AI Strategy
Build or Buy in the AI Era? The Demo Is No Longer the Hard Part
For most of the software era, the build-vs-buy decision was gated by feasibility: building was expensive and slow, so the question filtered itself. AI coding assistants have removed that gate. Almost anything is now buildable in a weekend, at demo quality. That has not made the decision easier; it has quietly broken the way most organisations make it. These are field notes on what we believe the question has become, offered as a thinking tool rather than a rulebook.
The question has changed: from capability to liability
"Can we build it?" used to do real work as a filter. In 2026 the answer is almost always yes, which makes it a worthless question. The question that still does work is: do we want to own it?
Buying a tool is renting a capability. Building one is adopting an operational dependent: it needs maintenance, upgrades, security patching, and it calls someone at 3am. The cost of building was never really the build; it is the ownership that follows. A useful habit in any build-vs-buy meeting: silently translate "should we build X?" into "should we be liable for X?" and watch how the conversation changes.
The second 90%: demos got cheap, production didn't
An old piece of software folklore says the first 90% of a project takes 90% of the time, and the remaining 10% takes the other 90%. AI assistants have made this worse, not better, because they made the visible part nearly free.
The working-looking interface, the convincing prototype, the demo that delights the steering committee. That now takes hours. What has not moved is the invisible remainder: integration with systems of record, edge cases, identity and access, timezone and sync correctness, monitoring, support processes, and compliance evidence. The cheapness of the demo creates an illusion that the project is nearly done precisely when the hard part has not started.
Most enterprise AI disappointment we observe lives in this gap, not in the model, and not in the idea. The leaders who navigate it well treat any demo as the start of the ownership conversation, not the end of the engineering one.
Core or context: the one-breath test
A decades-old distinction does most of the remaining work. Anything that makes your customers choose you is core. Build it, own it, pour your best people into it. Everything else is context. Buy it, rent it, integrate it, regardless of how interesting it would be to build.
The test fits in one breath: will a customer ever choose us because of this? A scheduling tool, an internal dashboard, a CRM, a form backend. Almost never. The system that embodies your differentiated process or product. Almost always.
The trap worth naming is that context is frequently more fun to build than core. Booking widgets, portals and dashboards are satisfying weekend projects; the unglamorous core work is harder. Capable teams drift toward building enjoyable context without anyone deciding to. The drift is invisible in any single sprint and expensive over a year.
When building is the right call
The same razor cuts both ways, and four situations genuinely favour building:
- When the thing is the product. If the capability is what you sell, ownership is the point. Build it.
- When a constraint forbids buying. Data residency, APP 8 cross-border rules, APRA outsourcing expectations: if no vendor can satisfy the constraint, the decision is made for you. This is constraint-driven building, not preference-driven building. The distinction matters in front of a board or a regulator.
- When you need 10% of a tool priced for 100%. A large platform licence to run one workflow is often beaten by a small, focused build.
- When integration costs more than construction. Occasionally, gluing three vendors together is more fragile and more expensive than one modest bespoke service. Rare, but real, and worth checking honestly rather than assuming.
Three questions for your next build-vs-buy meeting
Frameworks are forgotten; questions survive. These three do most of the work of this entire article in a single meeting:
- 1"If this breaks at 2am, who takes the call, and do we want it to be us?" This surfaces the ownership liability that build advocates rarely price in.
- 2"Will a customer ever choose us because of this?" The core-vs-context test, answered in one breath. If the room hesitates, it is context.
- 3"What does version two cost?" Everyone budgets version one. Almost nobody budgets the forever: the upgrades, the security reviews, the person who maintains it after its author leaves.
Key Takeaway
AI has made building cheap; it has not made owning cheap. The build-vs-buy decision now turns almost entirely on ownership: build what differentiates you or what regulation forces you to control, and buy everything else, however buildable it looks. And when the question arrives as "should we build X?", the most valuable move available to a senior leader is to re-ask it first: "should we own X?" The answer is usually already in the room.
Let's talk
Want to explore what AI could do for your business?
A 30-minute conversation about your business and where AI genuinely creates value — and where it doesn't. No pitch deck.