// MBA · AI · FullStack · +15 yrs
What I wish someone had told me earlier.
About me
Hey, I'm Fran — full-stack developer at Nieve Consulting. In my spare time I built Cliencer, a complete SaaS from scratch on my own. These articles are the engineering lessons I picked up along the way.
I downloaded a SaaS provider's documentation, fed it to parallel AI agents, and had a working backend in hours. Your code is not your moat anymore.
I built a Claude Code expert agent trained on a 20k-star community repo instead of official documentation. It found 25 issues in my production setup, including exposed secrets.
You're working on a large codebase and Claude Code keeps missing context. How placing documentation at each directory level gives the agent exactly what it needs — nothing more.
Practical patterns for AI-augmented development: how I structure agent workflows to deliver production code, not just prototypes.
Every public function has JSDoc. Every decision has a rationale. Real examples from a production SaaS on how self-documented code eliminates tribal knowledge.
Code grouped by the problem it solves, not by its technical role. A practical look at feature-based architecture with real production examples.
Reusing code sounds efficient, but it creates hidden connections. Here is why I prefer independent modules that can evolve separately -- with real examples from a production SaaS.
How organizing code around business domains -- not technical layers -- lets each area evolve independently, reducing risk and speeding up delivery.
If a solution is hard to explain, it's probably wrong. Concrete examples of choosing simplicity over cleverness in a real Node.js and Next.js codebase.
How enforcing single responsibility across middleware, controllers, and domain logic makes a SaaS codebase faster to debug, safer to change, and easier to onboard into.