We’re entering an era where producing software is becoming dramatically cheaper.
But producing more code has never guaranteed producing better systems.
The interesting question then isn’t simply how much programming AI can automate.
It’s what happens to software engineering when programming itself becomes less of the bottleneck.
Here I argue that software engineering has always been, fundamentally, a study of language. So I wrote a meditation on how we arrived at the concepts behind our most influential programming paradigms, and how their histories extend far beyond computers.
Because in a future where language can be generated and iterated upon at extraordinary speed, the remedy for inscrutable code and documents isn’t necessarily more code and documents.
Perhaps it’s to examine ourselves, to treat our cultural roots as dependencies, and to decide which ones we want to keep.
Computer Science is a Social Science
sadasant.ai/writing/computer…
𝙳☈ retweeted
most AGI pilled labs:
- typesafe
- general reasoning
- general intuition
- mit oasys
- amp code
this but just McDonalds, 7-Elevens, plastic bottles, generic sport t-shirts, asphalt, concrete and glass, everywhere
(I’m kidding)
Hey kiddo, don’t worry, nowhere is dramatically different from home. Always look for those 7-Elevens.
1
114
❤️
Replying to @tszzl
in particular, i am very proud of Daybreak Blue and the impact its had for defenders, which while it earns huge ire on here all the time for the bare minimum KYC requirements, is a victory born of a very specific technological and political posture at openai
3
210
Btw, many thanks to @UnderdogAI for underdog.ai/woof , it is incredible.
We’re entering an era where producing software is becoming dramatically cheaper.
But producing more code has never guaranteed producing better systems.
The interesting question then isn’t simply how much programming AI can automate.
It’s what happens to software engineering when programming itself becomes less of the bottleneck.
Here I argue that software engineering has always been, fundamentally, a study of language. So I wrote a meditation on how we arrived at the concepts behind our most influential programming paradigms, and how their histories extend far beyond computers.
Because in a future where language can be generated and iterated upon at extraordinary speed, the remedy for inscrutable code and documents isn’t necessarily more code and documents.
Perhaps it’s to examine ourselves, to treat our cultural roots as dependencies, and to decide which ones we want to keep.
Computer Science is a Social Science
sadasant.ai/writing/computer…
1
111
I might agree with you, @thesimaximalist , but a significant caveat is that I believe a precondition for high [local / individual] agency (we might call it just “trust”) is broad alignment on the underpinnings of groups of people: language, culture, priorities, preferences.
Nowadays, if an organization isn’t diligent ontologically and epistemologically, letting people and systems loose seems equivalent to accumulating timber and waiting for a spark to light it all on a beautiful and lush fire. A chain reaction, where every disagreement compounds until the entire system halts.
Loosely speaking here, but organizations have so far ostensibly run on vagueness, top down, in my opinion to maximize the degrees of freedom of the leadership. Some companies run on a few statements and some minimum policy, where the statements are catered to communicate updates briefly, let’s say on a PowerPoint presentation, to an audience that can’t generally invest itself beyond a few hours every quarter, let’s say, a board. And the policy documents have been focused almost to become exculpatory, to limit liability.
This limited depth on top-down structure works if the rate of errors is far below the cost of correcting them.
Let’s look at an everyday analogy: vehicles of any kind. If the vehicle is slow, you can get very creative on how you design it. Parades run on vehicles that are only meant for display- they are usually sluggish and usually have tons of stuff on top that work against the functioning of the vehicle. Best case scenario they are selected to carry a heavy load, but whatever cheap thing does the job is enough.
But if you want speed, then you are forced to consider environmental pressures that need preparation and many trials until the rate of error is reliably minimized. Otherwise you’ll hit a wall fast.
Now that code can be produced at inhuman speeds, why aren’t we taking proportionately that much premeditation before we trigger on that speed?
My bet is that we haven’t recognized the damages.
I’d expect more value in designing systems people can change independently. Generating a feature in minutes helps a lot more when shipping it doesn’t require synchronized changes across five services and three teams
1
217
Ok, so, for those of you who pay attention (thank you so much, sincerely! I mean it, thank you). Here’s what I’m doing.
I have plenty of material from past blogs, notes, documents, recordings, and more material coming (yes, including WSP and some bold claims about the future of software). Things that I want to put out there in the form of essays, from my experiences and interests. Collections of things that I believe are useful for others.
Over the past few years, I had published on Medium thinking that I wanted the attention of the executive and high managerial class. I wanted to understand what they cared about and how to write to them, while I fundamentally did not know how to write essays in English. So I used plenty of AI. I mean, I started vibecoding entire projects in 2024, why not also do this for writing. The point is, I have learned a lot, of what works, and fundamentally what not to do with AI.
I guess, for broader context, I come from Venezuela, and although I have an engineering degree, the depth and milieu of that environment mainly targeted local industry. Those conditions were against the things I desired. I wanted pursuits that would not flourish there. So, I eventually moved and made a life in the US. I managed to do this by being a good enough programmer and caring enough for the people around me. I gave talks at some point, I joined big tech at some point, I ascended through the ranks at some point, etc. But what I did not do was to invest time writing.
Moreover, I had grown awkwardly enough in my mind that if I didn’t learn how to write long-form now, I would be left to dig a deeper hole for myself. I would go insane, feeling nobody would understand me.
One critical lesson for me is that human-to-human communication should not be mediated with AI.
Some people reject AI for writing out of principle, and call it disrespectful, and they have a point. I wouldn’t personally feel it was disrespectful if the purpose of writing was to produce an output. Let’s say, just because a mechanical piece is printed by a machine and not labored by hand, why should I care? — But, later I realized that the purpose of communication isn’t to produce an output at all. That level of investment is limited to consumerism, to generic things that people seek for the purpose of leisure time, analogous to TikTok, and sure leisure time is valuable, but it is fundamentally non-conducive, it is transactional and, I figured, the second order effects, the conversation through time, the cultural impact of any single output, die rather quickly, creating the conditions for high churn, and on the authors a pressure for repetition.
I’m not writing here to present myself an expert on the subject of writing. You can call this just the bullshit that I believe and move on.
I figured that what makes a writing useful for the purpose of communication isn’t merely the ideas presented in which order, but something more succinct, like the uniqueness of the texture — of the voice of the writer, like a fingerprint we can smell subconsciously, but one which tells us the other person is alive and breathing at the time they wrote that piece.
I struggled to find that voice, but I believe that I finally did, hence why I made a new blog where I’ll publish new articles and republish old ones.
Each paragraph in these articles are written by my hand, and they should feel as such.
The images, the code of the blog, and a few other things will be vibecoded, primarily because I believe I should optimize my time for the actual content.
It isn’t meant to be perfect. It isn’t meant to pass whatever bar of writing you believe an article must pass. I set up my own goals. There isn’t a way to comment on the article directly. The code of the blog isn’t opensource. None of that is the point.
In my mind, I did what I had to do to get here, and I arrived at a valuable destination. Whether that is true or not, I leave it to the odds.
Cheers
We’re entering an era where producing software is becoming dramatically cheaper.
But producing more code has never guaranteed producing better systems.
The interesting question then isn’t simply how much programming AI can automate.
It’s what happens to software engineering when programming itself becomes less of the bottleneck.
Here I argue that software engineering has always been, fundamentally, a study of language. So I wrote a meditation on how we arrived at the concepts behind our most influential programming paradigms, and how their histories extend far beyond computers.
Because in a future where language can be generated and iterated upon at extraordinary speed, the remedy for inscrutable code and documents isn’t necessarily more code and documents.
Perhaps it’s to examine ourselves, to treat our cultural roots as dependencies, and to decide which ones we want to keep.
Computer Science is a Social Science
sadasant.ai/writing/computer…
7
343
Something tells me that the key to get there, the key to rediscover software engineering after neural networks (after they take over the world, in one way or another), is to recognize that everything, every subject, verb, object, property, parameter and function, is only local.
1
189
𝙳☈ retweeted
Reprogramming dormant neurons : Qwen TTS
Ever wondered why Qwen3-TTS sometimes breathes or laughs out of nowhere? Agent found neurons behind it.
The details:
Agent found 417 neurons to be specific (about 0.48% of the talker's MLP neurons) and 61 of them are dormant. Amplified together, they give the following effects effect:
1. Breathes as words are spoken
2. Hushed ASMR
3. Natural laughs as language is being spoken
4. Enhanced whispering
5. 2nd speaker appearing.
Comparisons and neuron map repo in the comments below.
Inspired by recent work on Lazarus by @DimitrisPapail , which reprogrammed dormant neurons.
𝙳☈ retweeted
The stupidest thing that's ever happened is that every verifiable domain is going to be hillclimbed by guys with thousands of GPUs. There's no trick to it. There's no new discovery. Just bigger layers and more data. A little spitshine to make it nice. That's it, and all verifiable domains are vacuumed up. This will be, for many people with more nobility than you or I could ever dream of attaining, tragic, but perhaps this is a tragedy we accept as the cost of "progress." But what we shouldn't accept is this idiotic ever-more-passive slamming of more and more data through more and more GPUs creating a permanent socioeconomical divide. That cannot be allowed to occur.

















