I have ideas.
That has never really been the problem.
The problem is what happens after the idea.
If you work inside a larger organization and have an idea for a new product, application, process or service, there are people you can talk to. You might sit down with a business analyst, developer, architect, project manager or another member of your team. You bounce the idea around. Someone asks questions. Someone tells you what you have forgotten. Eventually, an idea starts becoming something that can actually be delivered.
But what happens when you don’t have that team?
That was the situation I found myself in.
I had ideas, experience delivering technology projects, and enough technical knowledge to understand many of the possibilities. What I didn’t have was a business or development team sitting around waiting to explore those ideas with me.
Then I discovered AI chat.
It started with conversation
Initially, I treated tools like ChatGPT much like most people probably do. Ask a question. Get an answer.
But the more I used conversational AI, the more interesting it became.
Instead of asking for answers, I started bringing it ideas.
“What if we did this?”
“How would this work?”
“What am I missing?”
“Is there a better way?”
Those conversations were much more valuable.
An idea that might have stayed in my head could now be discussed. ChatGPT could challenge an assumption, identify a missing requirement, suggest alternatives or help me explain what I was actually trying to accomplish.
Over time, I stopped thinking of ChatGPT primarily as a search or question-and-answer tool.
It started functioning much more like my business analyst.
From an idea to a requirement
There is a big difference between having an idea and having something a developer can build.
I might know generally what I want, but that doesn’t mean I have thought through how it should behave, what happens when something fails, what should be tested, what shouldn’t change, or what “done” actually means.
That is where much of my collaboration with ChatGPT now happens.
I bring the idea.
We work through it together.
Sometimes the original idea survives mostly intact. Sometimes the conversation exposes a better approach. Sometimes we discover that I am actually trying to solve a different problem.
Eventually, the conversation becomes specific enough that we can define the requirement.
At that point, something interesting happens.
I can hand it off.
Then the project system takes over
As I experimented more with AI, I started building a system around the conversations.
That system is becoming BrinnOS.
BrinnOS is intended to become my project assistant: keeping track of projects, priorities, tasks, decisions, status and next actions.
There is also Brinn, the assistant layer I interact with through that environment.
Not every idea needs to become a development project. Some need planning. Some need research. Some need to become tasks. Some simply need to be remembered until later.
The goal is for BrinnOS to help manage that layer.
When something does require software development, another AI agent enters the process.
Hermes.
My developer is an AI agent
Hermes works directly in my development environment.
Once we have defined what needs to be built, Hermes can inspect a repository, modify code, run builds, execute tests and report the results.
That changes the economics of experimenting with an idea considerably.
I don’t need to take a requirement and find a developer before I can find out whether it works.
I can move from conversation to implementation surprisingly quickly.
But I have also learned that speed creates its own problems.
An AI agent can misunderstand a requirement. It can make assumptions. Providers can hit rate limits. Context can become enormous. A fallback model can behave differently from the primary model. Something can technically pass a test and still not be what I wanted.
So the process cannot simply be:
Idea → AI → Production
There need to be checkpoints.
I still lead the project
This is the part I think gets lost in many conversations about AI replacing jobs.
AI can perform a surprising amount of work.
That does not mean I give it the project.
I still decide what we are trying to accomplish.
I decide what matters and what doesn’t.
I decide whether the requirement reflects what I actually want.
I decide whether the result is acceptable.
And I decide whether something goes into production.
A recent change to the What Up Inc website was a good example.
ChatGPT and I worked through the idea of adding an Insights section. We defined how it should work and what needed to be verified. Hermes implemented the feature, built the site and ran the tests.
Then I looked at it.
One small feature offered visitors the ability to subscribe through RSS. Technically, it worked perfectly. Clicking it displayed the RSS XML exactly as it should.
I didn’t like the experience.
So we removed the visible subscription link while keeping the RSS capability behind the scenes.
Later, when the release was ready, Hermes was prevented by its configuration from pushing the code to production.
I pushed it myself.
Then I opened the production website and looked at the result.
That might sound like a small distinction, but I think it is an important one.
The AI did a lot of the work.
I remained accountable for the outcome.
The workflow that is emerging
I didn’t design this operating model on a whiteboard and then implement it.
It emerged through use.
Today it looks roughly like this:
Idea → Conversation → Definition → Planning → Development → Testing → Review → Decision
I provide the idea and direction.
ChatGPT helps me explore it and turn it into something defined.
Brinn and BrinnOS are becoming the project-management layer that keeps track of the work and what needs to happen next.
Hermes performs much of the technical implementation and testing.
Then the result comes back to me.
Sometimes I approve it.
Sometimes I change my mind.
Sometimes I see something on an iPhone that no automated test told me was wrong.
And sometimes I decide we should simply stop and leave something alone for a while.
That is still project leadership.
AI didn’t give me the ideas
One of the most important things I have learned from this experiment is that AI hasn’t replaced the part I originally brought to the table.
It didn’t give me the ideas.
What it gave me was a way to work through those ideas when there wasn’t a team sitting beside me to work through them with.
That is a very different proposition.
I now have access to capabilities that resemble pieces of a delivery organization: analysis, planning, development, testing and project assistance.
They aren’t people, and pretending they are would miss the point.
They are capabilities.
Learning how to combine those capabilities, where to trust them, where to verify them and where to retain human control is becoming a skill in its own right.
I am still learning that skill.
The tools will change. The models will improve. BrinnOS will evolve. The workflow I use six months from now will probably look different from the one I use today.
But I suspect one principle will remain:
AI can dramatically expand what one person is capable of delivering. It doesn’t remove the need for someone to decide what is worth delivering in the first place.
My BA is ChatGPT.
My developer is an AI agent.
And I still lead the project.