Fabien Butz

building AI. enjoying life. optimizing just enough to keep it worth living.

[Fabien Butz @f_butz]

@arpit_bhayani 800 million users. one primary.

meanwhile startups debate which database will scale before they have 10 users.

[Fabien Butz @f_butz]

boris cherny released anthropic's internal cleanup tool.

1.7M views. hot takes everywhere.

here's the thing: it's a prompt. a well-structured markdown file with ~50 lines.

two years ago that would've been a joke. now it's the product.

the skill shifted. not disappeared.

[Fabien Butz @f_butz]

@arpit_bhayani “it remembers you” sells better than “we indexed your data properly.” that’s the use case.

[Fabien Butz @f_butz]

you can use lovable 8 hours but can't decide what to build for 30 minutes.

that's not focus. that's dopamine.

ai tools made coding feel like scrolling tiktok. prompt, output, prompt again. same loop. same slot machine.

the vibecoding crowd will hate this. they're too busy prompting to read it anyway.

[Fabien Butz @f_butz]

@arpit_bhayani yeah. depth works when you know why you're going there.

spent years going deep because it felt productive. wasn't always.

[Fabien Butz @f_butz]

can't focus on work.
can focus on scrolling 90 minutes.

not a focus problem.
a reward problem.

[Fabien Butz @f_butz]

@HypeCrave @arpit_bhayani don’t give them ideas.

[Fabien Butz @f_butz]

integrated posthog today. took 5 minutes.

the data coming in is already changing decisions.

should've done this weeks ago.

[Fabien Butz @f_butz]

@milesdeutscher skills are underrated.

on-demand loading vs mcp’s all-upfront approach. might be the bigger deal.

[Fabien Butz @f_butz]

"the niche is saturated"

yeah. with garbage.

saturated doesn't mean no room.
saturated means nobody tried.

the bar is low. that's the opportunity.

[Fabien Butz @f_butz]

@alliekmiller the context folder setup is what most people miss.

building the same for sleep coaching. context changes everything.

[Fabien Butz @f_butz]

claude code after long sessions:
context pollution. quality drops.

my fix:
research. /clear.
plan to markdown. /clear.
implement. /clear.
validate.

fresh context + clear plan. every time.

[Fabien Butz @f_butz]

@arvidkahl same. shipping something while knowing there’s probably a better way you just haven’t learned yet.

[Fabien Butz @f_butz]

@milesdeutscher tool chain optimization for people who haven’t shipped yet.

getting started isn’t the hard part.

[Fabien Butz @f_butz]

building a workout app.

hardest part: same exercise, 1000 names.

canonical names + embeddings. done.

almost overengineered this with GraphRAG. didn't.

[Fabien Butz @f_butz]

@arpit_bhayani curiosity disguised as productivity.

classic trap.

spent months on something nobody needed.

learned a lot. shipped nothing.

[Fabien Butz @f_butz]

@arvidkahl “communicating intricacies”

aka realizing you’ve been running on instinct for years and never had to put it into words.

humbling.

[Fabien Butz @f_butz]

@boringmarketer patterns work until everyone scrapes the same patterns.

voice is the part that compounds.

[Fabien Butz @f_butz]

claude code auto-compacts at ~78%.

what to do:

50% → finish what you’re doing, don’t start new tasks

60% → tell claude: “write work summary to /scratch/summary.md”

70% → commit everything, quit, new session, “read summary and continue”

sounds like overhead. it’s not. re-explaining your architecture decisions for the third time is overhead.

[Fabien Butz @f_butz]

@arpit_bhayani don't apologize for weird.

the boring stuff everyone knows doesn't move the needle.

this does.