Back to Blog
·Marcelo Otero

What I Learned Shipping 7 Side Projects with AI-Assisted Development

AI
Side Projects
Product Management

Over the past year, I've shipped 7 side projects using AI-assisted development tools, primarily Claude Code and Cursor. These aren't tutorial projects or weekend experiments that never saw the light of day. They're live products with real URLs, real users, and real problems they're trying to solve.

The projects span a range:

  • Pinol: An AI financial copilot for Nicaraguan workers
  • The Original Snake: A pixel-perfect Nokia 6110 Snake recreation built as an SEO-first product
  • Tailor Resume: AI resume tailoring with job fit scoring (currently being rebuilt)
  • Amplituned: A web design and performance studio with lead capture and GA4 tracking
  • Otero & Vidaurre: A live law firm website with Keystatic CMS
  • Petal Coach: An AI-enhanced career self-discovery app built using GitHub Spec-Kit
  • FlashFive: AI-powered daily micro-quizzes (in development)

Here's what I learned.

AI Doesn't Replace Product Thinking

The most important lesson: AI development tools are incredible at generating code, but they can't tell you what to build or why it matters.

Pinol is my strongest example. The product serves Nicaraguan workers who need to make financial calculations (social security, severance, tax withholdings) that are notoriously complex and poorly documented. The key architectural decision was separating computation from interpretation: deterministic calculators handle the math, and the AI layer explains the results in plain language. AI never computes financial numbers.

That decision wasn't made by an AI tool. It came from understanding the user problem and the risk of AI hallucination in financial contexts. Product thinking came first; AI-assisted code came second.

Speed Is Real, But So Are the Traps

AI-assisted development is dramatically faster for certain tasks: scaffolding components, writing boilerplate, implementing standard patterns, debugging, and iterating on UI. A project that might have taken me weeks of evening coding now takes days.

But the speed advantage creates its own traps:

Over-building is easy. When code is cheap to generate, you can end up with features nobody needs. Discipline around scope matters more, not less.

Technical debt accumulates faster. Fast generation means fast accumulation of patterns you didn't fully think through. I learned to pause and refactor regularly rather than just shipping more features.

You still need to understand the code. AI can write code you don't understand, and that's a recipe for bugs you can't debug. My engineering background helps here. I read everything the AI generates and push back when something doesn't make sense.

Spec-Driven Development Works

For Petal Coach, I tried a more disciplined approach using GitHub Spec-Kit for spec-driven development: 8 user stories, 28 functional requirements, 57-item quality checklist, and 46 implementation tasks, all defined before writing any code.

The result was noticeably better. The AI had clearer context, generated more accurate code, and required fewer corrections. The upfront investment in specs paid for itself in reduced rework.

This mirrors what I do professionally. PRDs, user stories, and acceptance criteria aren't bureaucratic overhead. They're the instructions that help both humans and AI tools build the right thing.

SEO Product Thinking Applies Everywhere

The Original Snake was an experiment in treating a simple game as an SEO product. Every technical decision (page speed, crawlability, engagement metrics) served an organic search acquisition strategy.

This cross-pollination between SEO knowledge and product thinking is something I bring to every project. Understanding how content gets discovered, how pages get indexed, and how users find products is a product management skill that most PMs don't have.

The Real Differentiator

Plenty of PMs talk about AI fluency. Few have shipped multiple live products that demonstrate it. The side projects portfolio shows that I can go from idea to live product independently, make real architectural decisions, and iterate based on data.

But more importantly, these projects keep my technical skills sharp. When I'm reviewing a developer's PR, discussing API integration approaches, or evaluating technical trade-offs, I'm drawing on hands-on experience, not theoretical knowledge.

For any PM role that values technical depth, AI fluency, or builder mentality, this portfolio is a genuine differentiator.