joshheitzman

One thing is written directly into the constitution the other is not. How are they similar?

Guess who decides whether its commerce or gambling?

I said 'commerce' not 'finance' because finance is just one of many, many things that is within the purview of federal law due the commerce clause.

> And taking on the feds too, that is a gutsy move.

More likely a wasteful move. Federal preemption is well established.

One of the things that comes with experience is more things moving from your personal unknown unknowns bucket to your personal known knowns and known unknowns buckets.

The substance is hard to see through the conflation of CS and software engineering. Do you actually believe that all there is to software engineering is CS fundamentals?

It's notable that you said "out code" not "out engineer".

Steam can absolutely be replaced. There's lots that could be improved in the process of sifting through tens of thousands of games to find the ones I want to play. I have Epic's store installed on my gaming PC as I still play Fortnite. It has none of the things I want improved over the Steam experience. Literally none.

Put up or shut up Epic. I'm unemployed so you can hire me to do it for you, but I'll bet you'd rather just whine about it.

From the comments here I can infer that it has a schema and in turn I can extrapolate that this is as much "for humans" as many other config formats that are deterministically parseable.

A GUI is a config format for humans as it can you show what all of the configuration options are along with text that tells you what they do. A GUI can also ensure that there are no invalid settings that are discovered separately from the edit session (sure it may not do so, but it has the possibility unlike whatever random text editor the user opens the config in).

This assumes people don't increasingly let AI make decisions for them.

They've been offshoring as much as they can for decades. The H1-B program hasn't stopped it and if it were eliminated there wouldn't be a huge surge in offshoring. Many companies have already hit the limit on offshoring and had to pull things back onshore.

Speak for yourself. I'm reading these comments and not the blog.

You don't need a full browser or a full HTML implementation. You just need enough HTML support for markdown converted to HTML and wxWidgets has that level of support for HTML backed in. I've never seen an LLM generate HTML without it being explicitly told to do so, so I don't know where this notion comes from that full HTML/CSS/JS support is needed to render LLM text outputs.

I used the wxWidgets C++ GUI for my coding agent ( https://github.com/fluxmoot/fluximp ) not directly in C++ but in Python via wxPython. While wxWidgets doesn't have a full browser level implementation of HTML/CSS/JS it has enough HTML support for rendering markdown converted to HTML just fine. I'd imagine Qt would have no problem with markdown either.

> I actively use all harnesses

Several perhaps, but you definitely aren't using all coding harnesses actively.

Man-month or man-minute doesn't make any difference.

Can you add 27 more coders and cut the app build period down to 25% of what it was?

9 isn't a large number of people and its fairly well agreed on maximum limit for productive team size (the limit can be lower than 9) due to the number of possible communication channels. All that has changed is that a 9 person team can now do more with AI assistance then they could without it.

Not having to build still saves wall clock time. Also Python is a concise language (fewer tokens in context) with a monster ecosystem that avoids a lot of code having to written by agents (fewer tokens for the current task and fewer for future tasks by keeping the code base smaller).

As a counter-example, I use Deepseek v4.1 Flash to make progress on multiple projects (Python, C++, CUDA) every single day.

It is. Harnesses do not need to give LLMs such broad tools like the shell.

Let's bump that sample size to 2. I've observed the same and that observation has heavily influenced the design of my coding agent harness.

> No wonder Rust is becoming the default language for high-performance LLM-written code.

What's the evidence for this claim?

The implementation of the tools don't execute processes (i.e. the implementations aren't wrappers for shell commands).

I use that specific model as my primary LLM for coding and diagnostics on projects in Python, C++, CUDA, and for web research.

I'm also using it with my own harness (you mentioned "in my harness"), so I'll share two things I found about using open weight model providers. The first is that getting the configuration right isn't a per model thing, its a per model per provider per point in time thing. The point in time thing comes from middleware updates as those introduce or fix defects that can very seriously impact inference quality. The second thing is inference quality varies a lot between providers, even when they are ostensibly using the same quant of the same model.

If you want to try it something else for comparison I open sourced my harness a couple of weeks ago: https://fluximp.org

I recommend trying deepseek v4.1 flash from together.ai (not the 1017 one the one released last month). deepinfra.com is cheaper for that but its noticeably slower and the inference from together.ai feels like it might be better (not quantified; just vibes).

Yeah, I don't get either. I look at source control as the first line of defense for when agents screw up to badly. I'm not going let the agent mess with the easiest way to undo its screw ups.

From the docs it seems like that would just silently not install the blocked package, so while it would protect my dev environments regular pip installs would still pick up the transitive dependency.

> specific situations in which Claude Code will embed a user's GitHub CLI OAuth token into a response body, and in situations where response bodies are logged, these tokens may appear in LiteLLM logs.

That's pretty hand wavy. What did y'all have CC doing that it had your GitHub CLI OAuth token in its context at all?

Stuff like this is why my coding agent is a coding agent not shell agent that basically own the machine. Mine has no tool for shell use narrow tools specific things like ripgrep, finding files, and deleting files that limit the paths to explicit workspace directory.

I take care of commits, push/pull, building wheels, and publishing wheels myself. It really doesn't take the much time and the coding agent frees up a lot of my time.

Seems like pip doesn't have a way to exclude packages although it was requested at least once: https://github.com/pypa/pip/issues/10797

Just checked if litellm crept in with any packages I installed recently. Fortunately it didn't, but I'm working on an OSS AI agent harness so its certainly a concern and I'd like just flatout block litellm so I don't have to worry about picking it up as transitive dependency. Guess I'll have to vibe a script and remember to run it. Sigh.

Show the refund request rate.

By running my own agent harness that provides detailed API metrics in the GUI and logs very verbosely.

> Why these are not becoming standard or even required practices is beyond me.

It sounds like you haven't actually tried to do what your proposing then.

> tries to use the software using desktop control

This is not a solved problem yet. I've tried a number of solutions and none of them were actually good enough for what you're proposing. Don't get me wrong, I'm not saying its an unsolvable problem. Only that it isn't sufficiently solved yet.

I call it "like software rendering". Its certainly software rendering in spirit since no fixed function rendering hardware is being used.