prettyblocks

But many people are doing just that, automating workflows through prompt templates and skills.

I can’t help but wonder whether constant use of “agent” harnesses will lead to an atrophy of the software engineering (or really any field) muscles.

It will, but I'm not sure the impact of this will be all too great. We suffer from not knowing how to use an abacus because we have a calculator, and people who feel a pull to keep their low-level chops up will do so anyway.

The default system prompt is so lean:

https://github.com/earendil-works/pi/blob/main/packages/codi...

Yes, if you're just going down a list of jira tickets and implementing product features, but for many folks it's also an art medium.

Mostly yes, but I think it's both, because some people become life long fans, like the adult Pokemon card collectors.

In my experience most people with the type of critique I'm seeing from you have only tried it one time or have not taken the time to invest in an environment/process that will work for agentic coding.

My question is not so much about sharing a cherry picked example, but the question was more like "have you tried in earnest to make it work". That's the part that wasn't clear from your original post. But you say you have, and you weren't impressed. Fair enough. I'm not trying to convince you otherwise, but I encourage people to give the tools a fair chance before throwing up their hands and deciding it's meh.

Having said all that, you're right there isn't a shared experience.

I'm curious. What have you actually tried? Are you just prompting the LLM with one off tasks? For good results, you need to take the time to read the documentation for the harness you are using and configure your environment. This tuning can take dozens hours to nail down. Then there's the actual approach for working on your projects. Many people that have good results with agentic coding actually spend the bulk of their time in plan mode where they go back and forth with the LLM designing a granular playbook for the task at hand before they ever have it write any code.

#2 is not really an option though. It's more like #1 or #3.

If you use lmstudio it will tell you which models will fit in what you've got available.

It's great, but I wish I could use these things without it feeling like my laptop is going to melt through the desk.

You can also use JS to inject a CSP and it will enforce the strict most policy.

There are paid solutions for this, I've seen Exa recommended.

I've been relatively impressed with local llms, but by far the hardest thing for me is how hot they make my laptop run.

I believe it's 9k/year in parking revenue.

However, in hindsight, Leary proved to be a false prophet who helped destroy the 1960s movements by pushing young people to take a drug that fried their brains and diverted their energy from political activism.

This is a pretty bogus claim. False prophet maybe, but the idea that political activism fizzled out because everyone was eating acid seems pretty unsubstantiated.

Back in the day when I was in HS, kids would go to Borders and swap the stickers from cheap books onto expensive computer books.

For me the vision/OCR is much better than other models in weights class.

Yes ai can’t see, it only understands numbers.

I've also used AI to build frontends that I'm more than satisfied with, and I think it can "see" perfectly fine. The frontier models are multi-modal and pretty good at vision. You can hook up your coding harness to your browser which will take screenshots of your rendered frontend and modify the code accordingly.

I think it's safe to assume that "AI help was used" in everything these days and I don't understand why that should be a problem.

I think it really depends what time of year you go. Zion is beautiful, but when I went it was early in the season and a lot of the popular stuff was closed and it was very crowded everywhere. On the same trip I did all 5 parks in Utah, and I easily rank Zion as my least favorite, even though people seem to love it the most. I would definitely go back though.

In this case it's not basic stuff. You would need to prove that you own the actual bitcoin or transfer it for it to be collateral on a loan. It's the same as spending it.

Adam Back is already a high profile target. Unmasking him as Satoshi doesn't really change that for the guy that founded that company that leads bitcoin core development.

how is codex sandbox different from /sandbox on claude code?

It's the AWS/Serverless model.

I think good results come from spending a substantial amount of time reading the claude-code documentation and setting up an environment that lets you work in this new paradigm. Asking it a few things, getting poor results, and deciding "it doesn't work when you ask it to correct itself won't get you that far.

I asked claude:

take this game for ti-89: https://gist.githubusercontent.com/mattmanning/1002653/raw/b... 10eaae3bd5298b8b2c86e16fb4404/drugwars.txt

and make a single page web app in one single html/css/js file where it draws the ti-89 on the screen and you play the game in that calculator

---

It came up with this in one single shot:

https://drugwars-8od.pages.dev/

I imagine you can do this with a hook that fires every time claude stops responding:

https://code.claude.com/docs/en/hooks-guide

I'd like to read more about that if anyone has any suggestions.

I don't think you will get that anytime soon because for a model to work well with something like openclaw it needs a massive context window.

I think the biggest case for fine tuning is probably that you can take small models, fine tune them for applications that require structured output, and then run cheap inference at scale. "Frontier LLMs can do it with enough context" is not really a strong argument against fine-tuning, because they're expensive to run.