AI can already create binaries directly — by invoking the compiler the same way a programmer does. I don't think it makes sense to go further than that, and have AI generate executables without using a compiler, for the same reason they can't multiply large numbers without tools;

"Artificial intelligence is what it says it is — artificial. There is no ghost in the machine; there is only a machine in the machine."

I'm convinced that if you're building a large piece of commercial software, the only good way to use AI coding agents is as a power tool, without forsaking your ability to understand the code. On the other hand, if you're building disposable software — for personal use, for a

"AI systems amplify well-structured knowledge while punishing undocumented systems." 🎯 This sums up well how software development is changing, I think. Well-organized projects with clear documentation and legible code will have a big advantage. It isn't obvious how granular the

Having fun using Codex 5.3 ($20 plan) to improve my C++ Markdown parser. From 277 to 402 passing test cases since yesterday. AI really shines at this kind of work! I'm directing the effort and playing code janitor, but it even helps with those things too. github.com/preshing/plywo…

Git finally has an easy-to-use command line interface: pi -p "Rename remote origin to local"

This is using the minimalist agent harness pi.dev by Mario Zechner by the way. You can even use it "off the grid" by running qwen3-coder-next (or similar) on local hardware. There's no question that the commercial labs have much more capable models however.

So refreshing to hear practical technical discussions about AI amidst all the hyperbole and clickbait that have dominated 𝕏 in recent years. Keep up the good work Scott Hanselman 🌮 Big fan!

Doing some experiments today with Opus 4.6's 1M context window. Trying to push coding sessions deep into what I would consider the 'dumb zone' of SOTA models: >100K tokens. The drop-off in quality is really noticeable. Dumber decisions, worse code, worse instruction-following.

ThePrimeagen The definition of "AGI" is "Whatever I can say to hype the thing".

Coding by hand is still a great way design the API and internal data structures of a new module. It only makes sense to delegate tasks to a coding agent once a certain amount of the module has taken shape.

Interesting way to put it. I'd go even further: Not only is AI fundamentally a tool, but the human intellect is also fundamentally a tool. And the human intellect is what AI attempts to model. So how can an AI ever become more than a tool when the thing it models is also a tool?

I'm still scratching my head at the idea of spending thousands of dollars (or more) on tokens. The best reason I can think of is to run an agent swarm that searches for bugs/optimizations in an existing codebase. But even then, wouldn't such swarms be short-lived? Don't they find

Having done my fair share of cryptic C++ incantation 3am debugging sessions, I can relate. But C++ is really not that bad a language when you keep the more esoteric features to a minimum. Memory safety is completely achievable. High quality C/C++ libraries and development tools

I'm guessing this is the real reason why the latest frontier modes are only being made available to large tech companies.