Lately, I’ve been on a wild ride using Cursor AI to build an iOS app in Swift—even though I’ve never been an iOS developer before. In just one week, Cursor taught me Swift fundamentals, debugging techniques, and how to think in terms of project features. What began as a series of late-night experiments quickly evolved into a full-blown project filled with innovative ideas and plenty of unexpected challenges (including Cursor misclassifying my Swift app as a Next.js project!).
Key Learnings & Takeaways:
✅ Do: A Structured Feature Process Is Essential:
I relied on a framework that included:
• PRD.md A detailed requirements document with clear [MVP] markers for agents.
Combined Status/ Architecture.md: A live progress and design log capturing both day-to-day updates and overall system architecture.
.rules/cursor: A dedicated artifact to keep agent workflows focused and prevent misinterpretation.
❌ Don't: forget to start repositories and frequent commits, make agents save stable commands.
Vector Databases & Breadcrumbs:
Tools like Osmosis leverage vector databases to rapidly generate insights. However, in writing code and running autonomous agents, we still need extra “breadcrumbs”—think checkpoints created by Cursor AI or code markers like // MARK: - Error Definitions—so future agents have a clear path to follow.
The Rotating Cast of Agents Is Unpredictable:
Agents can forget, overwrite, re-organize, and misinterpret instructions. Even the most advanced models (from a low-cost DeepSeek to a premium OpenAI solution) require a human “super-orchestrator” to direct, intervene, and reject flawed outputs.
This Is Just the Beginning:
This iOS/Swift experiment served as a testbed to sharpen my skills before building more complex, LangChain-based applications in eCommerce. It’s a stepping stone that even justified the $99 I paid to Apple.
Looking Ahead:
Inspired by modular architectures like LangChain, I’m now exploring frameworks such as LangGraph. Imagine the next time Cursor’s agents misinterpret my Swift code and trigger npx create-next-app@latest—the promise of enhanced orchestration through robust inter-agent modeling is tantalizing.
If you’re working at the intersection of LangChain and eCommerce, an Engineering leader or just curious about how AI-assisted development can evolve traditional teams—let’s connect. I’m excited to share how human creativity, clear process documentation, and emerging frameworks are key to unlocking AI’s true potential.
Update February 12: If this area interests you, Langchain just posted their impressions of a single agent benchmark over time - As things stand right now, agents in general see degradation over time, highlighting roles of humans to keep things fresh
To see a long version of this post see my Substack where I will continue to post my journey.