DJ

A real living legend. I watched The Way of the Dragon dozens of times. RIP #chucknorris

@burkeholland @WonderingDavid Yes, same here! Hit Windows key, see "Sessions".

@WonderingDavid I...don't know. What is that?

Whats this? @burkeholland

@burkeholland @slicknet I find Sonnet 4.6 works best when Opus 4.6 orchestrates. Definitely not visa versa. Sonnet poisoned the rely between Opus and myself a few times. So, that ended fast.

@slicknet Really? I find Sonnet to be the one model I don't want anywhere near my code. Maybe I need to give it another go!

Claude 4.6 Sonnet has been pretty impressive in my recent testing. I'm getting Opus-like performance on long-running, complex tasks.

@burkeholland I switched to Claude Code in my vscode-insiders terminal. GHCP Opus 4.6 I use as my remote ochestrator now via Discord. I do use vscode ghcp chat if I need a third opinion. The cc vscode chat plugin isn't that great vs ghcp chat imho.

I'm a little new to the CLI game - admittedly - but I do not understand how you people survive with just a terminal. Madness. Don't you wanna see the git timeline? Don't you ever look at the files in the project? DO YOU NOT USE DEV CONTAINERS?!?!

@burkeholland If we could ever get real fusion off the ground. I'm assuming AI will be required to provide real-time magnetic confinement and instability control to keep the Tokomak from eating itself.

I have considered that the real value of AI might not be AI at all. It might force us to figure out how to make clean, dirt cheap energy. It’s highly likely that is what the payoff of all of this will actually be.

@burkeholland I've been enjoying pair programming for all its warts. GHCP helped me go from coding project CMS systems in my basement for 12 years to actually producing a revenue generating player tracking system a year ago. Location-based restrictions could use a tweak though 😅

If you are going to post about some "insane" AI workflow or 6K agents burning 4T tokens to build an OS how we're all cooked or whatever, you have to include a screenshot of what you built with AI or I don't believe any of it. I'll start since I'm the most guilty of this.

@burkeholland @microsoft I wish I could work there. I was at the Bellevue MS building with PeopleTech for about a year. Best time of my life. Did a General Motors project as UX/UI Generalist and Technical Artist. artstation.com/artwork/d84v0w

I think we may need more developers

@burkeholland Did you update your agent prompts? Mine seems to run better after I updated both Beast Mode and setup copilot-instruction to be complementary and not redundant.

I’ve been using Opus 4.6 all weekend and here’s my verdict: It’s a faster 4.5 Which is saying a lot cause 4.5 is no slouch. Only downside is it uses a MASSIVE amount of tokens. Just so many. You’re going to need that 1M window.

@burkeholland You read my mind. I can't even use Sonnet anymore. Costs 3x just to debug the slop it writes

What are you expecting from the rumored Sonnet 5? I personally would like to see Opus 4.5 level performance at Sonnet cost. But why do that at all?

@burkeholland @jpreagan @openclaw 💯

I'm jealous not gonna lie. I'm very close to doing the same thing, but I have a hypothesis that I'm going to work out this weekend. It goes like this... You don't need a complex agent. All you need is one that 1) knows how to use a browser 2) knows how to use and create skills as it figures things out and 3) has a long running memory. That's it. If an agent has just those 3 things, it should be able to do anything.

you guys can't be having all the fun without me @openclaw sitting on my desk rn it's going to be a good weekend

@burkeholland @code github.com/djcyphers/netw… I love these CLI tools! They can be portable to solve specific tasks. My use-case was diagnosing network issues quickly with Copilot.

Reasons I do not like it... ⛔ I really like the Copilot agent in @code. Especially for Opus 4.5. If you've used it then you know. It's crazy fast, super accurate. Perhaps I get the same experience in the CLI, but these things are so subjective. ⛔ I want to see my past agent sessions in the sidebar. I miss that so much. ⛔ The UI for the tools, custom agents and models is so nice in @Code. CLI's are just not good when you need multiple menus of things to be visible.

This is a @code layout I've been experimenting with a lot lately. wdyt?

@burkeholland Chrome MCP is such a game changer.

I just watched Sonnet 4.5 use the Chrome MCP server and click the "I'm not a robot" checkbox like it was light work

@burkeholland Jokes on them I've been writing garbage since MySpace, AI has now helped me 100x my GIGO powers 🤓👍

AI will never write code that is up to your standards of craftsmanship. Because you are a craftsman and AI is not. The question is whether or not that matters.

x.com/i/broadcasts/1… Yes.

@FATAL1TY Man, been having a blast playing Arc Raiders. Watch out for those clanker wankers! 🤣

$5 Million Gaming Grind for 5 skill points. Arc. @fatal1ty x.com/i/broadcasts/1…

@github @AnthropicAI @code github.com/djcyphers/netw… I made a copilot-cli wrapper that leverages skills. Thank you github team!

Stop repeating your instructions. 🛑 Copilot now supports Agent Skills, the open standard by @AnthropicAI. Package your specialized expertise into skills once and use them everywhere. Try it in: ✅ agent mode in @code Insiders ✅ Copilot coding agent ✅ Copilot CLI Here's how to get started. 👇 github.blog/changelog/2025…

@github I made a wrapper for it for use in Windows server networking. github.com/djcyphers/netw…

👋 Have you tried GitHub Copilot CLI yet?

@enteio Beatings will continue until morale improves...

"Google Antigravity just deleted the contents of my whole drive" 🫠 old.reddit.com/r/google_antig…

@burkeholland I would sell my left leg to get models with a larger context window. Maybe double the current in/out as a good compromise vs the 984k context on OpenRouter? I'm looking at Sonnet 4.5 specifically, but I'm sure others could apply. PS. OpenRouter wants two legs!

@burkeholland @code All my premium requests.

Trying out a new workflow in @code to maximize premium requests in Copilot: Budget: 2 Premium requests (Claude, GPT-5), Unlimited Included (GPT-5 mini). 1. Create Plan (Sonnet 4) 2. Convert plan into a set of "phase docs" that each define a single PR. Each doc contains ALL of the code needed for that PR and exactly what steps to take (GPT-5) 3. Implement phase docs one at a time (GPT-5-mini). This uses 2 premium requests and everything else is handled by GPT-5 mini which is included. So technically GPT-5 is writing all of the code, but mini is actually implementing it. This means I can move in small steps and iterate with mini but get the quality of GPT-5. Using this I built a Windows application last night that lets you resize Windows to a specific width / height. This interesting because it uses Windows API calls that I have no clue about, as well as things like XAML and WinUI which is also new to me. One evening from idea to working app. Next stage is code review and publishing to the store. AI is not hype, but the way we talk about it sure is. We need more practical workflows and less tweets about 100M MRR SaSS apps developed over a weekend. Start small, move incrementally, be smart with model usage and you can do a LOT. Disclaimer: I did call in Claude towards the end to do some UI cleanup because it's just so good at that. 3 Premium Requests burned on UI tweaking.