AI-assisted development works. But it works differently depending on whether you're building something new or evolving something that already has users, integrations, and history. Most teams don't realize they need different playbooks until the wrong one costs them a sprint.

In this hands-on session, we'll work through both sides using real prompts, real codebases, and real workflows. I build solo greenfield products at night and lead enterprise brownfield engineering teams by day, so we'll pull examples from both. The goal is not to crown a winner, but to show why each requires its own discipline.

We'll work through:

• Side-by-side prompting: the same feature request prompted two ways, one greenfield and one brownfield, and why the brownfield version starts with analysis instead of generation.
• Context debt: how to surface the implicit decisions in a legacy codebase before the AI overwrites them, using architecture documentation prompts you can take back to your team Monday.
• Local correctness, global incompatibility: the brownfield failure mode where the code is clean but the system breaks, and the review checklist that catches it.
• Greenfield verification: the discipline greenfield builders need when there's no existing system to catch their mistakes for them.

Both approaches have real strengths and real blind spots. You'll leave with practical workflows for each and the judgment to know which one you're in..