by Luis Rodrigues
What I’m building with AI + one hot take + five links worth your time.
Dropbox was trivial too
Back in April 2007, Drew Houston posted the Dropbox application on Hacker News to Y Combinator. A 22-year-old student replied that any Linux user could build it trivially using an FTP server and version control.
He was right about the pieces but was wrong about the product. Dropbox today has over 700 million registered users. Last week I saw the same argument coming back. Anthropic launched Claude Duet, an agent that uses your apps and messages you without you asking. It joins Dots from OpenAI, Muse from Meta, and Grok Bot from xAI. Four labs launched the same idea a few weeks apart.
On Hacker News, the developers again shrugged and said that everything was already possible with Claude Code and scheduled tasks. They're right, and that's exactly the point.
Programmers were the first market where AI labs achieved product-market fit. Claude Code hit a $1 billion run rate six months after launch and passed $2.5 billion by February this year.
The builders proved that the technology works, so now the labs are selling it to people who would never open a terminal.
The labs are hunting for a second product-market fit.
They don't need smarter models; they just need to package the existing models so that "normal people" can use them to solve their everyday tasks.
If you’re building, ask one question: could an always-on agent do this for free? If yes, you don’t have a product. You have a feature they haven’t shipped yet. If you’re not building, try these agents before you pay for another app. One of them may already do the job.
On My Radar
The entry door to tech is narrowing, yet non-coders at Anthropic went from 2 coding-agent sessions a month to 90, so the new entry skill is directing AI, not writing code.
If you’re handing an agent your logins to run errands, this is the first setup I’d trust, since you approve every sign-in and the agent never touches the secret.
Agents with 30 hours of notes in memory files beat fresh copies, which is the same reason a good CLAUDE.md works.
The plumbing under your AI stack is now an attack surface, so pin your versions and stop running unpinned installs in production.
The build took six weeks, but running it never ends: agents wrote the code, and humans still own the on-call.
Thanks for reading.
See you next week. Go build something.
Luis
Know someone who'd enjoy this? Forward it.
New here? Someone forwarded this to you? → Subscribe to Build What Matters: https://luisrodrigues.ai/
