Jonathan M
79 posts


@leonpalafox @freed_dfilan An agent harness is essentially a while loop with tool calls. Believe it or not, humans can write something like this fairly easily. LLMs did not "enable" this agent harness to be created. The ask was for things "way ahead" of human ability alone.
English

@jonathanmcdill @freed_dfilan Claude Code was literally the piece of software that put Anthropic in the positive cash flow column, the LLM was not it.
And the point isn’t even that, people are asking what products are being delivered vibe coding
English

@leonpalafox @freed_dfilan No it's a major accelerator for using LLMs. You're missing my point. But it's just not a revolutionary piece of software. There are many agent harnesses, human written and otherwise. The fact that Claude Code was vibe coded didn't put it massively ahead of the others.
English

@jonathanmcdill @freed_dfilan If you think all the infrastructure around the LLM is minor good for you, more power to you.
English

@SyntheticBeef @tomfgoodwin Yea but just because you’re doing something, does that mean you’re getting massive discernible value from it?
Also-you’re not using eldritch space orchestrator builders to write your orchestrator builders yet? Yikes. Gonna be left behind.
English

On a linear timeline, ai can exist in the middleground.
But that's not the timeline we are working with.
I wouldn't trust this Anthropic if I weren't living it myself. I've gone from using LLMs to write code to using agents to write code to using agents to write agents to using orchestrators to manage agents and next I will be automating building orchestrators.

English

For what it's worth, I continue to find it
Magical, but often less VALUABLE than expected.
Transformative, but transformations take a long time,
Incredibly fast to improve, in many ways, but how we use it remains the bottleneck.
It's astonishingly good at lots of things we don't really need.
Writing code or copy or images or making videos, or summarizing text, is not the great unlock of value we think. It's more like a party trick.,
Almost everything will end up in the middle, The demand will balance out. The costs will balance out.
We will slowly find a way to rethink some companies around it.
We will slowly find ways to reshape what companies do
Many companies won't change at all, and they'll be absolutely fine
Those who are first to use it won't really benefit, It's the companies that use it the right ways at the right time and know when to say no that will
Everything is almost entirely destined to be completely in the middle of expectations. The wisdom of crowds and all that
English

@leonpalafox @freed_dfilan The historical, revolutionary thing is the LLM. The thing humans created. Not the little agent harness thing that the LLM helped generate.
English

@jonathanmcdill @freed_dfilan One of the largest IPO in history and is not good
English

@haider1 There’s a kind of coping on the AI cheerleader side as well though. “If I’m all in on AI then I’m on the good, smart side of things. I’m accepting reality.” That’s still not going to save your job.
English

@freed_dfilan Im seeing a lot of the replies to this question asserting that LLM written software is “as good as” hand written software. This misses the point. If AI was truely as game changing as people say, (10x, 100x), we should be seeing a whole new class of software. But we don’t.
English

@leonpalafox @freed_dfilan And it’s not good. It’s not a revolutionary piece of software beyond the LLM powering it.
English

@freed_dfilan Anthropic literally said that the latest Claude release was mostly vibe coded
English

@tlakomy What about understanding code though? Not necessary?
English

@RoaringHammy If AI becomes good enough to take my job, then I can use it to create a billion dollar thing. So either way I'm good. I will have the same access that a company does to the magical thing if it replaces me.
English

@housecor Your new job is coercion into a subscription based tool that creates worse software... but faster!
English

@vanschneider I think the main thing is it’s not just “another abstraction,” it feels more like someone doing your job for you. Not much satisfaction in that. That week that it used to take to build something was a hard but worthwhile journey that you grew from. Now you just have a cheat code.
English

Spoke with many friends recently, designers and engineers, all people who've been doing it for 20 years or more and had a lot of success doing it.
And they all say the same thing. The AI stuff is genuinely useful right now. It's fast and things that used to take a week take an afternoon. Things you never even attempted because there was no time, now you can just do them. It's the biggest enabler ever.
But in the same breath, every single one also says that it's the least fun they've ever had in their entire career. They also mention it makes no sense to do it the old way. They're all in.
It's a strange paradox which I feel myself. Everything is possible now and I've never cared less about any of it. Both things true at once.
Not sure if thats just the feeling of the current moment, or if I just talked to people who're tired of the computer (since all of them been doing it for a long time).
English

@buccocapital I’ve been wondering about this problem too.
x.com/jonathanmcdill…
Jonathan M@jonathanmcdill
There is a strange paradox going on with the AI revolution when it comes to coding. By no means am I an ML expert, but it seems to me that the more we rely on AI, the less good it will be. See thread for why.
English

@tapasadhikary Yes, but it will be something worse. This is how these things go. Futurist theories usually don't pan out the way we think. Enjoy your keyboard while you have it.
English

@mischavdburg I think the point is that AI, by itself, pretty much produces slop. You said that you all but type the code yourself and then it's good. Yeah.
English

AI code isn't slop.
But most of what people show me is, and I keep thinking that's a workflow problem more than an AI problem.
When I give the model a real spec, tell it how I want my Python modules separated into neat files, what structure to follow, that it should stick to clean code, and then I actually review what comes back and understand it, the output is genuinely good. Well-specced, efficient, follows best practices.
I built the whole KubeCraft AI app this way, mostly thinking and prompting and reviewing rather than typing.
Maybe the slop was never the AI. Still turning that one over.
English

@forloopcodes I just wrote about this, any thoughts? x.com/jonathanmcdill…
Jonathan M@jonathanmcdill
There is a strange paradox going on with the AI revolution when it comes to coding. By no means am I an ML expert, but it seems to me that the more we rely on AI, the less good it will be. See thread for why.
English

@housecor I empathize as well, but appreciate this list of positives. Thank you for not saying "velocity" or "productivity". There are definitely parts of coding that suck that AI has made much better. But AI has somewhat taken the joy of honing your craft (if you let it).
English

I empathize with this.
What helped me turn a corner: I focus on the things I've gained:
I rarely get stuck anymore.
I don't have to type tricky syntax.
I can improve UX faster than ever.
I can easily automate tedious tasks.
I don't have to manually look things up.
I have a new "pairing" partner that's always available.
I can implement and compare multiple options quickly.
I can try big ideas that were previously out of reach.
I can use the best tool for the job instead of the one I know best.
Jeremy 💻🎮🔭@pcnerd37
@housecor All of the the agentic stuff makes me sick. I've been doing dev since 2004 and this is making me want to find another career. This is no longer software dev and I want no part of it.
English









