@SethSHowes Incredible, @SethSHowes !
This feels like the future becoming real — and accessible.
Soon it might sit on a shelf at home, just like a LEGO set )
@debelak_ja82216 …
Ironically, it feels like a return to the industrial era. I resist it a bit, but maybe it’s a phase we all need to go through with AI
I’m #BuildingInPublic — but not to sell anything.
No funnel. No paid subscribers.
I just want to share what I’m learning, maybe get useful feedback, maybe help someone who’s a few steps behind me ⤵️
@debelak_ja82216 Feels like an interview already 😄 thanks for the thoughtful questions.
Depends on the project — some things are ad-hoc, others are becoming structured.
I see a shift toward agents with specific roles and capabilities, forming a hierarchical system.
…
@debelak_ja82216 Hey — thanks for asking.
I’ve been using AI every single day for the past 2 years, and more intensively over time.
Not just for coding — writing, analysis, brainstorming, architecture, debugging ideas.
And yes, it’s always experimenting. Every day.
That’s the point.
@ZachChmael .. So “orchestration” = mostly me + terminal + structured prompts 😅
Trying to evolve it into something closer to a real router/orchestrator layer instead of manual juggling
@ZachChmael Thanks — really appreciate that 🙌
Right now it’s still pretty scrappy.
I’m bouncing between:
• tuning OpenClaw / ClawDBot with specialized agents
• manually switching windows
• SSH into my VPS where Claude Code runs
• Git repos for versioning
…
12/
So I’m building in public not to sell —
but to document how a non-programmer
is slowly building a personal R&D lab.
If you’re experimenting with routing/orchestration/agent structures —
curious what’s working for you.
11/
Resources are still limited.
I catch Claude Core windows.
Sometimes 30% of tokens go to discussion instead of execution.
I know this could be “solved with money.”
But not fully.
Because models have specializations.
It’s not one brain.
It’s a factory