What does an AI-powered dev workflow look like in practice — not in theory, but day to day, from a real senior engineer?

In this episode, I sit down with Brandon Romano, Senior Staff Engineer at Clerk, to walk through exactly how he uses AI to write code at a professional level. We go from first principles — how he started with a very short leash on AI tools and gradually built up trust — all the way to building a full Go backend API and a Next.js frontend from scratch in a single session.

🔍 WHAT WE COVER
→ Brandon's AI journey: from GitHub Copilot skeptic to near-full delegation
→ How to use Cursor effectively with Opus 4.5 for real engineering tasks
→ Why your database schema is your most important AI context signal
→ Using architecture.md and agents.md to give AI the right guardrails
→ Plan mode vs. agent mode — and when to use each
→ Writing PRDs as living artifacts to guide AI through complex features
→ Live CodeRabbit AI code review catching real bugs on a real PR
→ Building a full Next.js frontend, vibe-coded end to end

Whether you're just starting to incorporate AI into your workflow or you're a senior dev looking to level up your agentic coding setup, this one is packed with practical takeaways.

⏱️ TIMESTAMPS
00:00 — Intro & meet Brandon
05:00 — Brandon's AI journey (Copilot skeptic → full delegation)
09:00 — How to give AI the right context (rules, MCPs, guardrails)
17:00 — Live demo: building a Go backend API with Cursor
22:00 — Starting small: one function at a time
28:00 — Scaling up: building all remaining endpoints in plan mode
45:00 — PRDs as artifacts and the future of AI-native codebases
51:00 — CodeRabbit live review: catching real bugs in the PR
01:10 — Vibe-coding a Next.js frontend end to end
01:32 — Wrap-up and key takeaways

#AIcoding #CursorAI #SoftwareEngineering #NextJS #ClerkDev #CodeRabbit #AIDevelopment #WebDevelopment

Hamed Bahram

A senior engineer at Clerk walks through his real-world AI coding workflow, covering how he evolved from skepticism about GitHub Copilot to daily reliance on Cursor and Claude Opus 4.5. Key practices include starting with tightly scoped prompts and gradually increasing AI responsibility, providing rich context via architecture.md and agents.md files, always writing automated tests to validate AI output, using plan mode for complex tasks, and leveraging AI code review tools like CodeRabbit. A live demo builds a Go/Postgres cat cafe management API end-to-end, illustrating how database schema design remains a human-driven foundation while API boilerplate is delegated to AI. The conversation also covers model selection (preferring Opus 4.5 over 4.6 based on hallucination experience), using PRD files as planning artifacts, and the accelerating pace of change in AI tooling.

How a senior engineer at Clerk actually uses AI to code