Trying to Work Out GPTs vs Projects? Here’s What Finally Made It Click for Me

I learned the difference between GPTs and Projects the hard way, right in the middle of client work, when my system started to feel heavier than the work itself. At the time, I was designing a strategy for a client that involved a lot of complex information, competing priorities, and evolving thinking. I needed somewhere to hold all of that without losing coherence, and my first instinct was to use a GPT. I had been experimenting with them and they felt capable, so I assumed they were the right place to store complexity.

I started feeding everything into the GPT. Notes, background context, drafts, reflections, bits of data I thought I would need later. At first it felt productive, but before long I noticed I was constantly restating things, correcting drift, and trying to get the system to remember work it was never meant to hold over time. It wasn’t broken, but it was brittle, and I was compensating for that brittleness without really noticing how much effort it was costing me.

Around the same time, with another client, I was making the opposite mistake. I had set up a Project and was using it to store what was essentially a whole part of my business. Everything went into it. Strategy work, operational thinking, half written ideas, future plans. I treated the Project like a digital version of my head. Over time it became bloated and hard to navigate. The information was there, but the shape was gone, and I realised that memory without structure is not actually that useful.

That was the point where it clicked that I was trying to solve the wrong problem in both cases. I did not need more intelligence or more storage. I needed clearer boundaries.

What changed my approach was learning to define Projects much more tightly. Instead of thinking of a Project as a big bucket, I started treating it as a specific piece of work with a clear purpose. I learned that Projects work best when they are trimmed down into small, intentional chunks. Rather than putting all of Boonee into one Project, I now create Projects around defined areas of focus. A Project might be about better uses of cultural safety methods, or the development of a single framework, or a specific client engagement. Each one has a reason to exist, and when that reason has been fulfilled, the Project is done.

That alone reduced a lot of friction. I stopped losing work inside my own system because the system itself had clearer edges.

At the same time, I stopped trying to use GPTs as containers. I started using them as helpers. The way I now think about GPTs is as staff members who support my thinking, not as places where the work lives. Each GPT has a role. I bring work to it, rather than leaving work inside it.

For example, I might have a GPT that acts as a cultural safety expert. I build it deliberately, with clear expectations around language, care, risk, and responsibility, so that when I ask it to reread my work, it applies that specific lens consistently. Another GPT might be set up as a marketing expert, focused on clarity, tone, and audience fit. Each one is designed to think in a particular way, and that is its job.

The Project, meanwhile, holds the actual work. It contains the drafts, the decisions, the evolution of thinking over time. It gives me continuity. The GPT gives me judgement. Once I separated those roles properly, the system started to feel calmer and more reliable.

I stopped asking GPTs to remember everything, and I stopped letting Projects sprawl. I defined Projects with intention, and I designed GPTs with purpose. The frustration I felt in the early weeks made sense in hindsight. I had been asking the tools to do jobs they were not designed for.

This way of working has been particularly useful as a founder and someone who works for themselves. The biggest risks in that position are rarely technical. They are cognitive. Dropped context. Inconsistent decisions. Carrying too much in your head and rebuilding your thinking more often than you realise. When used well, Projects give you somewhere to return to without losing the thread, and GPTs give you support without blurring accountability.

It took me about a month to really understand this, not because the distinction is complicated, but because my habits were. I had to feel the friction before I could see where the structure was missing. Tools like those built by OpenAI reward clarity more than cleverness. They work best when you design them to support how you already think, rather than asking them to replace it.

Now, when I set up a Project, I am clear about what it is for and what it is not for. When I build a GPT, I am clear about the kind of judgement I want it to apply. That shift has made my system quieter, steadier, and much easier to live with.