Every AI decision eventually comes down to three options: buy something off the shelf, build something custom, or wait until the moment is right. Teams get into trouble when they reach for "build" by default, because building feels serious and impressive. Often it's the slowest, most expensive way to solve a problem that already has a tool.

Here's a cleaner way to choose.

Buy when the problem is common

If your problem looks like a lot of other companies' problems, someone has probably already built a good solution for it. Standard support chat, scheduling, transcription, common integrations, these are solved categories. Buying gets you ninety percent of the value in days, not months, and someone else maintains it. The temptation to build a "slightly better" version of an existing tool almost never pays off.

Build when the edge is yours

Building makes sense when the value comes from something only you have: your data, your specific workflow, your way of doing things. An agent built around your exact process, wired into your exact tools, is something no off-the-shelf product can match, because no off-the-shelf product knows your business. That's where a custom build earns its cost.

The test is simple. If the advantage comes from your own data and process, build. If it comes from features anyone could buy, don't.

Wait when the cost of being early is high

Sometimes the honest answer is "not yet." If the data isn't ready, the workflow keeps changing, or the team can't yet support a new system, waiting a quarter is a strategy, not a failure. Building on an unstable foundation just means rebuilding later. Waiting until the ground is firm is often the cheapest decision you'll make.

Buy what's common, build what's yours, and wait when the timing is wrong. Most expensive AI mistakes come from picking the wrong one of those three, not from picking the wrong model.