Bitcollage

ad astra per aspera 🚀 ✨🪐
can you? 🤖

[Bitcollage @Bitcollage]

Exactly 😅 Otherwise it is just prompt engineering. The problem I ran into is that bigger and more complex skills become slower and harder to control. Splitting them into smaller skills does not really help. I even tried defining state machines with transitions as structured tables inside skills. It simply does not scale.

So I use both approaches. Traditional skills still run at specific nodes/checkpoints inside deterministic workflows, returning structured output for the next step.

The cool part is that even smaller, faster models work well this way. The state machine validates their output and lets them retry until it is correct. They do not need to understand the entire workflow.

[Bitcollage @Bitcollage]

@futaspor Fake News!

[Bitcollage @Bitcollage]

Ready is not a feeling; it is a decision. Being stuck costs more than taking the leap.

[Bitcollage @Bitcollage]

@vineerpasam Your same dude:

https://t.co/PjG85u6JIB

[Bitcollage @Bitcollage]

@TheDavidHasbun Agree!

https://t.co/tbUoN96Lms

[Bitcollage @Bitcollage]

It is not as simple as you make it sound. People like @unclebobmartin and @Linus__Torvalds were still working with assembly long after higher-level languages like C were available. Uncle Bob even wrote his own tools to change the assembly code generated by compilers because he wasn't happy with it. I think we're in a very similar situation today.

[Bitcollage @Bitcollage]

I don't think you actually read what I wrote. It's not about reading or learning assembly or machine code. My point is that we're in a very similar situation today. The difference is that instead of assembly, we're now looking at Python, JS, Go, etc. What do you think people who care about building good software want? They want to make sure the code being produced is actually good. That's exactly where I see the similarity. How long this phase will last, I don't know. Maybe in a few years we won't even need to look at that code anymore.

[Bitcollage @Bitcollage]

It is not as simple as you make it sound. People like @unclebobmartin and @Linus__Torvalds were still working with assembly long after higher-level languages like C were available. Uncle Bob even wrote his own tools to change the assembly code generated by compilers because he wasn't happy with it. I think we're in a very similar situation today.

[Ryan X. Charles @rxcastrohacker]

deeply disagree. agentic engineering is the only skill that matters now. in the same way that almost no one needs to look at a binary, we *already* have tools that should be reading and writing all code, and they will only get better. do not waste time learning syntax when you could learn system design and architecture.

[Bitcollage @Bitcollage]

@rxcastrohacker @unclebobmartin @Linus__Torvalds Have fun!

[Bitcollage @Bitcollage]

Do not try to forgive them. Just let it go. They knew exactly what they were doing. They did it knowingly and on purpose, sometimes even while looking you straight in the eyes. And they were not even ashamed of it. Instead, they convinced themselves they were right and made you the one to blame.

[Bitcollage @Bitcollage]

🚀🚀

[David K 🎹 @DavidKPiano]

XState v6 is now in beta 🚀

Reduced API surface area (transitions are just functions), much stronger types, durable execution, new tools (CLI, viz, editor), and more to come.

Built for humans *and* agents 🤝

npm i xstate@beta

https://t.co/JJStSdXwwQ

[Bitcollage @Bitcollage]

@Armandotrue Best decision ever made and still happy.

[Bitcollage @Bitcollage]

@rauchg @tomastrajan Totally agree and still many others claim the opposite!

[Bitcollage @Bitcollage]

Playing dumb with an AI agent often works wonders!

[Bitcollage @Bitcollage]

Almost all of my skills (at least in Claude) are basically not really skills anymore in the traditional sense. I use function hooks from Claude Mods (not the classic hooks) to catch these events. From there, a workflow/state-machine takes over and everything runs deterministically. Only things that cannot be solved deterministically are handled by the LLM at specific nodes or checkpoints inside the workflow. So basically, the workflow controls the LLM, not the other way around.

[Bitcollage @Bitcollage]

@kentcdodds @ryanlanciaux

[Bitcollage @Bitcollage]

Do not give a single second of your life to anyone who cannot apologize, say thank you, admit their mistakes, or make things right.

[Bitcollage @Bitcollage]

Agentic engineering is challenging, creative, and a lot of fun. Especially when you are building on software and libraries that do not exist yet, but you really need them because the existing tools just do not do what you want.

[Bitcollage @Bitcollage]

I worked for years (~10) in understaffed teams. There was never enough time or budget to do things properly. Thanks to those damn MBAs in management.

Even hiring was always limited. And when I was allowed to hire, I often couldn't even choose the people I really wanted. So many talented people slipped through our fingers.

I tried to compensate for all of this with automation, but there is only so much you can automate.

Now I finally have my buddy Claude. I've been using AI from the early days, and Claude became my real coding buddy in November 2025.

For the first time, I can actually build the things I always wanted to build. I can do things the way I always thought they should be done. And I'm enjoying every single moment of it.

But on the other hand, I can also understand people who find AI coding boring. Maybe they had the freedom, time and resources all these years that I never had.

For me, AI coding is not boring at all. It's finally freedom.