you start with claude code or codex and it's cute, but then you realize - hmm configuration is cheap, the AI can do it!
then you start looking into MCPs and skills, fuck it, oh-my-pi looks awesome!
wait a second? I can just have AI make my own personal AI harness! Next thing you know, you're writing the 5th version of "little-coder" or similar using the Pi library
ahh shit, you just read an article that `tools` are actually crazy important for AIs, using `sed` is dumb when `hashline` + ASTs are way better, lets just start writing our own tools!!
...anyway I just use Zed, simple agent on the left, code on the right
i have some pretty complicated automated workflows that use `linear` + a orchestrator -> implementer -> reviewer -> releaser workflow, but it's less a dev stack and an AI factory
seems more like a culture problem, i have my calendar very public, all my junior devs know ill get on a zoom with no hesitation and they actually seem to enjoy the screen sharing, every zoom is recorded with AI summary/transcript so they’re more focused on asking questions instead of taking notes (and i think they’re really solid juniors and actually go back and watch)
there’s the whiteboard element but i’ve gotten pretty good at exalidraw and zoom annotating
add in the remote makes it kinda easy to not be distracting in meetings so i can easily DM them context on the side to get them ramped up easier
Tossing in my two cents here to agree with you. I worked remotely on and off from about 2014 onward until post-COVID RTO brought me into an office for 18 months before I became remote again. During that time (and across a bunch of companies) I went from desktop support to senior sysadmin to security on the cusp of senior security engineer.
In my experience the biggest factor in teams usually came down to the middle management layer. If their "style" was "watch over your shoulder, butts in seats" type of micromanagement then juniors didn't tend to progress unless they were self motivated to seek it out.
I'm sure this is quite a personal thing. I much prefer being in-person for that kind of interaction, and I don't think it's about efficiency as such, I just prefer being around people despite not being an extrovert - hybrid working is perfect for me.
Not quite as I understand it. The ternary approach bonsai uses leverages a FP16 scaling factor that each value in the ternary maps to. You're still using 16 bit multiplication, it's just that the weights are far more compressed.
fair, i think i was referring more to 1.58 bit architecture in general since the original paper (Figure 3) shows that we eliminate FP16 multiplication and addition just for INT8 addition. I need to dive deeper into bonsai overall if it differs
Zig's been around for ~10 years. It's more low-level and lightweight than Rust. Different goals, different trade-offs. If Rust is the new C++, Zig is the new C.
"technically" usually means something like "strictly", not "by a completely different metric". work takes time. zig has had a decade of work put into it.
Technically means according to a strict, often legal definition.
The strict definition being we don't count developments that happened before version 1.
Like when we talk about Rust, we don't mention the virtual threads or GC or the @ symbol for GC references. Even though those all happened during its development.
and when people talk about zig, they don't usually mention that zig used to have goto, casting syntax like `T(val)`, a rule that said you couldn't pass containers by value, language-level async, some truly awful syntax for what is now `try` and other operators, etc. both languages took time and work to realize that these features were not for them. very strange to deny that.
also, nitpick: they said zig has been around for ten years. this is, strictly, correct. the zig project has existed for ten years, just like how rust has existed for about 20, now. a project still exists if it is pre-1.0. nobody was talking about versions before you.
Deepseek proper is the only provider offering Deepseek V4 Pro at such an insanely low price. All the other providers are like 5x.
If someone says they're using Deepseek, it's probably a safe bet that they're sending money to China. People choose an inferior model (Deepseek) because it's cheaper. Not many people are willing to use Deepseek but also pay the 5x price to have an American company run it for them.
> Update: Since these measurements were taken, Claude Code has rolled out Tool Search with Deferred Loading, which loads MCP tool schemas on-demand and reduces context usage by 85%+. The context bloat described in Problem 1 is largely addressed for users on current Claude Code versions. The performance, debugging, and architectural arguments below still apply.
reply