Mason Stallmo

Mason Stallmo

@mstallmo.com

Something between Midwest Emo and Granola hippie Software Engineer | Photographer | Mountain Biker Rust 🦀 | AI | Distributed Systems Tech Blog: https://kernel-panic.leaflet.pub Personal Site: https://mstallmo.com GitHub: https://github.com/mstallmo

230 followers549 following641 posts

Just finished reading Soul of a New Machine by Tracy Kidder and there so much good stuff in there for the modern software engineering landscape. What is old is new again.

The book is excellent on the whole but what stood out to me most was in the chapter that talked about how people viewed computers in those days (late 70s - early 80s). People back then were saying literally the same things about computers of the day that are being said about AI now.

This both reinforces the idea that AI is merely the continuation of the information revolution and that the future with AI is far far less bleak than some people predict.

Today I had to say goodbye to my beloved Corgi Gracie. You were the beautiful, sweetest, best dog I could ever be so blessed to have. I hope the other side of the rainbow bridge is everything you could ever want. Love you forever ❤️

This is the dumbest thing in sports. None of these review committees ever have any sort of power to do anything. Someone at the NCAA needs to step up and do the obviously right thing.

Chris Vannini@chrisvannini.com

New: FBS oversight committee says it cannot overturn the Michigan-WMU ending. "The committee agreed that the instant replay protocols were not followed in the contest’s second-to-last play. Had the protocols been followed, the outcome of the game would have been different."

I have been saying for a while that AI needs to transition towards becoming an implementation detail. A product shouldn’t be interesting because it uses AI but it should do something awesome that would be impossible without it.

This is really really cool to see. I've been hoping for a long time to see Rust gain more of a foothold in how AI is built and this is a huge step forward.

The Rust Foundation@rustfoundation.org

NVIDIA, our newest Platinum Member, launched CUDA Rust today. You can now write GPU kernels in Rust and compile them straight to PTX. Learn more on their blog: developer.nvidia.com/blog/introdu...

People claiming DHH is “reinventing the personal computer” are straight up delusional. The guy reskinned Arch and installed some default apps. Call me when he’s written a new OS kernel that people actually use.

I have been using GPT-5.6 Sol during my Delta beta testing and I have to say I'm liking it a lot better than Opus or really any of the Anthropic models right now. It's not that their more capable but the speaking style is just so grading. It's really refreshing to have a normal sounding agent.

Got a chance this weekend to kick the tires on Delta from @zed.dev and I’m really liking it so far. There are some really interesting ideas here that I’m excited to see where they go!

Reminds me of debugging poorly performing code. Sometimes you find code that could be faster, you make it significantly faster, and the program still basically performs the same. Sometimes the thing you speed up was not the main factor of slowness in the large.

This post is unavailable.

Not sure if it’s due to repo specifics (code size, language, instructions, etc) but I’ve been noticing a huge difference in opus at home vs opus at work. Opus on my personal projects is much better to work with than opus at work.

I’m a simple man with simple needs. I just want one of these and an Oxide rack to deploy software on

Amos Wenger@fasterthanli.me

the cs-4 looks incredibly cool AND I lack the capabilities to predict what the world looks like with TPS numbers in the thousands for models like Sol-5.6. but I'm not sure I'm optimistic. www.cerebras.ai/cs4

Mason Stallmo reposted

Replying to @jackcarterbenjamin.bsky.social

One of my long held convictions is most people don't care about concepts like "openness" or "decentralization" or "interoperability" (people are busy!) but they do care about what those enable. It's the job of the much smaller number of people who do care to build the experience people will love.

This talk is going to be awesome, so sad that I won’t be able to attend this year 😭

Ada 🏳️‍⚧️@bilberry.games

I'll be speaking at #rustconf26I in three weeks! If you want to know what it takes to build a ONNX runtime from scratch, how to get the GPU to perform math from Rust, and if you should use cubecl, rust-gpu, or wgpu, be sure to tune in! Grab your seat TODAY: bit.ly/4wwduG6 #rustconf #rustlang

Patiently waiting on the private beta list. Really excited to try this out!

Zed@zed.dev

Today we’re introducing Delta. Delta is a multiplayer environment for coding with agents and reviewing what they build, powered by DeltaDB. The first invites to our private beta are going out now. zed.dev/blog/introducing-delta

Mason Stallmo reposted

This has been a delight to see come together. The joint keanu_woah.gif moments Annie and I have had learning about this space have been awesome. You’re gonna love it!

ngrok@ngrok.com

Compression and LLMs are trying to solve the exact same problem: predicting what comes next. @anniesexton.com's interactive essay walks us through the basics of compression to reveal its surprising overlap with every language model you've ever used.

AI or not the next big product move is giving more and more users tools that look like the tools we use as engineers. We are absolutely saving the best stuff for ourselves.

Ethan Mollick@emollick.bsky.social

A failure of ChatGPT Work & Claude Cowork is they assume that non-coders couldn't understand how to think about problems like a coder, so they hide all that stuff. They should instead explain choices like a good PM would (what should be delegated? what should be generalized? what tests to run?)

This is great and expresses a viewpoint that I have been woefully unable to put into words.

eliza🌻@elizas.website

with times being what they are, it seems like you basically can't have a public open source repo without receiving LLM-written PRs. thus, after a great deal of thought, i have adopted the somewhat idiosyncratic Mycelium AI Policy in my personal projects: github.com/hawkw/myceli...

Models are going to end up priced as a commodity (we’re already seeing this happen). It will be profitable but not have the large margins that are historically true for software. From a financial perspective models look a lot more like a hardware product than a software product. 1/

Ted Underwood@tedunderwood.com

Alternate hypothesis: Kimi K3 and Qwen 3.8 show that leading in model releases does not esp. matter commercially. MSFT and GOOG may well have figured out that what will matter is effective agentic integration with the vast user data in their core office suites. That’s hard but worth getting right.

View full thread

Compare this to the energy (oil) industry and there are a lot of similarities. A lot of capex is required to drill and refine oil. There is minimal differentiation in the final product (people don’t care who produces the gas they put in their cars) 3/

If you’re looking for traditional software margins the value is in the software that integrates AI into products and workflows. This is where products will be stickier and the capex will be in line with capex for traditional software products. This might be MSFT and Google’s play. 4/4

Models are going to end up priced as a commodity (we’re already seeing this happen). It will be profitable but not have the large margins that are historically true for software. From a financial perspective models look a lot more like a hardware product than a software product. 1/

Ted Underwood@tedunderwood.com

Alternate hypothesis: Kimi K3 and Qwen 3.8 show that leading in model releases does not esp. matter commercially. MSFT and GOOG may well have figured out that what will matter is effective agentic integration with the vast user data in their core office suites. That’s hard but worth getting right.

There are a two big factors at play here. 1) model output (tokens) are fungible. 2) Training and operating models are VERY hardware intensive. There is a large capex required to get into model development and serving. 2/

Really interesting thread. Going to elementary school in the 90s and early 00s I absolutely got dedicated typing classes. I was and am a better typer than either of my parents. I remember a lot of talk about the generation behind me being “digital native” and I suspect that is driving a lot of this

Jess Calarco@jessicacalarco.com

I wonder if students today would be less likely to lean on AI for writing assignments if schools had made sure they could touch-type (at least 60wpm) before having them submit all their assignments online. 1/🧵

I historically have been pretty unbothered by how AI talks but man Opus 5 has been driving me nuts lately. Hard to tell if it’s the model directly or just a bit of AI fatigue but it’s getting annoying

This is very much in the same vein as people who don’t understand what the first amendment says either. So many people get up in arms about what they “feel” the law says and not what it actually says.

ghost malone ergo propter bones@proptermalone.bsky.social

I think intellectual property is the dark mirror of immigration law, in that people’s intuitive sense of what existing IP law allows is far more restrictive than the reality while their intuitive sense of what existing immigration law allows is far more permissive than the reality

Anthropic shot up on the good will of telling the DoD to kick rocks and can easily lose that good will over open models. Tokens are a commodity and the competition is getting tighter.

Might just get some Taco Bell tonight in solidarity

public health guy@publichealthguy1.bsky.social

it does actually sound like taylor farms and the white house teamed up to try and blame taco bell for the outbreak. unprecedented corruption

I’m begging people to bust out an inflation calculator when talking about airline ticket prices in the past. $500 in the 70s is not $500 today. You can in fact still get the experience you long for from the 70s in first class for about the same real cost if not less.

This is why corporations themselves aren’t the problem but de-regulation and weakening government is the problem. I don’t want corporations to be more community minded. That’s what government is for. I want there to be a strong opposing force to the profit model that underpins corporations.

This post is unavailable.

The hug difference in response to AI from Linus and Daniel Stenburg vs the Zig guy is pretty telling. The creators of two of the most important software projects ever made recognize the usefulness of LLMs as their capabilities have evolved. Others have decided to stick their head in the sand.

IMO Zig should not be taken seriously or used for anything serious. It’s clearly viewed by its own creator as a hobby project, which is totally great for them. It does make it irresponsible to use in production though.

The only written data center critique I will accept must be printed on dead trees. Any critique in digital format is wholly unserious

PSA: No one has taken away your text editor or your compiler. If you don't want to use LLMs in your work you don't have to. Please spare the rest of us from your boring attempts at engagement farming. Thank you

Really excited to see what the PIT crew thing does under Mamdani. I've thought about the potential positive impact government could have via tech for a long time. There is an opportunity here to demonstrate what something like that could look like. If anyone is going to pull it off it's Mamdani.