Rafael Pinheiro

Ending AI slop in code.
Building dupehound: faster dog on earth 🐕

[Rafael Pinheiro @Rafaelpta57]

@doudou_19X @ycombinator Respect

[Garry Tan @garrytan]

Don’t be a mid startup

It’s true

[Rafael Pinheiro @Rafaelpta57]

@sarahwooders If you need a deterministic output, go with API. If not, an MCP might do it. One classic example is keyword research in tools like SemRush. You can't rely on MCPs for that.

[Rafael Pinheiro @Rafaelpta57]

@braelyn_ai many founders I know from Bay Area are doing the same. Honest question: why? And why Berlin?

[Rafael Pinheiro @Rafaelpta57]

No poetry, just wrappers

[Sahil Lavingia @shl]

When I was a kid, I wanted to be a rapper

Now I’m a wrapper

[Rafael Pinheiro @Rafaelpta57]

Just got this. Thoughts?

[Rafael Pinheiro @Rafaelpta57]

@shadcn Code smell is a thing

[Rafael Pinheiro @Rafaelpta57]

On OpenAI pitching ads in Cannes: disappointing and a bit embarrassing when smart people building breakthrough technologies still come up with one of the oldest and cringiest business model on Earth.

[Rafael Pinheiro @Rafaelpta57]

Love this. That is not a coincidence that OpenAI recently donnated to the Rust Foundation:

"Rust helps us move quickly on ambitious systems without having to choose between performance, security, and reliability” (Predrag Gruevski, Member of Technical Staff at OpenAI")

Dupehound is also written in Rust: https://t.co/KLJDIbVAQQ

[Matt Pocock @mattpocockuk]

Take a look at your favourite skill.

Go on, take a look.

Check for lines like:

- "Make the commit message very detailed"
- "Be thorough"
- "Make the implementation easy to read"

What do these lines have in common?

They're no-ops. They do nothing to change the agent's behavior.

Agents always write good commit messages. They always try to be thorough. They always try to create easy-to-read implementations.

Try removing them. Does the output change? No? Then the line was a no-op.

Agent-authored skills are LITTERED with no-ops. Entire paragraphs and sections of text that does nothing to change the agents' behavior.

No-ops make skills harder to evaluate, harder to maintain, and burn tokens uselessly.

Learn to spot them, test them, and remove them. Your skills will thank you.

[Rafael Pinheiro @Rafaelpta57]

@marconoiz Code review (especially in large codebases)

[Rafael Pinheiro @Rafaelpta57]

LLM’s > agents > harness > loops > ???

What’s next?

[dax @thdxr]

it's so crazy ai is bad at your job but good at everyone else's job

[Rafael Pinheiro @Rafaelpta57]

@steeldotdev Will do and leave you guys a feedback

[Rafael Pinheiro @Rafaelpta57]

@AlexanderTw33ts Free advice is worth what you pay for it.

[Rafael Pinheiro @Rafaelpta57]

@owenthcarey Dupehound is built in Rust (tree-sitter + Moss) to spot duplicated code in large codebases without using AI (local and offline). Give it a try: https://t.co/eQ5ijpX629

[Rafael Pinheiro @Rafaelpta57]

@markproduct spot duplicated code in large codebases

[Rafael Pinheiro @Rafaelpta57]

@wuweiweiwu @MomenticAI Dm'ed

[Aaron Epstein @aaron_epstein]

At the end of the batch, founders often ask about big looming decisions: should they be in-person or remote? stay in SF or live somewhere else? etc

The answer is: what would you want your competitors to do? Then do the opposite.