Adam Reed

@chadjanis @GetSequence is 9/10ths the way there they need to move faster tho, been using the service for years and they have been slow to improve. Feels like a hobby project still. In the age of AI they should be moving 10x faster.

What's amazing (and honestly unexpected), reading through the comments and quotes on this, is how quickly people (1) explain away the idea or (2) relate it to something they're working on or exists in the market and say it's already been done. There's a reason entrepreneurship is so difficult. Our biases and worldviews creep into the problem-solution phase. The closest thing I've seen from the comments is @GetSequence, but it's still not quite what I'm envisioning here. Sequence is likely built for people in the comments who have money and want to creatively design their financial system -- maybe not the target audience for this. The problem isn't "help me route money". Sequence solves that. The problem is "I don't have enough money". The solution is to help force money management and discipline to help more people have enough money using the limited funds they already have. Not granular digital budgeting (YNAB) or expense visibility (Monarch). The good news is that Sequence is likely using the same underlying infrastructure but deploying it for a different audience/use case. UX is critical for this to work. The audience size for "I don't have enough money" is probably 75%+ of Americans. And that's not even getting into the Business use case. Wishing good luck building to those who understand the mission and use case here!

@myfirstmilpod @chadjanis This is already a thing @GetSequence

Whoever builds this idea first is guaranteed a $1B exit in two years. @chadjanis just sold Gruns to Unilever for $1.2B. Here is his next top-tier idea that he's giving away for free: The problem: your paycheck hits your checking account, then you need willpower to allocate it. Put 10% in savings, pay bills, invest, whatever. Most people FAIL. The solution: become the distribution layer between direct deposit and your bank account. Before the money hits your account, it automatically splits: 20% → taxes (sitting in money market until tax season) 25% → rent/car payment/bills 15% → investments 10% → savings What hits your checking? $500 for groceries You never see the full amount. You can't fail at budgeting because there's nothing to budget. This works for businesses too. Founders want guaranteed profit extraction. VCs want forced financial discipline. Everyone buys. The infrastructure exists and the market is waiting. The first person to build this is getting acquired in 24 months. @thesamparr @ShaanVP

@TechnoTimLive Oh it does a whole lot more than that, been reverting this busted shit for an hour now

Just a heads up if you are updating Proxmox to kernel 7.0.0-3-pve. Some LXCs might not be compatible with that kernel version. This is the first time I have run into this but figured I would mention it in case you are wondering why some of your LXC services might be crashing after that update. VMs of course are not affected because they are fully isolated. To be clear you should update to the latest kernel, just be sure to check your LXCs afterwards.

I really just like to program Hands on keyboard, music, deeply thinking and enjoying the process

@_mtmk Better! I threw my idea on there on a comment as well.

@IAmAdamReed Fair. how does this sound? github.com/synadia-io/orb…

ICYMI we have recently released fair few Orbit .NET packages. They provide additional functionality on top of NATS Core and in most cases JetStream. #packages" target="_blank" rel="nofollow noopener">github.com/synadia-io/orb…

@Aaronontheweb @RebTechUPA My leftover ethereum mining equipment lol

@IAmAdamReed @RebTechUPA dude I have been WONDERING where to get a setup like that for chaining old cards together

Local AI 🤖 🙌

@Aaronontheweb @xoofx What kinda tok/s 🤔

@IAmAdamReed @xoofx I'm running 2 256k context length sessions per card right now. I could go higher if I reduced the KV cache quantization.

I should receive an AMD AI PRO R9700 32G VRAM today to test some tiny LLM models locally. It feels the best bargain these days for local inference. 😎 2 of them like this and it reaches the price of a single RTX 5090 and from the specs, it's not that far in terms of perf. We will see!

@engineers_feed I hear fetty wap. SEVENTEEN THIRTYEIGHTTT

What happens in your head when you add 37+48?

@Aaronontheweb @xoofx How do you get any context length on 32g? My initial Claude prompt alone is 52k

@xoofx This is what I've been using, 2X of these. I run a full instance of Gemma4 or Qwen3.5 on each and have a reverse proxy do sticky-session load balancing between the two

@Aaronontheweb Worse usb to pcie 1x risers

@IAmAdamReed you're basically stuck with PCIe

Stuff is still so early in the AI / LLM space - I'm sitting here doing a manual build of llama.cpp so I can get virtual memory management for the newest generation of AMD ROCm GPUs. This is the "compile your own kernel" era

@davidfowl I’m sorry

I'm building a native windows app for the first time in 10 years...

@davidfowl My higher ups say “your being pedantic the software works and the models are only getting better, accept it and move on”

There's a really high cost of duplicating code and we're all relearning that everyday with these coding agents. Until they get better at generating reusable code, bugs will run rampant. You can't "abstract it away" with more agents. You just get crappy software....

@thdxr I have to fight the urge to do this every day. Especially after AI is now my primary workflow. Every time I look at the code it wrote I cringe and click merge.

are there people out there who just want to refactor every day? just wake up and find the worst code and just chip away at it and clean it up wake up the next day do it again, infinitely improving things with zero external impact?

@Aaronontheweb This is why I gave up on my Claude code proxy. So much nuanced bs, and Anthropic hates everything.

The amount of minutiae involved in writing LLM harnesses is insane

@rick01 I do something similar with Blazor WASM and use NATS right in the frontend via wss.

@IAmAdamReed NATS owns the state and dispatches updates. Datastar lets me keep my UX a shell. Blazor renders HTML. Datastar merges it into the DOM. Zero client-side code. It's the easiest way to build reactive UX, imo.

They said AI programming would feel good. They didn't say it would feel this good.

@mitsuhiko Try cmux!

I think tmux is great software for an agent. But how people can actually work day to day in tmux is beyond me. It's such a horrible UX and hack.