This is a fascinating place in the stack to sit. I need to dig in a lot more. Nice start.
Good luck with the showing!
1,128 karma · joined October 31, 2012
I'm currently working on dev tools for agentic engineering and software-first simulated hardware design. Writing and researching as a means of growing.
Projects:
- Intraview (intraview.ai) give Agents the ability to co-create code tours in VS Code to speed up onboarding and alignment on code-bases
- Unfudged (unfudged.io) is source control without the commits so Agents (and you) can recover from mistakes.
- Writing / research (cyrusradfar.com) is where I do market research, guesstimating the future and keep searching the glorious unknown canvas of knowledge.
Find all my projects, writing and contact methods at https://cyrusradfar.com
Good luck with the showing!
I do think security is going to require more, not, less human investment as attackers may be running automated vulnerability screens from the outside that you must counter, as well. Without rigorous internal processes to manage and screen all changes and upgrades, companies risk leaving themselves open.
One design change which limits exposure is to have more local-first apps or experiences so there's less cloud / server to computer interactions to secure.
The argument I was trying to make is that we (in the US) have an economic and political system where decisions are made by the person/party with the most economic power. I shorthand by saying it's "by auction."
When you do that at a system-level you have a problem where the decision maker is making trade-off decisions for the system based on their best interests vs the system.
This problem happens in product/software companies at a different scale where orgs have different power and may influence a product in a way that fails the users and eventually isn't in best interests of the system.
I have been curious about the “what must be true” for AI data centers to be brought online. I wanted a systemic review to understand the technical supply chain to deliver modern inference.
What I found is the AI buildout is being slowed by people declining the offer, some on their porches, some on planning commissions, some at the trading floor of a capacity auction. There's also a few single points of failure in the supply chain.
The AI buildout, as I learned, is a bigger test of our entire economic, political, and societal system.
Hope you enjoy it.
After a trip to a textile museum, I had to research once I saw the obvious connection. I wrote this for anyone who loves tracking the evolution of ideas vs acting like everything is new and invented from nothing.
Looking forward to discussing and geeking out on computing history.
the article doesn't argue for a capitalist cohort to support any other. Rather it asks for limits to the capitalists reach, such that other systems can use resources in other ways and preserve non cash systems.
I think the insight for me in researching all this was just popping my bubble that thought of capitalism and what's wrong with it through the lens I was taught which was mainly political.
Once I reframed through the "right to not compete" or not participate in capitalism and retain dignity, the system and how it's taking us further from that ability became clear.
it’s not discussed in this post but in another right after I discuss the modeling I was doing on tech debt and finding the game to improve agent outcomes was reducing context.
functional programming accomplishes that. I can’t claim it’s the only way, but it’s one that’s well understood in the community
I've explored Clojure after talking to Metabase about how it had benefited them. That said, it was years ago so I can't claim it influenced this work.
The framework was designed to be a language agnostic way of sharing best practices to bias agent behavior towards a more scalable end. I initially used it when I was working with a team to do some massive refactoring/clean up across the codebase. We didn't come to an acronym but similar principles and it was "testable" and easy to push back on PRs that weren't aligned with the principles.
That said, it may be interesting to see if I could replace all that context and just say -- "code it like you would with Clojure"
Have you tried that?
I linked to the snap from archive.org and the 404'ing link is https://www.ycombinator.com/companies/delve
A port to Swift has been released April 1 by Tornike Gomareli (https://x.com/tornikegomareli) the community so now it's able to be used in Mac Native apps
- tweet: https://x.com/tornikegomareli/status/2039516374788124865
- repo: https://github.com/tornikegomareli/swift-pretextkitThat said, I don't "vibe" because it creates great code I love reading, but I can monitor and move the same metrics I would if I was managing a team.
I also use code tours a bit, and one of my first tools I needed and built (intraview.ai) was to support this need to get deep in the code the Agents were claiming was ready to ship.
It's a juicy story to talk about that hits a lot of checkboxes that make it viral --
1. the hustle culture they promoted online was gross
2. they followed the 30u30 Forbes pattern like Liz Holmes, FTX, etc.
3. they're a YC co, so their's plenty of popular voices supporting them
Finally, the security and compliance community is litigious by their nature and this startup, in general, was a net negative for a lot of people who do fractional / consulting work in security.
Four "labs" shipped self-optimizing AI, two in the last month. MiniMax ran 100+ rounds of autonomous scaffold optimization (30% gain). Karpathy's autoresearch (630 LOC) ran 700 experiments in two days. AlphaEvolve beat Strassen's 1969 matrix multiplication record. Microsoft's STOP did it academically in 2023. None retrained weights. They all optimize the agent layer: prompts, tools, sampling params, workflow logic. Schmidhuber's Gödel Machine needed formal proofs before self-modification. What worked was brute force with a 5-minute timeout.
Open question: what happens when the optimizer can also rewrite its own eval function?
Allows Claude to create code tours I can navigate with the Agent, give feedback, and iterate together. It's a nice inner-loop step to internalize what's changing and why.
I'm experimenting with a UX that I've been optimizing it for "speed reading" any online article. Initially, I made it for my 8 yo son, to see whether he found the experience more or less compelling to reading off a page.
Good news is he enjoyed it and was comprehending as well or better than reading it the conventional way. I paused from time to time to ask questions to confirm comprehension and he was actually absorbing material faster than his school thought he could read.
That led me to try locally, an LLM that'll generate questions and I'm considering a little App that pauses every N seconds with multiple choice questions to keep engagement and comprehension to make it a sort of reading game.
You can play with custom markdown here: https://cyrusradfar.com/labs/adhd-reader
There's a bookmarklet you can add to chrome/safari/etc., if you'd like to play with it across the web.
Feedback and ideas for directions to take this experiment are welcome.
Anthropic were very vocal, well before this happened, that they were against the use case.
I don't blame them. These use cases are like blaming MySQL for storing the lat/long of the school. AI can't be held accountable and the company was trying to protect us and, yes, it was too late.
tl;dr
App that only lets red light through the monitor. On Mac, has no CPU impact as it runs through display controller. Slightly different mechanisms on win/linux.
Free to install and test it out if you like it. Nothing remote in it including no telemetry. No accounts needed, totally free.
Tell me what you think (especially if it's late for you).
Screenshot of what my scren looks like:
https://x.com/cyrusradfar/status/2029478321277551005
or