I prefer luna for most development, especially when I am guiding the process. Sometimes terra. I have had terrible results coding with sol. It is way over-tuned on RL to make something that completes the task, no matter what. I end up with way too much code that does a lot of things I didn't ask for.

Try planning with Luna, implementing with Sol with guidelines to not exceed the given scope.

Sounds counter-intuitive at first, but Luna is overall better at sticking with what works. Sol is wicked smart but needs constraints.

IME you're supposed to have Sol drive Luna sub-agents to do 90% of the work. Sol should primarily be the verifier and goal setter. Use omp.sh with Task Delegation -> Always to strongly encourage Sol to drive Lunas. Also Luna prefers to be talked to with English in XML.

I love Luna too. An excellent model and still usually better value per dollar than Gemini if you pay for API tokens. Things may change with 3.8 - we'll know soon.

I setup Luna as main Claude Code driver (so zero anthropic api use) and it nailed crisply a handful of python tasks, gonna continue this way.

Why not use codex or an open source harness?

That is a very valid question. I happen to want to get Claude Code muscle memory under my belt for professional reasons in addition to get side projects advanced, could have settled for Codex else. Also OpenCode with eastern models gets part of the job done. In CC beyond using Luna for the cheap, I am using DeepSeek flash v4 for subagents, that is a further cost shaver. Not sure if in Codex I could do that.

The code I see is a lot like the pelicans. All of the code in codebases, good, bad and ugly, is slowly being replaced by whatever level of code ai is currently able to create. All code is now a slightly wonky pelican on a bike, but if you look closely, it doesn’t fully make sense. Since ai is converging on less wonky, but not internally consistent, we’re just moving on to what is possible with high volume instead of detailed quality. I think that is the ai software world as well.

Software that had a data moat because it was hard to integrate with or migrate off of will have that moat disappear. A web site is a client now. Building data migration too for all of you competitors is easier now.

I've just had a SaaS that I use decide to implement a 2.4x price increase. I reacted instead by taking screenshots of every page of the SaaS, downloading their API docs, exporting what data I could, and asking Claude to build a self-hosted clone based just on those files. I had a read-only version of my entire data history completed in a single evening. Even at Opus API rates, it cost me less than half the price of a single annual seat.

Heh and without api docs, just copy and paste the urls from network traffic and Claude will write a library for you.

One of the many SaaS products we use at Day Job chose to gatekeep its MCP behind an enterprise plan. A brief Claude Code session later and a better, more feature-full MCP than the official was reverse-engineered from internal APIs by Opus.

They will add flooded streets to the training simulation and this problem will go away. Eventually, the corner cases not in the training simulation will be so corner they basically never happen. Waymo can be incredibly successful without dealing with "surprise clown parade" or whatever.

Instagram is gone as well. Everything is fake in different ways. If the video isn't ai generated, then it's influencers acting out a scenario they think will get engagement. I realized that when it's a real video, there is a caption that says some scenario is happening, but there is nothing in the video that shows that is real. I think people are just reposting videos with different captions and testing out whatever invented scenario make the video have the most views.

If had to answer this question 2 years ago, I wouldn't have said software was a "don't see it's bad until later" category, with compilers and it needing to actually do something very specific. However, business slides are full of exacting facts and definitely never contains generic business speak masquerading as real insight /s.

This feels like telling a story after the fact to make it fit.

I agree, and by all accounts the success of coding agents is due to code being amenable to very fast feedback (tests, screenshots) so you can immediately detect bad code.

That's in terms of functionality, not necessarily quality though. But linters can provide some quick feedback on that in limited ways.

Yes. They probably are, but it's moderately hard to eat a healthy vegetarian diet if you look at what vegetarian athletes actually eat.

If llms are able to write better code with more declarative and local programming components and tailwind, then I could imagine a future where a new programming language is created to maximize llm success.

This so much.

To me it seems so strange that few good language designers and ml folks didn't group together to work on this.

It's clear that there is a space for some LLM meta language that could be designed to compile to bytecode, binary, JS, etc.

It also doesn't need to be textual like we code, but some form of AST llama can manipulate with ease.

At that point why not just have LLMs generate bytecode in one shot?

Plenty of training data to go on, I'd imagine.

The code would be un reviewable.

It would also be harder for the LLM to work with. Much like with humans, the model's ability to understand and create code is deeply intertwined and inseparable from its general NLP ability.

Why couldn't you use an LLM to generate source code from a prompt, compile it, then train a new LLM on the same prompt using the compiled output?

It seems no different in kind to me than image or audio generation.

...by a human :)

Hence very important in the transitional phase we are currently in where LLMs can’t do everything yet.

Would this be addressed by better documentation of code and APIs as well as examples? All this would go into the training materials and then be the body of knowledge.

readability would probably be the sticking point

> I could imagine a future where a new programming language is created to maximize llm success.

Who will write the useful training data without LLMs? I feel we are getting less and less new things. Changes will be smaller and incremental.

If a company can align it's business model with user goals, then it can work in the long run. Apple has somewhat aligned it's integrated hardware sales business model with user privacy. Google and Meta are advertising companies and capturing user data and attention will always drive the business.

Apple does ads as well, it just keeps all metadata to themselves.

Yes, but it's not a meaningful part of their revenue unlike Google where it's' their entire revenue.

They are very different companies in structure and it certainly is a "pick your poison" but it's completely stupid to act like they're the same on this front. Apple is better on user privacy

...unless you care about state actors, which you should, in which case your data is the US government's either way.

I humbly submit '1ndex' and 'ind0x', or '1dx' and 'i0x'.

While most people ran screaming from programming assembly aero142 thought it was a rather swell idea.

Frankly it's no worse than "mebibyte".

Pronounced "one-dex" and "in-dox"?