We May Be Building AI Development Tools Backwards

Most AI coding tools are optimized around a conversation.

I think that’s a mistake.




⸻




The Industry Assumption

Current workflow:

• Conversation


• ↓


• Context Window


• ↓


• Reasoning


• ↓


• Code

Most innovation focuses on:

1. larger context windows


2. better reasoning


3. better agents

The assumption is that improving the model improves development.




⸻




What Actually Happens

Real projects don’t exist inside conversations.

They exist inside repositories.

The repository persists.

The conversation does not.

Yet almost every AI workflow is built around the conversation as the source of truth.




⸻




A Different Mental Model

What if the project became the source of truth?

Instead of:

• Conversation → Understanding

Use:

• Project → Understanding

The AI session becomes temporary.

The project cognition becomes persistent.




⸻




Why Git Was Revolutionary

Git separated code history from developers.

History remained even when people left.

AI development may need a similar separation.

Project understanding should remain even when:

models change sessions end tools evolve




⸻




The Real Opportunity

I don’t think the next generation of AI development tools will compete primarily on code generation.

They’ll compete on memory.

Not memory of conversations.

Memory of projects.

The systems that can preserve project cognition over months and years may become more valuable than systems that simply generate code faster.




⸻




This is the direction we’re exploring with Contorium.

Website:

https://www.contorium.dev/

GitHub:

https://github.com/ContoriumLabs/contorium