AI made the first draft of custom software nearly free. It did nothing to the cost of owning that software for the next three years, which is why the build versus buy decision looks different but isn't actually simpler.
The Details:
The cost that used to make buying obvious was never the code itself. It was maintenance, security patches, the engineer who understood the auth layer leaving, and edge cases nobody documented. AI collapses the time to a working prototype but leaves every one of those downstream costs untouched, so a cheap weekend build can still turn into an expensive year two.
Buying carries its own hidden debt too. Average SaaS spend per employee rose nearly 22 percent in a year, and organizations waste tens of millions annually on licenses nobody uses. The cost doesn't vanish with either path; it just shows up as an invoice instead of a support ticket.
Vendor products carry scar tissue, years of edge cases already fought through. AI-generated code skips that scarring, which is exactly why week two of a new build is where things break. Studies show developers often feel faster with AI assistance while independent trials find them measurably slower, because speed without planning just feels like progress.
The smarter split is by layer, not by project. Commodity systems like payroll, billing, and authentication still favor buying, since vendors have already absorbed the painful edge cases. The custom layer that pays off is the glue between systems, the workflows currently stuck in spreadsheets because no vendor product ever fit them.
Bottom Line: The real question isn't whether you can build it anymore; it's whether you can sustain owning it.
Enjoy this article?
Listen to the Claude Code Conversations radio show or join the community.
