If claude only did what we tell him...
9 karma · joined November 28, 2025
I guess solution had not yet appeared in training set.
Let them taste their own medicine. :)
I would not understimate what folks in 70s had done.
Same story. The model works perfectly when called directly, but the agent integration doesn't actually execute the tools. It generates things like:
<tools> {"name": "read_files", ...} </tools>
as plain text and then just sits there until I tell it to "do it" again. It repeats the exact same fake tool call.
So instead of:
Qwen → tool call → grep/find/read → result → Qwen
I get:
Qwen → pretend tool call → nothing happens → repeat.
> Open documentation url for more info? (Y)es/(N)o/(D)on't ask again [Yes]: D
Update: after actually trying Aider, I don't understand the point of it.
It feels like browser AI chat moved into a terminal, except now it's on your own dime, with no freemium. Ask it `find test_main location` and it won't actually run `find` and inspect the repo. It tells you it can't access files you haven't explicitly added.
That's basically browser chat wrapped in a terminal with a few pre-existing skills.
Unfortunately, my experience was extremely disappointing. This happened on Fedora.
Instead of responding to simple prompts, it got stuck in what appeared to be an infinite loop producing nonsense like:
> Objection > work state > active > blocked > next move...
Even the response-copying functionality didn't work, so I couldn't easily copy the output. (Using Terminator Terminal Emulator)
And no, the model itself isn't the problem. I manually called the exact same llama.cpp OpenAI-compatible endpoint with `reply with Hi`, and it correctly replied `Hi`.
I also checked the OpenCode configuration and corrected the model ID to exactly match what the OpenAI-compatible `/v1/models` endpoint reports. Same result: endless nonsense.
So something in the OpenCode integration/agent prompting is clearly going wrong in this setup.
And please don't tell me to "read the docs." I've never read a single line of Claude Code documentation and have been using Claude Code without issues for several months. If Claude can make the experience that seamless, I'd expect a project with 200k stars to provide at least a usable experience.
Maybe I'm missing something obvious, but I wasted ~20 minutes trying to get a basic setup working.
Posting this in the hope it saves someone else 20 minutes.
1st one requires that you actually care
2nd one relevance depends on what are you doing, not everyone works on airplane software
3rd one, novelty for me
Employer monopsony can also produce less with fewer people and keep demand for labor below availability. This will also increase prices due to scarcity of the product. Just give me employer monopsony, government in their pocket, and I will give you N ways to screw the workers.
> workers share in those gains
Not gonna happen until workers obtain bargaining power and they are working hard to keep it that way.
> will it devalue labour or not
I agree with Kent Beck here: “The value of 90% of my skills just dropped to $0. The leverage for the remaining 10% went up 1000x.”
Also, AI makes workers do more for a less.
Yes, experts are not the buying ones. The buying ones are the paying ones. Stock holders. And they can lie them as easy as I can lie to a patient how I am curing 10k lines of code every day.
Then every expert who isn't influential enough to be on their payrole isn't an expert?
Dont listen salesman talking about engineering. Unless he has credibility proven by things he had done recently. Person might be zillionare but he cant buy credibility.
With scale problems arise.
He might be right in his context, but there is no silver bullet.
Some of us are less fortunate and work in companies where every deadline was yesterday. Every task is wildly different and often means jumping into someone else’s project that you are seeing for the first time. You don’t get unit tests. You don’t get proper documentation. You don’t get decent architecture. You don’t get any of the safeguards he is talking about. You get a Jira ticket and someone asking why it isn’t done yet.
The team that owns the project keeps it running like a broken company car. Nobody wants to fix it because there is always something more urgent. Then six months later you are somehow expected to apply all the best practices that should have been there from day one.
In that environment, making all of this a prerequisite for shipping is a pretty good way to end up on a PIP. And then out of a job.
I’m not saying he is wrong. I’m saying his argument assumes an engineering culture and organizational discipline that a lot of developers simply do not have. And more importantly, their companies have absolutely no intention of introducing.
“Just add tests, static analysis, architecture constraints and automated verification” sounds great until you are the person who has to introduce all of that into a 10 year old dumpster fire while being told the deadline is tomorrow.
There is no silver bullet. Sometimes you are just trying to ship the damn thing.
But if you try to delegate your work to reviewer, by dumping slop and waiting for reviwer to catch issues, so you could feed that in LLM and again do nothing, your contributions should not be accepted anymore. And this is not generated code issue, code generation just make it more common. Some people will always think they are smartest while everyone else is dumb, can't fix that.
To see that it works you first need to be some famous "programmer" who is on someone's payrole, basically doing their marketing.