From Vibe Coding to Agentic Engineering

A year and a half ago, I sat in front of Lovable, typed what I wanted, and watched an app appear.

It felt like magic.

I described a feature. It built it. I described another one. It built that too. Within a weekend, I had something that looked like a real product.

And honestly? For what I needed at that stage (validating an idea, seeing if something could work) it was exactly right. Vibe coding got me to my first product and I'm very grateful for that.

But then things got real.

I wanted to add sharing functionality. I needed proper error handling. Security wasn't optional anymore. I had users signing up, and suddenly "describe and pray" wasn't a strategy, it was a liability.

I'd fix one thing and break two others. I couldn't debug what I didn't understand. And the further I went, the more I realized: the tool wasn't the problem. My approach was. And markdowns alone don't help.

I was still vibing when I needed to be engineering.

The shift

The transition didn't happen overnight. It was messy. I moved to Claude Code and spent weeks figuring out what worked and what didn't. I dumped everything into a single CLAUDE.md file and wondered why it ignored half of it.

Slowly, I started building structure around the chaos:

Rules for path-based guardrails, so security standards loaded automatically when I touched sensitive files, not because I remembered to mention them.

Skills for repeatable workflows, so I wasn't re-explaining the same design patterns every session.

Hooks for quality checks, automated gates that caught issues before they shipped.

Subagents for autonomous tasks, letting agents handle scoped work independently while I focused on architecture.

MCPs for external tool access, connecting agents to the services they actually needed.

It took months of trial and error. But at some point, I realized I wasn't just using AI to write code anymore. I was orchestrating agents. Defining standards. Reviewing output. Making architectural decisions.

I'd gone from creative director to technical lead without fully noticing the shift.

This isn't just me

A week ago, Andrej Karpathy (the person who coined "vibe coding" a year ago) posted this on X (link):

Read that again. The person who started the vibe coding movement is now saying: the next evolution is agentic engineering.

"Agentic" because you're not writing the code; you're orchestrating agents who do.

"Engineering" because there's a real art and science to doing it well. It's a skill you develop. It has depth.

This isn't about vibe coding being bad. It's about recognizing that what got you started isn't always what gets you to the finish line.

Where are you in this?

If you're still in the vibe coding phase (exploring ideas, testing concepts, building your first prototype) stay there. That's exactly where you should be. Lovable, v0, Bolt: they're brilliant for that.

But if you've been noticing the cracks like builds that break in ways you can't trace, features you can't maintain, security you're crossing your fingers on 👉 that's the signal.

It doesn't mean you need to become a traditional developer. It means you need more structure around how you work with AI. More intention. More engineering.

The good news: you don't have to figure it out from scratch. I already made the mistakes.

Go deeper

If you're ready to make that shift, I wrote a comprehensive guide on how to set up Claude Code's customization layers (rules, skills, subagents, MCPs) so your agents actually listen and your builds are reliable from the start.

👉 Claude Code Customization Guide​

(Apologies for missing the link last time!)

It's the article I wish I'd had 12 months ago.

Oh, and you might have noticed this newsletter quietly changed its name from Vibe Coding Solopreneurs to Agentic Solopreneurs. Now you know why.

P.S.: Reply to this email if you're somewhere in this journey. I'd love to hear where you are and where you're trying to get to.