Why the build-vs-buy decision is drastically different now

Last week I posted a comic along with a hot take that ended up getting nearly 50,000 impressions in 24 hours (that's a lot for me).

The comic (below) is not actually an xkcd comic, but is meant to look like one. Someone modified a version of an xkcd comic from 2013 that was originally about whether the time spent automating a recurring task is worth the time it saves.

That original one really hit home for me, but this version does not.

Like many xkcd comics, I assume this was meant to be a bit hyperbolic, but this comic has been passed around the internet lately as evidence that building things with AI is inherently much more expensive than buying off-the-shelf software.

And that's just not true.

In just the last 8 months, we have built 6 incredibly complex SaaS products using AI. We use Spec-Driven Development, so I wouldn't exactly call our method "vibe coding" but it's not that far off.

Every one of them was built for tens of thousands of dollars in tokens, not hundreds of thousands, and certainly not the $5 million depicted in this comic! 🤯

On top of that, not a single one of them requires the exponential "maintenance & operations" implied here. In fact, most of our products require almost no maintenance at all (many people find this claim hard to believe, but I'm serious — I'll write more about that in the future).

I do think that if you are trying to build complex software using AI tools with absolutely no software engineering experience, you're probably going to have a bad time.

But we're really good at building software, with or without AI.

And with AI, we are building things better, faster, and cheaper.

So the answer to the question our clients have been asking for decades now is changing.

Should we build it or buy it?

Enterprise SaaS products are incredibly complicated (and costly) because they have to satisfy thousands of customers who have just as many different business processes and workflows.

That $100k per year platform often includes 100 features when you only need 12 of them.

And if something in that platform doesn't work the way your business needs? Good luck getting them to change it.

So you end up adapting your business processes to fit the software, not the other way around.

Why can't we just build this for ourselves?

Many teams we've worked with over the years have asked this question, but the answer usually came down to cost. You had to hire an expensive agency to build a complicated software platform that took 9-12 months to build and cost a million dollars.

But things are different now.

AI hasn't made “build” the right answer every time, but it has shifted the build-vs-buy line drastically. Building custom software used to be a six- or seven-figure project. Today, a much smaller team can build something credible for a fraction of that cost.

I'm not advocating for replacing everything with a custom-built version. Commodity software like payroll, email, accounting, and credit card processing is still going to be much cheaper to buy off-the-shelf.

But if your business advantage is a unique sales process, operations workflow, or something that is really difficult to shoehorn into a SaaS platform, the answer to that question might be much different now than it was two or three years ago.

What about ongoing support and maintenance?

AI is making that easier and cheaper, too. Debugging, refactoring, upgrading, and onboarding into unfamiliar codebases are way faster now.

I don't know what the people spreading this comic around are peddling, but saying that building software with AI is inherently more expensive is a wildly misleading take on the state of things today.

Cheers,

Ryan

PS. We're hiring a Growth Marketer! If you know anyone who is an experienced SaaS marketer, please forward them my contact info—I'd love to talk with them!