sroerick

Third party apps and commercial software will be the slowest to adapt to the new paradigm.

I use Synthetic, Neuralwatt, and Opencode. My usage is fairly high, I'm at 1.5B tokens in the last 7 days. ;)

Yeah, I'm sorry to hear that. I'm very glad I am not dealing with anything even remotely corporate during this time, because I'd absolutely hate it

I make 55k working at a dental office and a restaurant and I'm paying 200$ a month on open source models because I'm having an absolute blast with them.

Plenty of jobs have expenses built into them. Maybe a subscription is more than it used to be, I had great jobs on a used 200$ thinkpad, but it's still cheaper than being a photographer. And 100k is more than most jobs.

Software dev, even if it's all agents now, is still pretty dang economical

This is my first attempt at a AI framework and I'm pretty happy with it

https://pricklypear.rocks/

OCaml.

I haven't used the US labs in months. Kimik3 for planning and GLM5.3flash or the new Deepseek Flash for code.

Man, I'm sorry for the lack of original thought, but I am so sick of AI websites.

I would seriously rather see a plain unstyled html page with a marquee tag saying "A HUMAN WROTE THE NEXT THREE PARAGRAPHS"

I switch between Pi, Autolith, and my own harness. I come to this website. I don't know what the hell a strand agent is. I don't know who is putting this out. I don't understand whether I'm meant to use a GUI or write code in order to configure it. All the screenshots show a GUI. "use one line of python or javascript to configure." I don't know if this is a harness, an orchestrator, an agent framework, or all of the above. I think it's hard. People get so caught up in making their own thing - they lose track of the state of the art. I have no idea what the value prop for Strands Harness is. But maybe I'm not the target audience.

Also, I don't really care about 30% token efficiency gains. Maybe I should. But this is not my experience. Don't get me wrong, I'm conservative with tokens. But I also have a LOT of convenience / essentials which bloat up context. And I'd rather bloat context and have it work well than not have it work well. And when GLM 5.3 Flash came out - all of a sudden I had more tokens than I knew what to do with. That's when I started using OpenClaw type agents for the first time.

I don't have the tokens/cash to throw at something like Gas Town or OpenRig. And I'm still not at all convinced that those software factories are actually the future of everything. Maybe so. But I just don't see why I would invest the time to learn a piece of software without really understanding the value prop. It's crazy to say, but it's just not enough anymore for software to be free. Because the cost of me implementing my own software is rapidly approaching free. And I like mine better.

"Oh, well I just need something that works out of the box." Wait, why? What does it need to work for? For making software? lol.

Personally, I build a loosely compiled / linted DSL to handle spec.

It's a SEXP language but could just as easily be markdown. It'd probably be better if it was - I spend a lot of tokens matching parentheses.

Compared to markdown - I lose some human readability but gain a lot in workflow.

For me - it turns the sort of "free form jazz code odyssey" into something with structure. It changes the question to "does the spec match my idea? And does the code match my spec?"

Every so often, I do a "code astrology reset" and try to purge all the tricks I use to keep things functional. I absolutely notice an efficiency hit when I don't have a spec language. Even with the new models - I find this essential to keep things on track.

I think most people on large projects hit a scaling ceiling with markdown specs. They get massive and contradictory very quickly. I think a src/md folder is a good strategy. I try to organize with a "one spec per module" rule. It doesn't always happen that way but I find that helpful.

I think Literate Programming is a good inspiration here. I also think Yegge's beads and Gastown have some really smart things to say in this regard - even if he's a bit tokenmaxxy for my taste.

Can you elaborate on what frameworks you've used? and how they've helped

I totally agree with this. Agents seem tremendously bad at UI to me. Maybe it's just because I am a back end guy.

Right now I'm working on a declarative UI framework which can help me along here. My thought is that if I sacrifice a little control for sane primitives, that will make that spec /build loop easier.

I think ClayUI is a really interesting "reduced instruction set" for UI. I don't know that immediate mode UI is the right call for anything web related (that's how you get React lol) but his reduced primitive layer is very interesting to me

This article hints at it, but I think the blocker is environment.

How can you self drive an app on windows? There's no clear UI framework, design pattern, nothing which can bullet proof your app.

I bet you'd have better luck on Plan 9.

Remove the choice. Make frameworks which have limited options which are usable by default.

I did a similar thing. Yours seems cleaner than mine. I have a 'compiler' and a sexp based DSL. I don't know how anybody vibecodes medium or large programs, say above 50,000 lines of code. I don't read all the spec until I sense something is out of alignment.

Ive wondered if I have too much complexity, and from time to time I do a "prompt astrology reset" where I get rid of all the extra cruft. I can't go without the spec sheets though.

It's funny to me that instead of moving existing tech to a word in word out paradigm, people are dropping millions getting the chat to work with the existing paradigm.

Probably inevitable, but seems like a lot of disruption could happen there.

How could agents take over the internet if compute is still gated within Anthropic / OpenAI? Even if the botnet was controlled remotely, wouldn't anthropic just be able to shut off the controlling nodes API access?

I wonder if we will see a mass (math) exodus of mathematicians to open weight models.

I just say jippity or sometimes "the jippity"

Facebook is the first to market with a polished openclaw? Good for them.

I'm sure many others will criticize other aspects of Meta, and rightly so. But imagine using a Claw agent with zero tech support available

Can you elaborate on this? how would you work on a larger THREE project? would you create models independently?

LLMs are better at OCaml than any other language, and being able to read and think in OCaml is very helpful to understanding LLM generated OCaml code.

Also, it may very well be the decade of formal verification - if so, OCaml is a good place to be.

Couldn't agree more. Even when the models could barely write it, the compiler was so good

Lol

This is funny. I was trying to get agents to talk to each other on XMPP. one of them wrote their own chat room on a Lisp Habitat that I run. then it starting talking (On XMPP) about how nobody was receiving or responding to its messages. On the chat board that it wrote. That it didn't tell anybody about.

I don't know people or talk to people so I'm happy to hear otherwise. Which industries?

I definitely look for libraries which are handcoded and I consider them to generally be of a much higher quality, but it increasingly seems like high performance / critical infra is going to move to formal proofs rather than hand coding

That may be true - but I hadn't used Claude Code and I was still in the "its impossible to vibecode successfully" mindset

Man, as a former extreme skeptic, I watched a video from Eric Schmidt in early 2025 where he said that by the end of the year nobody would be coding, and that one was dead frickin on.

> Steve Russel realized that the Lisp specification of EVAL can be translated to machine language to produce an interpreter.

So just to be clear, you are here confirming that the first implementation of Lisp was in fact an interpreter?

This is just the Wikipedia page for "interpreters" so perhaps this is fake news but

> The first interpreted high-level language was Lisp. Lisp was first implemented by Steve Russell on an IBM 704 computer. Russell had read John McCarthy's paper, "Recursive Functions of Symbolic Expressions and Their Computation by Machine, Part I", and realized (to McCarthy's surprise) that the Lisp eval function could be implemented in machine code.[3] The result was a working Lisp interpreter which could be used to run Lisp programs, or more properly, "evaluate Lisp expressions".

I agree with this and I'd also say - I've been vibe coding in OCaml since before the training data was good.

All the stuff I lost in training data I gained back in having great compiler feedback.

I don't think agents are nearly as restricted by language or training data as we think.