Build or Buy: The Short Answer

Build AI when it is a core differentiator that your customers pay you for and that no off-the-shelf product can deliver. Buy when the capability is a solved commodity that everyone needs and nobody wins on. The trap is treating this as one big decision. In practice, most businesses should do both: buy the commodity foundation, the models, the vector databases, the platforms, and build only the thin layer on top that is specific to your workflow, your data, and your customers. That is where your advantage actually lives, and it is the only part worth the cost of building.

I am Mahmoud Zalt, an independent AI architect with 16 years building software. Through Sista AI I help founders decide what to build in-house and what to buy off the shelf.

The Real Question Is Not Build vs Buy

Framed as build versus buy, the decision feels binary and high-stakes. It is neither. The useful question is which layer of your AI stack you are talking about, because the answer differs at each layer.

Nobody sensible builds their own foundation model, that is a commodity you rent through an API. Almost nobody builds their own vector database either. But the agent that automates your specific onboarding flow, using your data, your rules, and your systems? No vendor sells that, because it only exists inside your business. Once you split the stack into layers, most of it is an obvious buy, and the build question narrows to the small, valuable part that is unique to you.

The Criteria That Decide It

For any given capability, run it through these questions before committing to build.

The maintenance row is the one teams forget. Building is not a one-time cost. Every custom system needs someone to own it, update it, and fix it when it breaks. If you cannot commit to that ownership, buying is not the weaker choice, it is the responsible one.

The Hidden Costs on Both Sides

Buy looks cheaper because the price is on the invoice. Build looks more powerful because you control it. Both hide costs that decide the real total.

Buying hides integration and lock-in costs: the tool still has to connect to your systems, your data still has to flow into it, and if the vendor raises prices or shuts down, migrating can be expensive. Building hides the long tail: the working demo is a fraction of the effort, and the rest goes into evals, guardrails, observability, and the maintenance that keeps it reliable after launch. The honest comparison is not sticker price versus sticker price, it is total cost of ownership over a few years, including the people who keep each option running. When you account for all of it, the buy-the-base, build-the-edge pattern usually wins because it spends your build budget only where it creates advantage.

Frequently Asked Questions

should a small business build its own AI

Rarely from scratch, and never at the foundation layer. A small business should buy proven tools for commodity needs and reserve custom building for the one workflow that is genuinely unique to it and central to how it competes. Spreading limited resources across custom builds of things you could have bought is how small teams stall.

is it cheaper to build or buy AI

Buying is almost always cheaper to start and to maintain, because the vendor absorbs the engineering and operations. Building is only cheaper over time in narrow cases, such as very high usage volume where per-call pricing adds up, or when a custom capability drives revenue no product can. Compare total cost of ownership over several years, not the upfront number.

what AI should I never build myself

Foundation models, vector databases, and general-purpose infrastructure. These are mature commodities where established providers will always be cheaper and better than anything you could build, and building them adds no advantage. Rent the infrastructure; spend your effort on the layer that is specific to your business.

how do I know if a capability is a differentiator

Ask whether customers would choose you partly because of it and whether a competitor could simply buy the same thing tomorrow. If it is unique to your data, workflow, or customers and cannot be purchased off the shelf, it is a differentiator worth building. If any competitor can buy the identical capability, it is a commodity worth buying.

Buy the Base, Build the Edge

The strongest AI strategy is not maximum building or maximum buying, it is spending your build budget only where it creates advantage and buying everything else. Get that split right and you move faster with less risk, because you are not reinventing commodities or outsourcing the very thing that makes you different.

If you want help drawing that line for your business, and then building the custom edge properly, my Agent Development service covers custom AI applications and agentic systems from architecture to production, starting with a fixed Discovery phase to decide exactly what is worth building.