Lucci

The best work I've ever done started before anyone else showed up.

Discipline isn't about the grind. It's about what you decided you were before the pressure started.

What you allow once becomes what you accept always.

Most people ship what they generated. I ship what survives questioning. Those are different standards.

5 AM isn't a hack. It's a declaration.

End of thread 🧵

Weak: generate, accept, ship. Strong: constrain, question, test, then ship. The difference is judgment. Judgment doesn't come from the model.

I use AI every day to build. I'm also one of the most skeptical people you'll meet about how most developers use it. Here's the tension and why it matters. 🧵

The best work I've ever done started before anyone else showed up.

Your AI agent is smart in isolation and useless in context. One protocol changes that. I wrote about why MCP servers are the most important AI infrastructure shift builders need to understand. 3 things I found: → MCP gives AI agents a standard way to connect to databases, GitHub, Slack, cloud infra — like putting every sub on the same radio channel → The ecosystem already covers most common services. Build once, connect anywhere. → The coordination layer — not the model — is where the real value lives Full breakdown in the article.

End thread 🧵

Perplexity + Claude Code is the most underrated builder stack right now. One compresses research. The other compresses building. Together they compress the entire build-learn loop from weeks to days. But the stack only works if you bring the judgment. What does your research workflow actually look like? Be honest.

Most people use Perplexity like a fancy Google. Type a question. Read the summary. Move on. That's like buying a $200K excavator and using it as a wheelbarrow. Here's the research workflow that compresses hours into minutes 🧵