There’s something fascinating happening in the world of AI-augmented development right now, and it reminds me of an old debate that I think we’ve been interpreting wrong.
Since Anthropic released their articles on Model Context Protocol (MCP), token optimization strategies, and agentic workflows, we’ve seen an explosion of methodologies. Everyone’s publishing their approach—their systematic way of structuring prompts, organizing code generation, architecting with Claude. And I keep seeing the same pattern: everyone wants to be the cathedral.
But here’s what struck me: we’re not actually heading toward a cathedral vs. bazaar world. We’re heading toward something much more interesting—something that looks a lot like coaching trees in football.
The Cathedral That Never Was
The original cathedral and the bazaar framing was about open vs. closed development models. But what we’re seeing now isn’t a battle between monolithic control and chaotic collaboration. It’s the emergence of distinct lineages—methodologies that get forked, adapted, and refined as they pass from developer to developer.
Think about it: you have the foundational models at the top—Claude, OpenAI, the big players establishing the fundamental capabilities. Then you have the first generation below them: lead developers and solo practitioners who’ve spent months developing their own systematic approaches. They’re working with real clients, solving real problems, and crystallizing their learnings into repos full of prompts and architectural patterns.
These aren’t just random collections of code. They’re philosophies materialized in markdown. They’re systems of thought about how to structure an AI-augmented development workflow.
The Disciples and the Clone Army
Here’s where it gets really interesting: late one night, a junior developer or an entry-level engineer hits a wall. They’re trying to figure out how to architect something, how to structure their prompts, how to think about the problem. They stumble onto a repo that speaks exactly to their situation. They clone it. They run with it.
That developer just became a disciple of a methodology they don’t even realize they’ve adopted. They’re now part of a lineage—influenced by the internal workflow practices of some company they’ve never heard of, shaped by constraints and priorities they don’t understand, carrying forward patterns that emerged from contexts completely foreign to them.
And you know what? That’s not a bad thing.
Why Proliferation is a Feature, Not a Bug
In football, we don’t think it’s a problem that there are multiple coaching trees. We don’t worry that some coaches came up through the Bill Parcells tree while others descended from Don Coryell or Norv Turner. Each tree represents a systematic approach to the game—offensive and defensive philosophies that get tweaked and refined with each iteration. Each disciple adds their own twist, adapts to their personnel, modifies based on their context.
The West Coast Offense didn’t kill football. It became one option among many, and the game got richer for it.
That’s what’s happening now in AI-augmented development. We’re in the early days of methodology formation. The solo developers and team leads publishing their repos aren’t competitors in a zero-sum game—they’re the founders of coaching trees.
The Market Impact
The immediate effect is already visible: the floor is rising. Junior developers with access to well-structured methodologies can suddenly punch above their weight class. The “copycat crowd”—and I mean that descriptively, not pejoratively—can deliver better results than they could six months ago.
This creates an interesting dynamic. The value isn’t just in knowing how to code anymore. It’s in understanding why a particular methodology works, when to apply it, when to deviate from it. It’s in having the experience to know which coaching tree’s playbook fits your current situation.
The commodity isn’t the methodology itself—it’s freely available on GitHub. The value is in the judgment to select, adapt, and evolve it.
The Trees Are Growing
We’re watching the real-time formation of schools of thought around AI-augmented development. Some will emphasize token efficiency. Others will prioritize code readability. Some will be opinionated about architecture, while others will be flexible about structure but rigid about prompting patterns.
These aren’t competing visions of the one true way. They’re different answers to different questions, shaped by different contexts, optimized for different constraints.
The developers who thrive won’t be the ones who found the one true methodology. They’ll be the ones who understand the lineages, who can trace the reasoning behind different approaches, who know when to be orthodox and when to innovate.
Not Cathedral, Not Bazaar—Ecosystem
What we’re building isn’t a cathedral or a bazaar. It’s an ecosystem of practice. It’s messy and it’s beautiful and it’s exactly what should be happening.
So publish your repos. Share your systematic approaches. Start your coaching tree. Because somewhere out there, a developer is going to find your methodology at 2 AM and it’s going to be exactly what they need. They’ll clone it, they’ll use it, they’ll modify it, and eventually they’ll teach it to someone else.
And that’s how the craft evolves.
What coaching tree are you part of? Have you found a methodology or workflow pattern that fundamentally changed how you work with AI? I’d love to hear about the approaches that have shaped your practice.
