For decades, the default for most companies acquiring new software was straightforward: buy it. Building was reserved for the largest enterprises with serious IT budgets, or for problems no vendor could solve. For everyone else, the calculus was simple — a SaaS subscription was faster, cheaper, and lower-risk than asking a team to build something internally. Buy what you can; build only when you must.
That default was rational for a long time. However, it's no longer the only reasonable answer. Quietly, over the past two years, the economics underneath it have shifted — and the organizations that notice first will be the ones best positioned for what comes next.
The buying wall
The case for buying was always speed. License a platform, deploy it across the company, capture quick wins. And that worked — for a while. But the gap between what off-the-shelf software delivers and what a well-designed custom solution can deliver has been widening for years, and AI is widening it further. Generic platforms, applied to the average company's average problems, produce average results. They work everywhere, which is another way of saying they fit nowhere in particular. And when every competitor in your industry has bought the same platform, that platform is no longer a source of advantage — it's table stakes.
The deeper limitation is that vendors design for the median customer — the one they can sell to most easily. Your competitive edge, by definition, is not the median. It's the thing only you do, or the way only you do it. No vendor will ever build for that, because the market for "exactly your operation" is exactly one.
What actually changed
Here is the shift that the headline numbers obscure: the cost of building has collapsed.
Five years ago, building a custom internal application meant a multi-month engagement with an outside vendor, a six-figure budget, and a long wait for something that might or might not fit when it arrived. Today, a small team with the right design instincts and access to modern AI tools can prototype the same thing in weeks. HBR describes the new reality plainly: companies now have the option to build, compose, collaborate, or buy outcomes rather than make do with existing SaaS offerings.
The largest, most sophisticated companies have already moved. McKinsey's internal AI platform, Lilli, is used monthly by 75% of its 40,000 employees. Morgan Stanley's in-house AI has saved its coders 280,000 hours this year by converting legacy code to modern standards. Walmart's Trend-to-Product AI slashes clothing design timelines from six months to six to eight weeks. None of these companies needed to build. They chose to — because what they built fit them, and only them.
The trend is now visible across industries: most organizations begin by adapting open-source algorithms to their proprietary data, and as they mature, evolve toward building custom models that support their differentiated value proposition. The same source notes the limit of the alternative: while embedded AI can provide early value, relying only on third-party tools may limit long-term progress.
Maturity, in other words, leads to build.
The new bottleneck is design
If anyone can now generate code, the question becomes: what should you build?
This is where the conversation shifts in a way most executives haven't yet absorbed. The barrier to implementation has fallen, but the barrier to good design has not. A small, capable team with the right tools can now produce a working prototype in a fraction of the time it used to take. What's far harder — and what doesn't get easier just because AI got better — is identifying the right problem to solve, designing an interaction that people actually want to use, and recognizing when the output is good enough to put in front of real users.
Design is now the scarce resource. Not engineering capacity. Not access to models. The ability to look at a workflow, understand what people are really trying to do, and shape a tool around them — that is where value gets created and where most projects fail.
What this looks like in practice
At WSOM, we've been living this shift. xLab — the experiential learning lab embedded in the school — has been building tools that no vendor was going to sell us, with small student teams, in months rather than years.
One is a lobby kiosk featuring AI avatars of a WSOM faculty members, greeting visitors and answering their questions about the school. Five students built it over a single summer. Visitors often stop to take it in — a small, recurring reminder that AI has moved from headlines into this building, where people learn, build, and solve problems.
Another is a course scheduling tool. Assigning which faculty teach which courses in which rooms at which times is a problem the school handled manually for years, because no commercial product understood our constraints. We built one that does, and it now runs a process that used to consume weeks of administrative effort.
The third is a case study simulator — an instructional tool that lets instructors create case-study experiences using AI avatars to generate interactive interviews that help students learn. Commercial alternatives exist, and we considered them. But we built our own. It fits our pedagogy, our cases, our students. It belongs to us. And the version we have today is one we can keep improving, on our own timeline, in our own direction.
Each of these projects shares a pattern. Small student teams work together to produce tools that fit their users exactly — and the work itself was experiential learning at its most direct. Real projects. Real problems. Real solutions, used by real people. Students didn't just study how AI is changing business; they shipped something that proves it. That double return — a working tool and the talent who built it — is an outcome no off-the-shelf purchase can produce.
The invitation
This is what xLab does. We prototype ideas. With AI as accelerant, we can prototype more of them, faster, and with sharper fit than was possible even two years ago. The lab pairs design thinking with student teams and AI-assisted development to turn an idea into a working tool in weeks.
For organizations watching the AI landscape and wondering whether the right move is to buy yet another platform — there is another path. The companies that will be best positioned three years from now are the ones building, today, the handful of tools that make them distinctively themselves. That work doesn't require a Silicon Valley budget. It requires partners who know how to design. xLab is one of those partners.
The quiet shift is already underway. The question is whether you'll see it in time to be early.