About a third of organizations told McKinsey this year that they skipped buying at least one piece of software because AI coding agents could build it instead. That's self-reported, and "at least one" could mean a single internal dashboard. The follow-up nobody asked is the one I care about: how many of those tools are still running, and who gets woken up at 3 AM when the damn thing falls over?
Writing the code was always the cheap part. The expensive part is the years after: security patches, upgrades, outages. One large GitHub study found that autonomous agents raised code changes about 240%, while finished releases rose only 30%. Agents sped up writing far more than shipping.
So maybe the build-versus-buy line moves. Owning code gets nearly free. What you'd still pay a vendor for is someone else carrying the pager, the alert that drags a human out of bed.
Before you skip the purchase:
- Name the person on call. One human gets the alert. "The team" is not a name.
- Price year two. Hosting, upgrades, security patches and the occasional outage all add up. The build is usually the smallest cost.
- Rebuild only what you use. Copying the features you need costs far less than cloning the whole product (Malo, Yu & Qiu).
- Check whether AI runs inside it. If the tool calls an AI model every time someone uses it, that bill never ends.
- Be wary with official records. Small tools with one clear job favor building. Systems that hold the company's official data, especially regulated data, favor buying.
- Split it. Build the thin edges and buy the core.
- Bring it to renewal. Proof that you could build it gives you leverage with the vendor.