The App Isn’t the Experiment. Distribution Is.

AI has made building software radically easier. We know this because there's a million posts online these days telling us 'how I built this with AI'.

No-code started the shift. Vibe coding accelerated it. More people than ever can now turn an idea into a working product.

But there’s still a significant difference between building an app and building a business.

Once the app exists, you still need customers.

That’s the part I’ve decided to experiment with.

My access to AI is heavily restricted in my day job because of security policies, so I wanted a practical way to learn more about agentic workflows.

Instead of using agents primarily to build an app, I wanted to see whether they could help with the harder work around it, particularly content, marketing and distribution.

Half Nine is my test vehicle.

It’s a deliberately lightweight habit tracker that sends three questions through WhatsApp every evening at 9:30pm.

But the app isn’t really the experiment. The system around it is.

I’m building an agentic workflow that can research content ideas, create and manage a content catalogue, analyse what performs and recommend the next experiments to run.

I started with:

• Zero followers
• Zero existing content
• No paid advertising
• No asking friends or family to share it
• Instagram as the initial distribution channel

It’s a genuine cold start.

I’m currently two weeks in. The numbers are still small, as you’d expect, but the signals are beginning to improve as I test different formats, messages and publishing times.

Every Sunday, the agents email me a report explaining what happened, what they learned and what I should do next.

My constraint is that I’ll spend no more than one hour each Sunday carrying out those actions. Over time, if the recommendations prove reliable, I’ll allow the agents to take on more of the execution themselves.

I’m giving the experiment until the end of the year.

If it works, I’ll have learned whether agents can do more than help build software. They can help build distribution around it.

If it doesn’t, I should at least have a useful record of where agentic workflows still fall short.

Either way, I’ll share the numbers and what I learn along the way.