Manuel Odendahl

*taps the sign*

or inspiration fuel for my hardware projects

I'm just making 80ies ads for my software now

@Pshichenko Gpt5 is basically programming AGI for me. The limit is whatever I can think of prompting it with, within my knowledge of the “what is realistic”, which is maybe a weird way of putting it? I wouldn’t expect “solve climate change” to work, for example.

Codex is inspiring a new definition of AGI for me. AGI is something that can get the job done within reasonable time or with reasonable accommodation no matter what job. It must not spiral out in confusion as things get incrementally harder.

Philosophical foundations of mathematics x.com/i/broadcasts/1…

Philosophical foundations of mathematics x.com/i/broadcasts/1…

building a screencast software x.com/i/broadcasts/1…

Building a screencast solution x.com/i/broadcasts/1…

I wrote about going from YOLO agent slot machine programming to prompt engineering. Link below...

@boristane I just go to the coffee shop/park with pen and paper (tablet). gogogolems.substack.com/p/slowing-down…

nobody is working towards making planning suck less - I can't be the only one spending my entire day reading and commenting markdown files

here's an interesting idea, within a company / team: - a single person is responsible for all the production code, kind of the way an architect was - everybody else can slop around as much as they want, the key is that at the end comes out a good document / a good feature (could be just a screenshot or literally 2 sentences: "yes this is good"), and the single integration person is responsible for prompting it reliably into the main codebase, instead of reviewing PRs.... that integration human role can be passed around, or some people will find that they really enjoy it (I know I do), and it removes that whole "review 200kLOC of slop": from the get go you know they're destined for the trash can, so there is no animosity or so. The responsibility is to get the "core" of your work across.

@mitsuhiko This is why I like saving the agent output to my tablet and stepping away from the computer. The chat interaction ux pattern just makes me miss things that are evident once I step away from the computer and read the thing with a pen in hand.

There is this moment where after an hour of “discussing” with the clanker you realize that the damn thing started hallucinating and you became dumber in the process too. This is me discussing changes of trait bounds with the clanker and it just does not understand.

Deploying to the cloud with k3s broke websockets so y'all have to wait for the killer coding agent demo, but thought i'd share for posterity, the federated module + cloud deployment was a big ass step from my symlinked local mess.

Of course it comes loaded with apps because why not.

It's finally getting pretty real. go-go-os with a few generated apps running, each app can be a HTTPS backend or a federated module or a generated piece of code. All code with introspection+docs (for agents and humans), all apps run in a quickjs VM with FRP pattern. more pretty below 👇

@ekzhang1 IMO no pull request should come without a full interactive playground demonstrating the feature at hand (and proof that it is exercising the real thing and not just making a pretty mock)

@ekzhang1 I’ve been doing interactive environments (think Bret victor) for most topics I am interested in and for my own agents PRs and it’s a big rabbit hole. Example: #linebreak" target="_blank" rel="nofollow noopener">pretext.yolo.scapegoat.dev/#linebreak

had a random thonk today after scrolling through lots of demos I used to think about coding as “not only can we solve problems, we have the unique power to make digital media. coding lets us make objects of _interaction_ that people can use and enjoy” but with the LLMs getting really good at specialist web development it’s no longer the case, I guess anyone can make interactive artifacts with computers :) — exciting times it does mean I have kind of a missing hole in my life philosophy that I need to think about a bit more though, what do i value most

I wish we talked more about prompt engineering. Start with the desired outcome: a set of tokens that contains say, the entire source code of the thing we are trying to build. Only then start thinking about what the context needs to contain to create said outcome. And where do we get said context from? I wonder if spending 2 years building context windows by hand (“$(cat foobar .c) using this file create a new file that matches the api: $(gen-api tmpl)”) just makes me default to that kind of thinking, and then map it to agent behavior and figuring out which tools to give them.

I wonder when this (here purely an example) is going to happen unprompted and if it will ever make sense to leave human out of the loop in those cases: Models don’t do well with imperative deployment (think using a cli to deploy etc…), do great with declarative / gitops. No agent currently will go “fuck this coolify shit let’s go kubernetes”. But most task I have right now, if I were to give them more autonomy, would require this kind of decision-making. As far as I’m concerned coding AGI is here, and I feel it actually being quite relaxing: alright, now I can ferret out what these decisions are.