marstall

how do you define the specialists? a prompt? what do you mean by using protocols as guardrails? Also, can you give me a TLDR on how these voice interfaces work? What kinds of problems are you solving, and how, in an ELI5 way. I might be a lot dumber than you give me credit for!

cool. i like what you've done here with a virtual panel that can answer questions ... i have similar ideas for kabuki syndrome (currently: https://www.thekabukipapers.org). talk? marstall at gmail.

i've been using Cascade (a third party harness) since the 3 week period in 2023 when it was hot. I think it's called something else now. Devin? Things got confusing there for a second and I stopped paying attention.

Anyway very happy with it, I use it as a plugin to RubyMine and Webstorm.

One of the primary advantages is being able to choose your model - and it often has free deals for newer models that are running promotions. Whenever I switch to Claude Code it seems clunky. Would rather use Claude with Cascade.

is it just for coding? the docs don't mention code, just "agents"

"force phone providers to collect identity information from ordinary people before they can acquire or renew service with a phone carrier."

don't see the harm in this? isn't this already the case for 99.9% of phoneline havers already?

I'm the creator of a newsletter using AI to make the science behind a rare disease, Kabuki Syndrome, more accessible.

In this blog post I walk through the steps I took to translate scientific papers into sparkling prose that maintains tight links between generated sentences and the passages they're based on, and touch on some of the broader issues related to what an editor is in a publication that is AI-generated, and why I think rare disease reporting is a uniquely appropriate and ethical use of AI!

Philip Ball helped me understand this in his book "How Life Works." He makes the point that you can't understand language by understanding letters. Walking up to someone and just saying "L" won't have any effect. And given just an understanding of the alphabet, you would never be able to use that on its own to understand a whole book, you have to move up. The whole has its own kind of existence. I think about how hard it is for one individual to influence a culture. You kind of have to speak and operate at the highest level to change things. I know that's super vague but those are just the terms I've been thinking about it in.

yes it all comes back to iteration, the original "vibe coding". for me, programming has always been about making it up as i go along. like an artist starts with one stroke, i started when i was 10 years old typing '10 print "hello, world" 20 goto 10' and i've never really stopped programming that way 47 years later. For me programming is the same as refactoring, they both happen in a continuous Zone throughout the day. The idea of spending this big period at the beginning Defining the Architecture then letting AI fill in the blanks makes no sense because I only know what the architecture is, what the product is, as part of a process of typing all day for days and weeks and months, that never ends.

i've actually been wanting something like this.

issue seems to be there is way off the site? top story about thinking machines has no way to click through to original link AFAICS

I've been reading "Elusive Cures" by Nicole Rust, about the failure of neuroscience to cure major ailments like schizophrenia and alzheimers, despite decades of work modeling neurons and other brain systems (costing far in excess of $500M).

Here are some thoughts that that book sparks ...

"the whole is other than the sum of its parts" - someone, based on Aristotle. The way Nicole Rust puts is that the whole->part relationship is one way. In other words, you can determine the parts from the whole, but not the whole from the parts. A cell is a complex dynamic system with many overlapping and interacting feedback effects and diverse homeostatic drives. Its state emerges into its own entity that, once formed, bears only a tenuous relationship to its parts.

Understanding of our bodies and minds may be more tractable at more common levels, levels where the life system is at (whole), not where it was (parts) is where I think she's going: language, art, kinship, etc. But I'm not done yet.

> the OS layer will "see" everything and be able to ingest everything you do into a context

that makes sense, i could see it being useful. though it would be a feature easily replicated by Apple or Android if it got some traction.

Maybe this idea of apps being "replaced by agents" is less literal than I'm thinking.

Like with music, maybe it could be an agent that in general helps me be a cooler, more informed music listener by constantly thinking about new ways of accessing myriad sources of info about what's cool, or what I would think would be cool.

here are some iphone apps i use often:

- Spotify

- Le Monde (french news)

- Maps

- Find My

- Amazon

Anyone have a take on how these (maybe not precisely) could be "replaced by an AI agent"?

TNR - DNR

it may be hard to measure, but it's definitely helping everyone who uses it to be more productive in certain aspects of their work. that's clear to me.

but it might be, say, 20% more productive in 20% of your workday, or 1000% more productive 2 days per month (the days when that perfect dream spec hits your desk that you can just paste into claude and get a slick working system back), which works out to just 4% more productive overall, or whatever, which is hard to measure with all the noise.

in the end companies will pay for these tools because their employees will be demanding them, same as they demand other things that make their workday more pleasant - email, coffee, air conditioning, a conveniently located office, etc.

that said, I see the intelligence itself being rapidly commodified/free. the companies that extract rent in the sector will be the ones that effectively bundle and sell corporate-friendly features with the core intelligence - compliance, tracking, productivity, systems integration, authentication, etc. etc. etc. Which is a competency companies like Salesforce, Microsoft, Google already possess, so they are likely to win. Plus a weird Euro variant of course.

I got 20% autistic, 80% Irish?

| iterate on their product faster. i've run into a situation where it really slowed iteration down, because i wasn't able at some point to explain in english what I wanted it to do and had to go into the code, which, lo and behold, I didn't understand. Ended up scrapping everything that had been generated and starting over by hand!

and he wrote them when he was completely blind!

or democracy/civilization attack. DHS/Putin siccing this kind of thing on democratic voter rolls around Nov. elections could be the least of it.

i could see a long tail of impenetrable chaos as private correspondence gets hacked, ppl get divorced, fired, fight back, flood the zone with their own reputationslop so they have a grounds for denial, decide to take it ALL down to distract. recursive waves of tyranny/chaos. this isnt the singularity we were promised!

yes - i've thought that could work. returning to a more protected object oriented programming model (with hard-defined interfaces) could be a way - "make these changes but restrict yourself to this object" etc.

Let's take that vibe coded product and iterate what it gave you 100 times, as you tweak it to fit your vision. When you do that 101st iteration, can you prevent it from breaking something else, or changing it in a way you don't like it?

What if it doesn't understand what you're asking it to do and keeps failing and you have to keep rolling back? Can you understand the 20,000 lines it's generated so you can make the change yourself without tearing your hair out? Can you fix bugs in it that it can't, without starting from zero and having to understand the whole codebase?

with these posts I always wonder, what happens when this code runs into a customer? Or 1000 customers, or a million? All with their own divergent needs year over year.

I have just gotten off 3 years as a developer for that kind of project, and I used the best AI tools diligently every day. It often saved me time. Like from some small drudgery of half day of flailing about in config land. Or it could generate some nice rails controllers and a javascript front end from a well-written spec. writing tests was also a strong suit.

but just as often it failed to understand the depth of the product and its myriad concerns and led me down the garden path, reducing my efficiency.

Aside from that, a large part of my job was the parts that weren't coding - wrestling with specs that were far from ready for primetim, chaotic internal processes, deployment, internal coordination/communcation, talking to customers, etc.

In the end it seemed like it saved me maybe 20% of my time overall. Nothing to sneeze at.

I get that greenfield apps that have no customer contact can be created with a phrase now. That's pretty amazing. But I would love to see Opus 4.6 up against a real beast of a codebase that you're far from a master of.

I take it; still obsessed.

holy crap, that's amazing. i struggle to get my head around the agent hype - but i guess this video helps a little. I have Agent FOMO, I guess.

i wonder if the skills will divide a bit. That there will be those who still program by hand - and this will be a needed skill, though AI will be a part of their daily toolset to a greater or lesser degree.

Then there will be the AI wranglers who act almost like DevOps engineers for the AI - producing software in a different way ...

yet tim cook still feels the need to attend the melania premiere

"barratry" - thanks for that one, too!

I have tried the Cascade toolbar with RubyMine and it works well ...

The Myanmar story was definitely the worst (Mark Z + callow execs being willfully ignorant as Facebook clearly inflamed ethnic cleansing there and caused many deaths).

Later in the book, the China story was a close second. In order to get into China (to "grow") - exec team agreed to host Facebook's servers in China where the government could get access to customer private data, so they could stifle dissent.

Tons of other weird/bad/embarrassing stuff too. The author, a member of the core executive team, was seriously complicit but redeemed herself in my view with this no-holds-barred account of the complete lack of ethics up top.

In general a damning portrait of the executive team as just not giving a shit about anything except for growth and willing to actively participate in dictatorship in order to make it happen.

I am still a WindSurf user. It has the quirk of deciding for itself on any given day whether to use ChatGPT 5.2 or Claude Opus 4.5 in Cascade (its agentic side panel). I've never noticed much of a difference, they are both amazing.

I thought the difference must be in how Claude Code does the agentic stuff - reasoning with itself, looping until it finds an answer, etc. - but I have spent a fair amount of time with Claude Code now and found that agentic experience to be about the same between Cascade and Claude Code.

What am i missing? (serious question, i do have Claude Code FOMO like the OP)

are you talking about the hashes (##, ###) etc in the subheadings? I think that's an intentional design thing, a bit of a nod to the back row, if you will.