You might be interested in the book @swardley.bsky.social and I are writing at moldabledevelolment.com
07.03.2026 05:04
👍 0
🔁 0
💬 0
📌 0
You might be interested in the book @swardley.bsky.social and I are writing at moldabledevelolment.com
Perspective
"Please continue"
That's what the documentation for Claude Opus 4.6 says when you happen to get empty messages from the magic box.
I'm not sure whether "Please" is necessary, but I didn't dare try anything else for fear of upsetting the machine.
Nice one :)
The greatest architectural decision of our time: where we value the human in the decision making process.
Understanding systems is not the only problem to solve, but it is the most costly and consequential one we face today.
I missed your reply.
It does look like it's about figuring systems out. I argue that it starts from the wrong baseline and draws the wrong conclusion.
A short thread about the baseline:
bsky.app/profile/tudo...
The alternative: deterministic contextual tools. Thousands per system. LLMs can help.
"How else?"
The starting point is to accept the problem. Then we must make it a subject of conversation. And if you cannot quite find the starting words, borrow them from the book @swardley.bsky.social and I are writing in the open:
bsky.app/profile/tudo...
8/
We cannot pretend it does not happen, and we cannot pretend it is not important. Because of that, we have to make our way out of it. And simply asking LLMs is not that way.
7/
If we would not aim to build a whole new world in digital form, this would be of little consequence. Alas, we do. If it would be an isolate case, we could perhaps just accept it and carry on. Yet, the past success of StackOverflow doubled by the numbers in those surveys document pervasiveness.
6/
These are the very premises of engineering. And you have given them up.
"And what does this have to do with LLMs?"
If the baseline we compare against is a state of despair, a faster answering is actually not progress. It's just faster serendipity.
5/
That is not a sign of sound engineering. That is a sign of despair.
"Why despair?"
Because you gave up hope of the possibility of understanding cause and effect. Of comprehending how the situation can be corrected and what the impact of that action is.
4/
"What does this have to do with anything?"
Everything. Consider this: You have a problem with a system you engineer. And the best possible option is to go somewhere else to ask a question and hope that someone that has never seen your system has the solution...
3/
"What do you mean?"
In 2020 and 2021, StackOverflow asked an interesting question "What do you do when you get stuck?". The top considered alternatives were between googling and visiting StackOverflow.
2/
"Claude Code is the most effective diagnostic and debugging tool in computing history"
You might want to check the baseline you compare against.
1/
1) Software engineering is not about building features
2) Refactoring is a business problem
3) Software engineering is not an engineering practice
4) LLMs will not replace software engineers
5) Architects don't make decisions
by @swardley.bsky.social & @tudorgirba.com
medium.com/feenk/rewild...
Beautiful summary! I’d love to learn more about that work you describe in which you use LLMs to accelerate your data understanding.
I’d love to hear more details about what you find interesting about it.
@swardley.bsky.social and I are documenting a different perspective at moldabledevelopment.com. Especially the 6th chapter tackles this very problem. If you feel inclined, I'd be happy to have a conversation about it.
I understand why one would say that Claude Code is the most effective diagnostic and debugging tool. It's indeed much faster and better than going to StackOverflow.
But that perspective is misinformed:
bsky.app/profile/tudo...
That said, you should still heed the overall advice of exploiting LLMs.
I am not sure I understand the comment :). Would you want to elaborate?
I do agree that LLMs and agents are useful in software engineering. I still disagree with the overall argument:
bsky.app/profile/tudo...
@swardley.bsky.social and I just published a chapter that tackles this but from a different angle:
bsky.app/profile/tudo...
I'd be interested in your opinion.
The instinct might be to want to write code, but that's just because nobody talked about the part that was most expensive all along: figuring systems out. Because we did not talk about it, it's highly uneducated and unoptimized. That is why Claude can pass as a great debugging machine.
When myths fall apart, new opportunities arise.
"What opportunities?"
Figuring systems out without reading code. Oh, and without asking LLMs to summarize them.
"And how is that?"
Go read the chapter. It's long, but I promise it's not boring. And it has pictures, too.
moldabledevelopment.com
4/
While seemingly quite different, we argue that they all are rooted in misunderstanding the nature of software engineering. Once we accept that software engineering is primarily a decision making activity, and that we can understand systems through contextual tools, all these myths fall apart.
3/
"Software engineering is an engineering practice"
"LLMs will replace software engineers"
"Architects make decisions"
"Dealing with legacy is hard"
"Non-technical people can’t understand code"
"Technical debt exists"
2/
@swardley.bsky.social and I just published Chapter 6 of our Rewilding Software Engineering. It's about 8 myths in software engineering.
Linked at: moldabledevelopment.com
"Software engineering is about building functionality"
"Refactoring is not a business problem"
1/
Nice idea
Cloudflare outage was due to a small change.
Azure outage was due to a small change.
Amazon outage was due to a small change.
Please, tell me again how details don't matter in software engineering.
It seems that AI does not magically solves the legacy problem.
by @raymyers.bsky.social
ainativedev.io/news/ai-hate...