Build to find out.

I want to explore what comes next by making things. A working experiment gives a question something concrete to push against. It can expose a possibility, a limit or an assumption I need to change.

Start with a question

Could an agent complete a whole task with a clear point for human review? Could an interface present only what a person needs for the job at hand? Could a tool help someone investigate a problem outside their usual expertise?

I would choose one question and a small, realistic task. Before building, I would write down what a useful result looks like and what evidence would make me reconsider the idea.

Try the part that might be different

An agent experiment might prepare a report from permitted sample data, identify gaps and stop for review before sending anything. The question would be whether it can handle uncertainty and hand back a result someone can trust.

An interface experiment might offer different controls for two versions of the same task. I would want to see whether that adaptation helps someone finish, or simply makes the tool harder to learn. These are examples of experiments to try, rather than results I’m claiming.

Keep the awkward evidence

Record corrections, interruptions, supervision and failed attempts alongside the successful output. How much effort did the result need? What happened when the input changed? Could someone else repeat it?

Experience helps me choose those questions. AI helps me attempt more answers. I want both the enthusiasm of a first demonstration and the patience to find out what survives a second use.

Carry the learning forward

The apps on this site are early work from that exploration. Future experiments may take other forms: an automation, an adaptable tool or a different way to organise work. The useful thread is what each attempt teaches about the next possibility.

Build something small. Observe it carefully. Change the question when the evidence asks for it. That is how I want to explore what comes next.

— Vince