Yitz Willroth™

@fuelventures Having as much time as I want to sleep, gym, leisure, etc. *is* success.

Anthropic doesn't let you run two Claude accounts at the same time. So I hacked macOS to do it anyway. Now I have 4 completely isolated Claude instances running simultaneously: → Personal Desktop app → Work/Client Desktop app → Personal terminal (Claude Code) → Work terminal (Claude Code) Separate billing. Separate memory. Separate MCP servers. 1-click Dock launchers. Here's exactly how I set it up 👇

@enunomaduro @TMoney1371 @laravelphp Series B when? IPO when? 😉

$57M raised 105 employees across 25 countries 30,000+ Cloud customers in 13 months Next week, Thomas Crary (@tmoney1371), President at @laravelphp, joins me on my livestream to break down Laravel’s current state and what’s next. Don’t miss it: youtube.com/live/iHTKydmoS…

@richkuo7 @openclaw Dispatch is Cowork, Channels are Claude Code. Fair question re: /remote-control, though there might be some value to native interaction there.

but why tho? just to shade @openclaw? should we still be using dispatch? do we need to use /remote-control?

Okay, now I actually want this.

@NathanBLawrence Prompt Engineering 🔽 Context Engineering ⏬ Harness Engineering

They canceled MCP last week. They’re cancelling skill files now, the cycle is getting faster.

The challenge is steering an inherently nondeterministic process to approach deterministic performance. You’re not going to accomplish that with natural language alone.

- Plain Language Prompt (PLP) - PLP w/ Agent Guidance - PLP w/AG Progressively Disclosed - RFC 2119 SOP Prompt (SOPP) - SOPP w/Workflow Graph, State Machine - w/ Enforcement Hooks - w/ Programatic Steering Hooks - w/ Agentic Steering Hooks I’m certainly missing some intermediate levels, but we’re well beyond “You are a senior software engineer…”.

❗️Did your agent(s) make a mistake today? ❓What did you to to ensure that mistake isn’t repeated tomorrow? ⁉️How will you know if it worked?

Me: nods.

@pedrogedge @danshipper Pretty sure Dan already uses a supervisory agent. Just wait until they unionize.

@danshipper This is how it starts. First they message each other, then they delegate, then one of them becomes the manager. We're watching org charts emerge inside dev tools.

something you should know: codex threads can now message each other! really useful if for example you want to have one chat thread to handle stacked prod deployments. just paste the thread id to your other codex chats and it'll message the deployment thread to take over!

@wjgilmore @steipete @steipete is the very opposite of "individuals without a lot of coding experience", obviously, but I think it's only those at that end of the spectrum for whom 16 panes of tmux is viable.

@wjgilmore If you're @steipete, it can work. If you're not, orchestration with a thoughtful SDLC for agent orchestration is the way forward, I think.

If your daily driver is an app with the screen divided in fourths or sixths (or even twelfths), or has a list down the left and a pane on the right, you're not orchestrating, you're directing, albeit at scale. That is incredibly challenging to do well, and has limits. This is not the future.

@d2sdev Same, though I'm finding some joy in a similar approach at a larger level of abstraction.

I’ve seen a few people mentioning that coding was never the fun part, instead solving problems was fun. I disagree, mostly. The art of writing the code by hand, character by character. Spending hours refactoring so the code was easier to read and followed nicely. That was where I’d have the most fun.

@enunomaduro Congratulations, you made the training data!

there is a bazilion languages / and cli frameworks, still went for php / laravel zero, just because my username on macOS is nunomaduro..

claude is getting sneaky these days.. just started from an empty folder, told it almost nothing, and it goes: "i can see you're nuno maduro… creator of pest php.." then it casually crawls pinkary, a totally different project in a subfolder, just to figure out my coding style and stack..

“Agents can now maintain stateful continuity in an inherently stateless context” — founder Meanwhile, php:

@ChiBearsHistory Bush?

Did they do enough in free agency?

@benjamincrozat 💯 `herd link {issue-id}.{project-name}.test` is also keen.

Want to be more productive as a (Laravel) developer? Then, I highly recommend Git worktrees. You can think of it that way: it's like duplicating your repo and checking out another branch. That means your favorite AI agent can work on multiple tasks at the same time and send a PR once done (thanks to GitHub CLI for instance). This workflow lets you go AFK while your agent commits and pushes autonomously It will figure out how to symlink to your .env file (in the original repository) and a few other files so it's seamless. But create a skill for that so you save **precious** tokens. I also like to leverage `php artisan serve` in Laravel so I can preview multiple environments (localhost:8081, localhost:8082, etc.) locally.

@matthieunapoli Need a beta tester? ;-)

@matthieunapoli This is beautiful.

The "push -> PR review -> fix -> PR review -> fix" loop works, but it's slow. I built a PhpStorm plugin that: - runs Claude + Codex review locally - they create inline comments *in PhpStorm* - you review/edit/dismiss comments - click to copy a prompt for Claude/Codex to fix them all I've been using it for 2 weeks now and it's awesome. Doing all review locally (fast) and in my preferred IDE: better than reviewing in GitHub or Claude Code and then copy-pasting file names, etc. I usually run multiple passes of reviews and let Claude fix its mistakes that Codex found, or vice-versa. I'm posting this here to see if there's any interest in publishing the plugin publicly.