Back to Blog

My AI-Assisted Development Workflow: Shipping Full-Stack Products Faster

November 30, 2025

My AI-Assisted Development Workflow

I use AI tools every day — Cursor, Bolt.new, Lovable, and LLM agents — and they've genuinely changed how much I can ship. But there's a line between using AI as leverage and abdicating to it. Here's the workflow that keeps me fast and honest.

AI and code

Where AI genuinely helps me

  • Boilerplate and scaffolding. Starting a Django app, wiring auth, standing up a Next.js route — the boring 80% that takes hours by hand.
  • Explaining unfamiliar code. Pasted a library you've never seen? A quick summarisation beats reading 500 lines.
  • Fast prototypes. For validating an idea or showing a client a direction, generating a working slice is unbeatable.
  • Tests and fixtures. Generating test cases for edge conditions I haven't thought of yet.

Where I refuse to let AI decide

  • Architecture. The trade-offs between Django and Next.js, where to put business logic, how to model the data — that's my call.
  • Security. AI tools hallucinate "secure" code all the time. Auth, permissions, input handling — I review these line by line, and because I do bug bounty, I even attack AI-generated code like a target.
  • The final review. Every line goes through my brain before it ships. If I can't explain it, it doesn't merge.

My actual loop

  1. Break the feature into small, verifiable slices.
  2. Have the AI draft the slice with explicit constraints (framework, patterns, test requirements).
  3. Review the diff for correctness, security, and style.
  4. Write or verify the tests — happy path and the failure path.
  5. Ship small and iterate.

The pitch

You are helping me build a feature in [stack].
- Follow existing project conventions.
- Do not introduce new dependencies unless I ask.
- Prioritise security (input validation, auth checks, safe queries).
- Show the changed files and explain what each does.

That last line is the kicker: if the tool can't explain it, neither will I — and that's the moment to slow down.

What it changed in my output

Since adopting this workflow, I've shipped production e-commerce, inventory, and ordering systems with far less grind. The key is that AI doesn't replace review — it accelerates the gap between reviews. The moment I treat generated code as reviewable draft instead of finished work, everything got faster and safer at the same time.