The Build/Buy Decision Is Now a Three-Way Door

The build vs. buy decision used to be two doors. Now it's three, and most teams are still walking through the wrong one.

The old question was simple: does this capability exist in the market, or do we build it? Buy if a vendor has it, build if they don't. That framing assumed the two options were the real choices and the middle was a compromise.

AI broke that. The middle is no longer a compromise. It's the default, and it's often the right answer.

Why the two-door frame fails now

The two-door frame assumed the thing you were deciding about was stable. Software used to sit still long enough to evaluate it — pick the package, install it, and it worked the same way next year. AI doesn't sit still. The model changes every quarter. The vendor's roadmap moves. The capability you evaluated in January is a different thing by June.

So the classic buy decision, “the vendor has it, buy it,” now has a hole in it. What the vendor has is a model, a set of prompts, and a service around them. What you need is a capability inside your business, with your data, your workflows, your review process, and your constraints. The gap between what they sell and what you need is where the middle lives.

And the classic build decision has a hole in it too. The raw model you can license is a commodity. The differentiation isn't in the model — it's in the data, the evaluation, the integration, the operating layer. Building “AI” from scratch when the model itself is a commodity is building the wrong thing.

The question isn't whether to build or buy the model. It's what to build, what to buy, and what to assemble around the part you license.

The three-way frame: capability, control, speed

The frame that works has three questions, and each one points at a different answer.

Capability: does this need to be a differentiator? If the capability is table stakes, like a summarizer, a chatbot, or a classification layer, buy it. Nobody wins by hand-building a commodity. If the capability is the thing your business is built on, meaning the core workflow that makes you money, you need to own more of it, because a differentiator you rent is a differentiator you don't have.

Control: what happens when it matters? If the capability touches your data, your customers, or your compliance obligations, you need control over the parts that touch those things — the prompts, the evaluation, the escalation, the audit trail. Control isn't an all-or-nothing state. It's a list of the parts you can't delegate.

Speed: how fast do you need it, and how fast does the market move? If a vendor can get you to value in weeks and building takes a year, the speed question has already answered itself. If the market is moving so fast that the vendor's roadmap is your roadmap, buying means you're permanently riding someone else's schedule — which is fine until the schedule matters.

The hybrid pattern that actually works

Here's the pattern I keep seeing in the organizations that get this right. It's the same shape every time.

They license the model, the raw intelligence, from someone who does that better than they ever could. That's the buy. Then they build the layer around it: the data pipeline, the evaluation set, the escalation paths, the human review, the integration into their actual workflow. That's the build. And they assemble the third layer, the operating model: who reviews, who decides, how quality is measured. It's neither built nor bought. It's managed.

The model is a commodity. The layer around it is the product. The operating model is the moat.

That's why hybrid isn't a compromise anymore. The two-door frame treated the middle as “a little of both, done worse.” The three-way frame treats it as the only answer that respects what actually changed: the intelligence became cheap, and the work around the intelligence became the entire game.

How to make the call

Walk your next AI decision through the three questions and watch what falls out.

If the capability is commodity and speed matters, buy it and move on. You don't need a committee for this one. If the capability is a differentiator, plan to own the layer around a licensed model — and start the build now, because the integration is always slower than the demo. If control matters because of data or compliance, write down the parts you can't delegate before you sign anything, and let that list drive the architecture.

The three-way door isn't harder than the two-way one. It's just honest. The old question pretended the choice was between two clean options. The new question admits what's true: you're going to buy the brain, build the body, and operate the whole thing. The only real decision is where each line gets drawn, and the three questions draw it for you.

Build what differentiates. Buy what doesn't. Assemble the rest. That's the whole playbook, and it's the same answer whether you're evaluating a chatbot, a copilot, or an entire platform. The door in the middle isn't the compromise anymore. It's the way through.

Think this argument fits your event? Tell me about the room — the calendar is selective.