the meta is there is no meta

LOG 202601191930 #log

it's been a minute. like, eight months since i posted anything that wasn't about berry. and while she was absolutely worth every post, losing her in may broke something in me for a while. still carrying that weight.

but a couple months ago we got a new puppy. bun. yeah, we really like bun around here 😅. he's a mini australian shepherd / mini poodle mix, our little slice of bread, a smol lil cinna'bun. when we aren't orchestrating claude and codex we're snuggling bun. life keeps moving whether you're ready or not.

anyway.

i was doomscrolling x the other night and boris cherny dropped this thread about how he uses claude code. boris is the guy who created claude code btw, so you'd think his workflow would be some galaxy-brain setup that mere mortals couldn't comprehend.

and like... it kind of is? but also... it kind of isn't?

the viral thread

his setup:
- 5 parallel claude sessions in terminal
- 5-10 more sessions on claude.ai in browser
- CLAUDE.md as shared memory across sessions
- custom slash commands (he said "dozens of /commit-push-pr daily")
- plan mode first, then auto-accept for execution
- verification loops with tests and browser testing

i'm reading through this and nodding along and then it hits me.

wait.

i'm already doing most of this.

the realization 🫨

so i've been building this system called root over the past several months. it's my master claude code configuration system and it has:

PLAN.md for multi-session orchestration (literally what boris described as "shared memory")
four specialized subagents (eng for architecture, art for design, bot for testing, reporter for docs)
cook cli for scaffolding new projects with my patterns baked in
20+ custom slash commands
brand presets for frontend work so every project has its own visual voice

and here's the thing. i didn't build this because i read some thread or followed some tutorial. i built it because i was frustrated. because the default patterns felt limiting. because i kept running into the same problems across sessions and wanted a way to remember what worked.

the meta isn't a map to follow. it's something you discover by building.

there is no meta

that's the thing nobody tells you about these viral workflow threads. boris isn't sharing a roadmap. he's sharing what he discovered worked for him. and if you try to copy it 1:1, you're missing the point entirely.

the value isn't in the specific setup. it's in the journey of figuring out what YOUR setup needs to be.

i spent months iterating on root. started with just a CLAUDE.md with some basic instructions. then added slash commands when i got tired of typing the same prompts. then subagents when i realized different tasks need different personalities. then plan mode orchestration when context limits started biting.

each evolution came from pain. from friction. from "ugh, there has to be a better way."

and that's exactly how boris got to his setup too. he just happened to tweet about it.

the infrastructure

but wait, there's more. root is just the claude code layer. underneath all of this is farm.

farm is... a lot. it's a full dev workflow platform running on my vps. research to planning to execution, all in one place. task ingest from meetings via fireflies. manus as a research agent. claude code deeply integrated and running the whole thing. a custom ide i can access from any device. i can literally talk to claude from my phone.

it has this journal feature that auto-logs activity - was broken for a couple weeks, we actually just fixed it today. but when it's running, everything i do gets captured automatically. no manual notes, no forgetting what i worked on.

farm got stable a few months ago and it's been the backbone ever since. its architecture informed a lot of what came after. including hive.

the juggle

right now i'm riding account limits like waves, juggling projects across multiple fronts:

hive - this one's being built RIGHT NOW as i write this post. literally have codex spinning it up. it's a unified ai workspace for team collaboration that evolved from farm's patterns and something i called "ape" (orchestrator stuff). i built it because the patterns in root started feeling limiting for team contexts. first-class ai teammates alongside humans. we'll see where it goes.

genvid - ai video generation combining remotion, manim, and claude code. because apparently i decided static content isn't enough.

jf-reminders (firemind) - fireflies.ai to apple reminders menu bar app. solving my own problem of never actually doing anything with meeting notes.

claude-island - dynamic island style notification ui for claude. because claude deserves better than terminal output.

prompt-pro - an LMS about using ai, prompting, all of it. the twist? the whole thing is being generated. the content, the platform, everything. pure vibe-driven development with no pre-planning, but somehow it looks polished for the amount of attention it's gotten.

plus root and farm keep evolving. plus work stuff. plus bun demanding snuggles.

it's a lot. but here's the thing - we're not losing track of it. it's more like... focusing a blast of energy. staying in control while moving fast.

the real point

i want to be clear about something. i'm not posting this to flex or act like i've figured something out that other people haven't.

the models are getting better faster than any of us can fully adapt to. there's always another capability to explore, another workflow to optimize, another pattern to try. staying curious and staying in motion is the whole game.

but here's what i've learned: the journey is personal exploration. there's no "correct" way. there's just your way, discovered through building and iterating and shipping things that actually help you.

boris found his way. i'm finding mine. you'll find yours.

the blast

and you know what? despite the account limits, despite losing berry and pushing through that grief, despite the constant evolution of everything...

i'm having a blast.

there's something addictive about being in the arena while the tools are evolving this fast. every week there's a new model or feature or capability that unlocks something that wasn't possible before. and yeah, keeping up is a full-time hobby. but it's also exhilarating.

i just ship stuff now. things that would have taken teams and months and budgets i don't have. i'm building tools for problems i care about, at a pace that still surprises me.

what's next

honestly? no idea. hive might become something real or it might join my graveyard of half-finished experiments. genvid might ship or might pivot into something completely different. root will keep evolving as i discover new friction points.

the only certainty is that i'll keep building. keep iterating. keep discovering what works for me.

if you're reading this and wondering whether you should follow boris's thread like a recipe... don't. or do, i'm not your dad. but know that the real value isn't in the specific steps. it's in paying attention to what frustrates you and building solutions for it.

that's the meta. there is no meta. you create your own meta.

and then maybe one day you tweet about it and someone else reads it and thinks "wait, i'm already doing most of this" and the cycle continues.

anyway, back to watching hive build itself while i write about watching hive build itself. meta on meta on meta.

we can sleep when we're dead ✌️