Using AI as a sounding board when you build alone

Most of my decisions get made in my own head. On a walk, or at the kitchen table late at night with a coffee I shouldn't be drinking at that hour. There is no co-founder to argue with, no standup where someone finally says the thing I have been avoiding. When you build alone, the hard part isn't the work. It is that every call runs through one brain, and that brain has been awake since six and is quietly attached to whatever idea it had first.

So over the last year I have started using AI as a sounding board, mostly for pushback. I don't ask it what to build or whether to put the price up. I ask it to find the hole in my thinking before I spend three weeks finding it the expensive way.

What a sounding board actually does

A sounding board does one quiet job. It reflects your thinking back at you with enough distance that you can finally see it properly. That is harder to do on your own than people admit. When the idea is yours and you have been carrying it around for a week, you slowly lose the ability to tell the difference between this is good and this is mine.

Using AI as a sounding board works when I treat it like a sharp colleague reading a half-formed plan, not an oracle handing down a verdict. I am after a better question, not a faster answer. Most of the value shows up in the first few minutes, before any advice, when it asks the plain thing I had walked around for days.

The personas I actually use

Luna, the app I am building, leans on this idea, so I have spent a fair bit of time on it. The thing that made it click was personas. Instead of one flat assistant agreeing with everything, I set up a few characters and let each one read the same problem in its own way.

• The sceptic. Its only job is to argue against whatever I want to do, as well as it possibly can. If the strongest version of don't do this still doesn't land, that tells me something useful.
• The annoyed user. The person who downloaded the app, hit one bit of friction, and never opened it again. They don't care about my roadmap. They care that the thing felt slow on a Tuesday.
• The calm one. Less clever, more steady. Good for the nights when the real question isn't is this a good feature, it is why am I panicking about a feature at eleven at night.

Three voices on the same problem get me most of the way to the conversation I would have had with a good team, without the calendar invite. None of them are smarter than me. They are just not me, which on most days is the entire point.

I am not asking the model to be right. The job is to make me less wrong, faster.

Where it genuinely helps

The biggest one is naming the real question. Half the time I sit down sure I am choosing between two features, and a few minutes of being asked what are you actually worried about reveals I am really deciding whether anyone wants the app at all. Better to learn that on a Tuesday morning than after a fortnight of building.

It is also good for the late spiral, the one where a small worry has quietly grown teeth. Saying it out loud to something that won't get bored or worried with me usually shrinks it back to its real size. Half the time the honest reply is that I am tired and the decision can wait until morning, which is exactly what a steadier colleague would have said.

And it catches the thing I am protecting. We all have a favourite idea we defend a beat too quickly. A sounding board that keeps gently asking why notices that flinch before I do.

Where it falls down

It agrees too easily. Ask a model a leading question and it will happily tell you that you are onto something brilliant. It can be confident and wrong in the same calm voice it uses when it is confident and right, and there is no tone change to warn you which one you are getting. Left alone, that turns a sounding board into a very polite mirror.

So I have learned to set it up to push. I give it the strongest case against me and ask it to make that case properly. I ask for the version of events where I made the wrong call, and what I would have missed. I tell it plainly that agreeing with me is the unhelpful answer. It takes about three sentences of setup, and it changes the whole conversation from flattery into something I can use.

Keep it yours

There is a catch with all of this. The half-formed, slightly embarrassing version of your thinking is exactly the stuff you don't want sitting on someone else's server, quietly turning into training data. A sounding board only works if you are honest with it, and it is hard to be honest with something you suspect is keeping a copy.

That is the whole reason Luna's reflections run the way they do. The personas, a therapist, a friend and a coach, work on your entries while everything stays on your device. Your words are private, on-device, with no tracking and no ads, and you can export the lot to Notion or a CSV whenever you like. You own the thinking, not me, and not a company you have never met.

I still make plenty of bad calls. A sounding board doesn't fix that, and it was never going to. What it has changed is how long I sit alone with a question before I get a second angle on it. That used to be days. Now it is about ten minutes, and for a one-person studio that is worth a lot.

Anyway, that is how I have been working lately. If you try it, start with the sceptic. It is the one that earns its keep.

Benjamin

Founder & sole developer, Novaire Digital