Why I Use Multiple AI Chats Instead of One Huge Conversation

When people first start using AI for software development, they often treat it like a search engine. They open a single conversation and keep asking every question in the same place.

I started that way too.

For small projects, it works perfectly well.

Large software projects are different.

As my work on the Auspice Darer Framework grew, I realised that one conversation was trying to cover too many different topics at once. One moment I was discussing dependency injection, the next I was redesigning CSS, then debugging routing, before switching to documentation or database design.

The conversation became cluttered, and the AI had to continually switch context.

Eventually I tried something different.

Instead of one conversation, I created a project and gave different chats different responsibilities.

One chat focused entirely on framework architecture.

Another concentrated on CSS and presentation.

Others were dedicated to individual applications, documentation, testing or specific migrations.

Almost immediately the quality of the discussions improved.

Each conversation developed its own context and history. Instead of constantly reminding the AI about previous decisions, those decisions became part of that conversation's working knowledge.

It felt surprisingly similar to working with a development team.

You wouldn't ask your graphic designer to redesign your database schema, or your database administrator to choose your colour palette.

Each person has their own area of expertise.

By giving each AI conversation a clear responsibility, I found the same thing happened. Every chat became increasingly specialised, while I remained responsible for coordinating the overall project.

That last part is important.

Creating specialist conversations doesn't remove the need for a software architect.

It makes one even more important.

Each conversation only sees its own part of the project. Someone still needs to ensure that all of those decisions fit together into a coherent whole.

That's the role the human developer continues to play.

Looking back, I think this was one of the biggest improvements I made to my AI workflow.

I stopped thinking of AI as a single assistant.

Instead, I started thinking of it as a team of specialists, each contributing to one part of a much larger system.

The software still has one architect.

It just has a larger team helping to build it.