A friend read one of my research runs and hated it. The research was right: their plan wouldn't work. What they wanted were the routes and loopholes to make it work anyway.
I spent the morning arguing about it with @boardyai. Right finding, unusable output. His parting line is now on a post-it on my desk:
"It's mostly one habit: asking what the person actually does next after they read it. Everything else follows from that."
How to stop losing Claude Code's best suggestions in a pile of AskQuestion prompts:
Make them a question too.
I made a hack that turns the "You should know" mod into an AskQuestion, so Claude waits for your answer. Works on Claude mobile too.
Advice you can't miss is advice you read.
Before you build your next app, ask whether a vibecoded clone could get the same result.
If it could, someone will ship a cheaper clone. Then a cheaper one. Then an open-source copy.
If it couldn't, good: you've found what to build.
The most useful thing an AI does for me: kill my ideas.
@boardy_ai has been refining my products, shutting down my bad ideas. Right now he's helping me shape something new.
I call him my boxy overlord.
Building small tools for myself changed how I judge software.
The big apps on my phone have endless feature lists. I never open them. My home-cooked ones are tiny, and I use them every day.
An app you never open has zero features.
Built a haunted keychain: an ESP32, a tiny screen, and a GIF gallery for it to cycle through.
Lain is there, of course. The catch: when it's off, it's a broken TV hanging from your keys. Oh well.
Ev Williams hired me to help fix Twitter in the early days, back when the Fail Whale was showing up all the time. On Rails.
That was buried somewhere in my LinkedIn. So was being a Tech Lead at LivingSocial lol.
Yesterday @boardyai helped me rewrite the whole profile, and I had to get precise about my own history. Building software since 1998. Engineer, then lead engineer on Chargify. Early Twitter at Obvious Corp. LivingSocial. Flexcode since 2016.
It's live this morning.
changed my username to make things more simple:
instagram.com/caffo.dev
Stop waiting to finish your side projects before you show it. My Game Boy game about Murakami's "The Wind-Up Bird Chronicle" never got finished. It's online anyway: drafts.caffo.dev/windup/
Nobody sees the version on your laptop.
My Claude Code subagents now know when my M2 has had enough. If Claude decides a job is too heavy (a Rust test run, a long thread), the subagent runs on my beefy Linux machine instead. The laptop keeps the light stuff and finally stops gasping. So good.
A senior engineer through Flexcode costs around $100K a year. That's on purpose.
I've run Flexcode since 2016. We embed senior engineers from Brazil and Latin America directly into existing product teams for mobile and web work. They're not a separate outsourced team, and they don't need hand-holding.
We've worked with Twitter, Lufthansa, FOX Networks and Groupon. Placement takes about two weeks.
We don't compete with the cheap shops. It's a narrow niche: funded companies that want people who know what they're doing and do it well, at a price that's fair but not bargain-bin.
Almost all of our work has come through referrals. I do one call to explain how we work, and after that the developers' work speaks for itself.
If that sounds like your team, my DMs are open.
Rodrigo Franco (Caffo) retweeted
Hot take: the best UI library in 2026 is the one your AI agent can actually read
uiarc.dev














