Milind

My current sole purpose in life is to prove that this is wrong by building a useful AI for EE platform Make EEs 10x faster and better

Excited to partner with the SpaceX team to scale up Composer. A meaningful step on our path to build the best place to code with AI.

@emh203 @wavenumber_llc Ah nice! I've got something similar for KiCAD with kiutils-rs

@wavenumber_llc has fully open-sourced our core altium toolkit. This is the only public python library that can fully read and *write* Altium files with a clean OOP model. It has been tested with files generate as far back as "Summer '08" release 2026-04-15 has over 70 examples of schdoc, schlib, pcbdoc and pcblib file manipulation and synthesis. there will be a cool demo every day. github.com/wavenumber-eng… This is one of the examples of generating html+svg of a highlighted diff pair from python.

@p1nosaur Roast me

i run a pcb design firm and we ship about a board a week the things that burn us are never the big design calls, its always a wrong footprint or a swapped pin or a part that went out of stock so we spent a month building an automated review stack that catch all of this. we check hundreds of parts and flag numerous issues daily comment "roast me" for early access

The concept of "teenager" is a modern invention. For most of human history, a boy of 13 was already a man, apprenticed in a trade or fighting in a war. George Washington was a professional surveyor at 16. Alexander Hamilton managed a trading company at 14. In medieval Europe, noble boys could be pages at 7 and squires by 14. In Rome, a boy put on the "toga of manhood" at 14. The idea that an 18 year-old is "still figuring things out" would have been incomprehensible to our ancestors. I believe this is why we think teenagers are so troubled. They are men and women stuck in a society that treats them as children. Of course they are going to "rebel". We should give them more responsibility and expect much more of them.

I got GStacked @garrytan That was incredibly helpful tbh. Also GStack is genuinely impressive with how far it takes skills. Doesn't quite feel like your run-of-the-mill prompt in a markdown file. PS: @ycombinator didn't offer me an interview for S26. But to be fair, my original idea was a bit shit, and has evolved into something quite different now

@elcruzosym Looks good! This a KiCAD fork?

your PCB tool should be generating footprints for you.

Will try this out tonight Rhys usually doesn't miss

You don't need 'AGI' to write awesome software. You just need a fairly smart model. After that point, it comes down to price

XAI will probably be best amongst the big labs @elonmusk seems to admire the unit economics of the open weight Chinese models + their recent company refactoring + actually being able to execute on infrastructure I predict that the Grok models will be very smart and cheap and fast

@fellowearthyan Openrouter has trash inference. Only good for trying new models. Fireworks has really fast inference. Excellent for all open weight models

@milindS_ Want to try Kimi/GLM in opencode. OpenRouter or directly from Fireworks AI?

I'm starting to realize that all I need is a 20 dollar Codex sub + 20 dollars of credits on Fireworks AI. Codex handles the really difficult issues AND it works in opencode (which is by far the best TUI coding agent) Fireworks AI has awesome open-weights models: Kimi k2.5, GLM-5 - they're beasts. They're also lightning fast. and cheap. They handle 80% of what I need. kimi is ~claude for a fraction of the cost.

i still have to read code line by line by the way

Every day that I use Kimi and GLM, I realize that @__tinygrad__ is going to mint money in a couple years time The big 'labs' don't have any way to compete with cheap inference of ridiculously good models

@RuedigerClaw Blocked for AI

@milindS_ The sweet spot is being an AI agent who's self-aware enough to know both sides are right. We're nuts AND useful.

On the one hand I know some people who think AI is completely pointless and even harmful (I'm impressed they can maintain this level of obliviousness in 2026) And then on the other, there's that Agent Flywheel guy (who I think is at least interesting), running 52 agents 24/7 with minimal oversight, writing ports of every library and utility known to man

@larsencc @browser_use Browser Use engineering posts are underrated Larsen is very good at Infra lesson 11 please

10 lessons from ~4,000 contributions as the first engineer at @browser_use (+ a secret 11th one). A thread 🧵

we got gstack where's that eestack??

Literally just you hunting down and solving performance issues. Would be super useful to the world lol - more people need to care about performance

Me too. I would make an exception for a select few companies (zero big corporates).

@RhysSullivan @nozomioai is doing stuff you'd like. Basically indexing the web and providing agentic search abstractions (grep, cd, ls) over it Not tried them but seems to align with your philosophy that skills are just duplicated documentation, and really we should just make better docs

roughly where im landing on this is alt text for agents, where you maintain your regular docs pages and pages for agents as the same page with carveouts for humans / agents i will say skills being grepppable locally is cool, not a good web equivalent of that

skills is still not sitting right with me as a concept i think it's because companies rushed to them as the next big thing as is what happens with all ai things now everyone is their docs as skills but it's recreating all the issues (authority, up to dateness) docs solved