JamesSwift

Well in a lot of ways this demo actually is a mix of jev for simple stuff + a very intelligent harness to fill in the rest. Pay attention to the "jev counter" on when api calls actually happen and the kinds of decisions its making. Its very rarely in a tight loop, and when it is (eg in an item menu) it tends to randomly walk through options.

Its very cool, but the harness is doing a _ton_ of heavy lifting.

Sure it can, just ask it for next char or "done" in a loop

> I'm guessing, AI slots invading the store.

Well, same as my lack of sympathy for github and their "inability" to deal with "unprecedented AI generated traffic", I dont buy it. Seems really simple to setup an algorithm of "existing app from pre-AI is submitting another update gets put in the priority queue while everyone else gets the shared queue".

Sure it works fine because you havent yet experienced how much better it can be. I encourage you to stop after the 'plan' part, read the plan, and tweak it. Have part 1 of the plan be it coming up with a framework to evaluate success, part 2 be it spitballing solutions, and part 3 be taking the best solution and "productionizing" it. You will get much tighter output and the brainstorming wont all muddy each others context as it tries different solutions. Bonus points for having part 2 fan out on its own, though thats much harder on shared hardware.

> How do you all keep your context so low? Complex tasks just balloon it in my experience.

By breaking the problem into discrete steps and aggressively restarting the prompt from the current state after completion of said steps

"Whats new" is, I hate to break it to you, 99% AI-related. Its a pretty unique point in time here and likely will continue to be for at least 12-24 months. This is likely to go down as being as reality-altering as the internet but with the added wrinkle of things are able to progress/iterate _much_ faster. Not only are the models progressing at breakneck speed, but our understanding of how to use them effectively is as well. So theres legitimate "discoveries" happening every week.

You get other things like mutual exclusion, a clean baseline regardless of whats in the main directory, and potentially in the near future a way to tie into lifecycle hooks for eg creation/deletion [1]. That last part is nice in the new agentic AI era where you might want to setup/teardown local services that are isolated to those directories.

But yeah the filesystem cost is not ideal. I was just thinking about how to exclude my worktrees from timemachine backups automatically.

[1] https://lore.kernel.org/git/7c8b4673-37ac-45fa-ad8c-a1dc09af...

> Money is fungible and instantly redeeming rewards for a statement credit is almost always the optimal way to use credit card rewards.

Uhh I have no credit cards where redeeming for credit is the optimal play. Every single one has 'offers' that give a further multiplier on the dollar amount of the points. Eg redeem $80 of points for a $100 home depot gift card.

That seems narrow minded. Theres both "learnings directly related to the thing studied" and "learnings downstream from the thing studied". If you imagine mathematics being a huge sudoku puzzle of unknowns you are trying to fill in, each previously empty square you are able to fill in (or gain a smaller bound on) has implications in all sorts of other areas.

Time-to-fix is lower, but time-to-new-bug is also lower

Ehh I love Casey and have learned a ton by watching how he thinks about things in his handmade hero series, but he is fairly narrow minded in his views of dev. Not that theres anything wrong with that, for the kinds of dev he does his approach is very good. But its not generalizable.

Yeah I had the same thoughts about what they chose to DIY vs hire out. Insulation is so simple, and trim can be done at your leisure. But thats also not likely to save much overall since the contractors also know how simple those are.

I assure you there are legions of nixos users doing similar things but getting fully custom configs on the other end

To more fully flesh out what my thoughts on this:

I think because more devs cant run things locally, we are not getting the "open source gains" that you would get by having the network effect of more eyes on the work. Theres still little nuggets of gold in there though (eg antirez's DS4 project). But once more people can get their hands on hardware, I think the flood gates will open. And once that happens, I dont necessarily think people will run locally. But that will enable fungible service providers which we already see, but at a much larger scale.

To a lesser degree, open weights also leads to an eventuality of these two companies not being able to sustain their capital investments. Not to mention their pricing models are unsustainable as they currently stand. So they are being eroded from the outside-in, while also trying to fight the internal pressure to continue to raise prices/margins.

EDIT: I actually think the harness is more of a future for these big players. And theres real value in providing a high quality harness. So maybe they move to SAAS for claude code in the future. Who knows

They are both on borrowed time. The only thing keeping them afloat business wise is that current hardware costs are prohibitive which keeps local models artificially constrained. The next era will be open weight / local model for the masses and SOTA for the big players.

I mean what Im hearing is "once we have AGI it will be easy". Sure, agreed. But AGI is going to unlock a lot more than software dev including having people just one-shot Lovable replacements. I also just fundamentally disagree its going to happen without another several "attention is all you need" events all chained together.

I think you are doing the equivalent of looking at a hockey stick graph and extrapolating to infinity on where we are headed. We are already hitting the asymptote and the line is starting to go horizontal. Theres plenty of gains to be had by current state of AI, I'm not trying to be a wet blanket. I just dont think those gains are going to be had by "the masses using AI". Its going to be highly technical people doing highly technical things at scopes and speeds previously unimaginable.

Im bullish on more people being able to build software than pre-AI but bearish on the general population being able to. The unfortunate truth of AI is that it requires you to think like a dev to be effective. And thats just not how a lot of people think or want to think. Thats fine. But that means AI wont be a huge unlock for the masses.

I was wondering why I suddenly started seeing these in my feed, I figured I would have to write my own extension to fix it.

Oh well, I cant be mad at the uBO team. Much appreciation for all they have given me over the years

This is pretty common in rum drinks, especially tiki. Like you said, rye offers something unique separate from traditional “bourbon” and so the flavor profile of the end drink is different

Famously, stock is settled on a delay (and generally doesnt involve physical products that are not fungible). Im sure theres a lot to glean from how they handle concurrency but Im not sure they are solving the same problems.

Awesome, thanks for taking the time to put this together, that really is impressive pre-LLM. We wont be at this level for a while but its great to have resources to look at for examples of others have solved things.

Im thinking through this for my team, any previous writeups on it? Or do you mind brain dumping the high level of how you had it setup from a design perspective : D

The whole point is to force the cloud providers to go through a contract negotiation for licensing rather than keep the door open for them. Nothing prevents the parent project from specially licensing to individual parties.

You can easily spin up a linux builder vm or I just have a dedicated remote builder on my network which I offload to

Intel is very very close to dead now as apple has officially EOLed them and software communities are following suit (eg homebrew and nix). I just picked up an apple silicon because the writing was on the wall that it could be any day that I could no longer get “real work done” on the intel machine

Most generics code is for library authors. Having generics allows you to write ergonomic algorithms for eg channels instead of forcing every single app developer to reinvent the wheel and run into the same footguns

The true power of LLM powered interfaces is the malleability of the whole thing. Sure you can write traditional software to handle “more like this”, but how hard would it be to handle “more like this but uptempo”, “more like this but not the same artist”, ad infinitum?

I have a bunch of skills Ive created for random work that have “predefined interfaces” like a single url. I very frequently trigger these skills with freetext prompts instead and get all the intelligence of the skill while bypassing what it was actually designed for. This aspect of malleable API is extremely powerful and is probably the only reason I keep AI in the loop of some of my custom software vs removing it and avoiding the cost/nondeterminism.

Im wrapping a lot of that up into "intent". If they intended it to rhyme, then you will necessarily "lose something" in translation. The same with elegant word choice to fit within certain metre, etc. Sure, a direct translation might be faithful to a more esoteric word choice but it completely loses sight of why that word was chosen in the mother language.

This also doesnt touch on translation also having an element of the audience in it as well. You see this in debates on what the best translation of the Odyssey is, with some preferring one version for being more "classical" in language and others preferring a more modern take. I would argue they can both be right, as its perfectly valid to choose turns of phrase that the intended audience will understand vs those which they wont. Language evolves over time.

Translation = words + intent. Literal translation frequently loses the intent encoded into the chosen language, especially for creative writing. If I translate "youre a real piece of work" literally into another language, Ive likely lost the entire point of the phrase. Should we prefer that translation instead of one that either spells out the underhandedness of the comment in the other language? Or which replaces it with its own approximately similar idiom?