Kyle Rubenok

@LukeParkerDev This brings me back to the QLab days so hard. Miss this stuff.

Okay so you can control your lighting tech console with opencode now. ‘Make that beam red and make it fan faster’ ‘Set the fader to 40’

@michael_silver @kyeburchard FA (slop cannon), FO (SRE), then pretend to grow up but decide to be hot instead.

@kyeburchard Oh the many hats I’ve worn…

Hate to say it, but it's true. There are now only 4 jobs: slop cannons, SREs, hot people, and grown ups.

@tomwarren Not exactly the kind of ground control we were aiming for. Team is looking into it. 📫

NASA’s mission to orbit the Moon is being interrupted by Outlook (New) and Outlook (classic) both refusing to open 😂

@OmarShahine Incredibly handy for so many random things: amazon.com/WOLFBOX-MF50-E…

Do you know what is worse than Claude Code throwing API Error: 500? A crumb stuck under your MacBook Pro Delete Key making it not work. Need compressed air stat.

@matvelloso @turbotax 1) 🖕Intuit, they're the reason taxes are annoying to file. freetaxusa.com is great and well... free for federal and hilariously cheap for state. 2) Did this last year as well, lots of good insight, interested to see how this year's models do!

BTW, I highly recommend doing this. I found several obvious problems that @turbotax was unable to find (even when deterministic rules should be able to point to those, and I was paying for their services). These issues would have costed me quite a bit, not to mention risks with auditing. I am actually surprised how little Turbotax is able to catch. Also a bit baffled that they are not already using AI for final review checks and validation on top of it. The LLMs did a better job at guiding me on where to fix issues in Turbotax than Turbotax did itself on its own data.

@Salaman ADOPT TROUT

chipjobs 🤤

@kyeburchard @ericzakariasson The 5.x chat/instant models that serve most normie ChatGPT requests are _very_ different than the reasoning models. Sloptimized for both speed, cost and vibes.

@ericzakariasson how do you avoid this when people use gpt models in cursor? do you eval for it?

whats going on with chatgpt these days? almost all responses ends with clickbaity questions "if you want, i can tell you the one mistake that almost everyone forgets"

@jeffwilcox Ironically, forced WFH for the next two weeks after choosing to RTO for the last two years while we move. Will probably catch you on the highway if you keep that schedule up though 🫡

RTO day 1: do I wear black and mourn all that is lost, as someone who has been remote long before the pandemic but spent the first half of their career commuting to the sleepy zip code of 98052?

@matvelloso @bdsams

@bdsams People at Microsoft do it, but as a side job, so they can move faster. At Google moving fast is the main job, not a side quest.

Microsoft would never 😑

@jeffwilcox Someone got a new🪚💻 They're pretty nice. Much better than my old one.

My first Dell since high school, and, my first Windows ARM64 device!

@kyeburchard The black box will grow ever larger

@krubenok yeah it would not surprise me if some of the techniques were learning from the orchestration layer move into the model layer over time

has anyone tried building an LM architecture that uses a small model to select the most relevant context and provide just that to a larger model? re-rankers work great for RAG, but I wonder if a similar concept could be applied to the model calls themselves

@kyeburchard Oh interesting. At what point does pre tokenization begin and model orchestration end 😵‍💫. The future is cool.

@krubenok I was imagining it built into the base model as a pre-tokenization step. might make prompt caching way harder to implement, but just thinking about ways to build next gen models with orders of magnitude larger context windows

@burkeholland DevOops*

We're all DevOps now

@burkeholland Embrace the Mac whisper dictation hot key 💌 goodsnooze.gumroad.com/l/macwhisper

Mac dictation just transcribed me as saying "Jeep Eat Tea 5 3 Kodex".

@pierceboggan @code Way ahead of you, cutting the PR now

@krubenok @code Because it's a setting, you can put it in .vscode/settings.json and troll your whole team with different phrases :)

New in @code Insiders: Replace Chat thinking phrases with the `chat.agent.thinking.phrases` setting.

@thsottiaux @matteocollina Biggest flex here is that he called it a "Codex Subscription". ChatGPT who? What's that?

@matteocollina Pi is nice!

I’m really thinking of switching from Claude Code to Codex+pi as my daily driver. I guess I should just bump my Codex subscription to 200$ and try?

@michael_silver @joinkaizen @sharanjhangiani take notes to up your b2b saas-posting 👀

Everyone on this hellsite is talking about "taste" But have you tried selling B2B SaaS over Carbone?

@steipete github.blog/changelog/2026… HomeAssistant has been testing this with decent results. Sounds like the foundations of what you need?

PRs on OpenClaw are growing at an *impossible* rate. Worked all day yesterday and got like 600 commits in. It was 2700; now it's over 3100. I need AI that scans every PR and Issue and de-dupes. It should also detect which PR is the based based on various signals (so really also a deep review is needed) Ideally it should also have a vision document to mark/reject PRs that stray too far. This can't be fully automated, but even assisting would help. The closes I found is an obscure oss project. How's no startup working on this?