstacle is fixed. Just ask the AI agent to bypass it for you!
I'm reasonably opposed to LLMs, mostly from a human perspective. They reduce us to our output without factoring in any of the actual thoughts and experiences that go into that work, it's literally not in their training data.
It's unfortunate that we live in a world that has decided to push in this direction.
I have started a work project that is 100% vibe engineered, as an experiment to see where it would land. While it seems to be working, I can see that it has made so many mistakes along the way that it is still unclear if we can dig ourselves to production ready.
I was at an art museum recently, and was looking at a marble statue. And I thought to myself: this artist probably spent more time sculpting the big toe than anyone has spent appreciating the entire statue.
That speed is what we used to have in software engineering. So much careful craftsmanship, no line just yeeted out. And now we're just heading into a direction with no clear end, just prayers that bigger and better models will lift us forward.
What's the point of being human if we're so gung ho to remove that which separates us from other animals (our intellect)?
We’re currently entering that time and people are quickly learning that empirical evidence is the new output. It’s the difference between “I think the bug is here” vs “I fixed the bug and here’s the evidence I gathered from user telemetry”
This is pre-AI speak. Post-AI assumes we all maintain multiple concurrent chats and use that to search for information. “Claude says” becomes the new “I googled this”.
I started writing this as satire and I don’t like how realistic this gets.
I still don’t like it, if only because the power imbalance seems somehow worse than with Google.
Disclosing that you got some info from an AI is way better than not. We’re all using AI in some capacity just be transparent about it. If you didn’t read the findings then say it. As always there’s nuance and it depends
No problem, the next time you have an issue or ask if anyone has any idea, I'll just ignore you and mind my business, confident that you can do the prompting yourself. Voilà, problem solved.
If someone answers your question with "Claude says..." it's a clear indication that you should've asked the question to an LLM yourself before bothering a human.
This is just the new version of LMGTFY.
> AI is the reason our software stack is overengineered to the 10th degree
There's definitely times where the question is 'hard' enough (usually having to do with something so business specific the LLM won't have a great answer) where a true human answer is needed, but I do regularly still get asked questions of 'why is x failing' with a datadog link. And if you shove the exact logs they've already found that contain a stack trace, the LLM instantly tells you exactly what was wrong.
It's become so frustrating when literally everyone is just using Claude for everything. Simple messages and updates are fed through Claude. It's come to the point where any human language in corporate is punished.
Anyone with intellectual curiosity and any sense of subtlety or balance is operating in a separate reality from people who are so AI-pilled and very proud of it and there's basically no possible cross talk (usually because the AI-pilled people are the ones who own all the shares).
I say this as someone who really likes LLMs and what roles they'll play in software engineering in the future (and already are, right now).
I can't wait until the AI craze is in the same spot that the crypto craze is currently. People are still using and building with crypto, but the sale of ape PNGs has mostly stopped, and now if I have a problem that a blockchain could solve, I could actually talk about it instead of accidentally picking up a battle standard in the middle of a trench war.
Then, of course, all of those speedbumps are demolished in record time, only for the AI agents and not for the humans, regardless of what caveats remain unsolved, with little regard for whether or not it meets compliance standards.
The obstacle is fixed. Just ask the AI agent to bypass it for you!
It's unfortunate that we live in a world that has decided to push in this direction.
I have started a work project that is 100% vibe engineered, as an experiment to see where it would land. While it seems to be working, I can see that it has made so many mistakes along the way that it is still unclear if we can dig ourselves to production ready.
I was at an art museum recently, and was looking at a marble statue. And I thought to myself: this artist probably spent more time sculpting the big toe than anyone has spent appreciating the entire statue.
That speed is what we used to have in software engineering. So much careful craftsmanship, no line just yeeted out. And now we're just heading into a direction with no clear end, just prayers that bigger and better models will lift us forward.
What's the point of being human if we're so gung ho to remove that which separates us from other animals (our intellect)?
I have been in meetings where the conversation has been 'my Claude has got feedback', 'great, I'll get my Claude to take a look'.
When do you just cut out the human middleware in that process?
I started writing this as satire and I don’t like how realistic this gets.
I still don’t like it, if only because the power imbalance seems somehow worse than with Google.
If someone answers your question with "Claude says..." it's a clear indication that you should've asked the question to an LLM yourself before bothering a human.
This is just the new version of LMGTFY.
> AI is the reason our software stack is overengineered to the 10th degree
Complete BS.