Eight days, about thirty dollars, and I had a working CRM that real people could use. That's the part everybody wants to hear, so there it is up front.

Here's the part that actually matters: building it turned out to be roughly thirty percent of the job.

Can you build software with Claude AI without being a developer?

Partly. I built a working, tested CRM in about eight days for roughly thirty dollars, but I have a technical background and I verified what the model told me against other sources. Without both of those the timeline changes. The build is also only about thirty percent of shipping a product.

Eight days, thirty dollars, one working application

It's called Chumley. It's a deliberately simplified CRM for sales reps who aren't technical, which is a group every CRM on the market claims to serve and none of them actually do.

The problem I was solving is one I'd watched for years. A rep running jobs out of a truck signs up for a CRM, hits the eleventh required field before they can write down a lead, and goes back to a notebook. Every time. The software wasn't too weak, it was too much.

So Chumley is one screen. Deals are cards. A lead goes from nothing to on the board in about ten seconds. There's nothing to configure and nothing to learn.

Fully functional. Tested. Merchant services connected, so it can take money. Completely set up to add users.

Eight days.

The two things that made eight days possible

I want to be careful here, because this is where most articles like this one start lying to you.

I already knew enough to know what to check

I have a pretty technical background. Not a computer science degree, but decades of being close enough to the work to read what came back and have an opinion about it.

If you don't have that, your number isn't eight days. I don't know what your number is, but it isn't mine, and anyone telling you otherwise is selling something.

I never took its word for it

This is the transferable part, and it's the one I'd actually want an owner to take away.

I have no fear of asking questions, and no fear of going and checking the answer somewhere else. When the model told me something, I confirmed it against other resources before I built on top of it.

Most people do one of two things instead. They trust it completely, which is how you end up with a confident, broken application. Or they don't ask at all, because they've decided the whole thing is hype.

Neither works. The useful posture is in between: ask it everything, believe none of it until you've checked.

What Claude AI for small business is good at, and what it isn't

Good at: drafting the whole thing, explaining what it just did in language you can follow, holding a long build in its head across days, and doing the boring wiring nobody wants to do.

Not good at: knowing what your users actually need. It doesn't know that a rep in a truck will abandon the form at field eleven. I knew that, and I knew it because I've spent years around people who run jobs out of trucks.

Also not good at telling you when it's wrong. It's wrong with exactly the same confidence it's right with. That's the whole reason the checking habit isn't optional.

Then the real work started

Chumley has a few users. None of them are real.

It was mostly a lesson for me in how to build an app and do it the right way, all the way through to connecting merchant services. It works. It's ready.

And now the hard part starts, which is sales and marketing. That's probably more like seventy percent of the equation.

Eight days and thirty dollars bought me the thirty percent. The rest is finding the reps who need it, getting in front of them, giving them a reason to switch off a notebook they're comfortable with, and getting them to pay. None of that got cheaper this year. None of it has an eight day version.

What this means if you're about to pay someone $40K

Build the thing. Seriously. The economics changed and they changed in your favor, and if you've been sitting on an idea because the quote was forty thousand dollars, that number deserves a second look.

Then budget for the part that isn't the build, because that's where the money and the time actually go. The software was never the hard part. It's less of the hard part now than it has ever been.

If you want the wider version of this argument, AI for small business covers where it belongs and where it doesn't. If the question is which model, the honest answer is that it matters less than you'd think. And if what you actually need is somebody to work out where AI fits across the whole business rather than just marketing, that's a fractional AI advisor, or this explains the role first.

Go price your idea again. The build quote you got last year is probably wrong now.

  • A working, tested CRM took about eight days and roughly thirty dollars to build with Claude.
  • Two things made that possible: an existing technical background, and verifying every answer against other sources instead of trusting it.
  • The model is wrong with the same confidence it is right, which is why the checking habit is not optional.
  • Building the product was about thirty percent of the job. Sales and marketing is the other seventy, and none of that got cheaper.