Nacho CollI've been shipping software for over 25 years. I've survived the waterfall-to-agile transition, the...
I've been shipping software for over 25 years. I've survived the waterfall-to-agile transition, the monolith-to-microservices scramble, and the cloud migration era. But what's happening right now with AI-assisted coding feels different — and not entirely in a good way.
Let me show you three things I've witnessed in the last six months. Each one is real. Each one is becoming common.
A team uses Cursor to build a new authentication feature in 3 days instead of 3 weeks. The demo is flawless. Sprint review: applause. Two sprints later, a completely unrelated AI-generated change silently breaks the auth flow. Nobody knows why. The AI that wrote the original feature is stateless — it has no memory of what it decided or why. The team spends 4 days debugging code nobody fully wrote and nobody fully understands.
A 400-line PR lands in review. AI-generated. Fast. Clean-looking. The junior developer approving it hasn't written half of those patterns before. They approve it anyway — it passed the linter and the tests. Six weeks later it's in production and it's a security hole. The velocity was real. The understanding was not.
Sprint 1: an AI agent implements a data model decision — polymorphic associations over separate tables — for good reasons documented nowhere. Sprint 4: a different AI agent (same context window, fresh session) refactors the same area and quietly reverses the decision. Now you have both patterns in production. The codebase has architectural schizophrenia and no human made either call.
I use AI coding tools every day. They are genuinely transformative. The problem isn't the tools — it's that we're using them without a framework for long-term sustainability. We have Agile for human teams. We have no equivalent operating model for AI-assisted teams working on real business software over time.
That's why I wrote the Agile Vibe Coding (AVC) MANIFESTO. It's an open framework built on top of the original Agile Manifesto, extending its philosophy to environments where AI systems participate in producing and evolving code.
The MANIFESTO centers on four values (in the spirit of the original Agile Manifesto):
Full principles at agilevibecoding.org — also on GitHub where you can contribute.
This is a living document. The original Agile Manifesto was written by 17 practitioners in a ski lodge over two days. I don't have a ski lodge — but I have this community. I'm building the AVC MANIFESTO in the open, with real practitioners, based on real pain points.
Here's what I need from you:
→ Have you hit any of these three walls? Tell me in the comments.
→ Do you think a principle is missing or wrong? Open an issue on GitHub — I'll credit contributors in v1.0.
→ Does a principle feel overreaching? Tell me why. I want the hard pushback.
→ Want to be listed as a co-author of MANIFESTO v1.0? Star the repo and open an issue with your contribution.
In my next post I'll be launching OpenAgile.ai — the tool I built to implement these principles in a real sprint workflow. But the Manifesto comes first. Tools without principles are just more noise.