Build vs buy just flipped: what a third of companies skipping a software purchase means for your business
Aviv Ben-Dov · Founder & CEO, Orchestra-Labs
Build vs buy is the oldest question in business software: when you need a new capability, do you pay for an off-the-shelf product or build your own? For decades the answer for most companies was buy, because building was slow and expensive. In 2026 that math changed. AI coding agents dropped the cost of building custom software far enough that a third of companies now skip a purchase they would have signed without thinking a year ago. Here is what that shift actually means for a small or traditional business, and where it helps you rather than distracts you.
The number that made everyone look up
The headline came from McKinsey's State of AI 2026, its annual survey of more than 1,700 respondents across nearly a hundred countries. Nearly a third of organizations, 32%, reported deciding against buying at least one software product or feature because AI coding agents let them build the functionality themselves. Among the highest performers the figure was close to half. The signal is not that software vendors are finished; it is that the price of building crossed a line. Because building used to take a team and months, buying almost always won for anything outside a company's core product. Now, for a narrow feature, that default no longer holds automatically - and a lot of buyers are noticing at once.
Why this is not just a big-company story
It is tempting to read a McKinsey survey as news about enterprises with large engineering benches, and to assume it does not touch a distributor or a law firm. That reading misses the point. The same economics that let a bank skip a feature purchase also let a smaller company skip one - and smaller companies feel each SaaS subscription more sharply, because a per-seat tool that solves one workflow can cost more per year than the process it automates saves. Moreover, the evidence is not limited to McKinsey. A Retool 2026 report found 35% of organizations have already replaced at least one SaaS tool with custom-built software, and most expect to build more. The direction is the same across the data: the buy-by-default reflex is loosening.
What "build" really means when you have no R&D team
Here is the honest translation, because the word "build" is where small businesses get the wrong idea. For a company with no developers, build vs buy does not mean hiring engineers or writing code in-house. It means that the specific gap you were about to fill with another subscription - the extra CRM add-on, the standalone form tool, the manual copy-paste between two systems no product connects - can instead be a bounded AI agent built on the systems you already run. An implementation partner builds and hosts it inside your environment; you own the workflow outright. This is exactly the decision our custom agent versus off-the-shelf copilot guide lays out: if a task ends in text, a copilot is enough; if it ends in an action across your systems, a custom agent usually fits better and, increasingly, costs less.
Where to build, and where to keep buying
The mistake would be to swing from buy-everything to build-everything. That is how you end up rebuilding an ERP badly. The useful distinction is between systems of record and the workflow layer on top of them. You should almost never rebuild the systems that hold your data - accounting, ERP, CRM, email. Buy those, keep them, and never let their data leave your control. What is now worth building instead of buying is the thin layer of glue and judgment above them: routing an incoming order, decoding a non-standard RFQ, reconciling two reports, chasing an approval. That layer is where off-the-shelf tools fit worst, because it is specific to how your business actually works - and it is exactly what an agent handles well, as our back-office automation guide describes in detail.
The catch nobody prices in
Building is cheaper to start than it has ever been, but "cheap to build" is not the same as "free to run." A custom agent still needs an owner, a defined success metric, wired-up access to your data, and human approval on anything that writes or spends. Skip those and you rebuild the exact failure that sinks bought-and-bolted AI too - the pattern behind McKinsey's other 2026 finding, that only a minority of companies turn AI spend into real earnings, which we covered in the AI earnings gap. The advantage of building is not that it removes discipline; it is that a bounded, well-owned agent gives you a workflow shaped to your business with no recurring seat and no data leaving your walls. Those are the same guardrails that separate the AI projects that last from the ones we watch stall in why most AI agents never reach production.
What should you actually do this quarter?
- List the subscriptions you are about to renew or add. Flag any that exist to solve one narrow workflow rather than to store your data. Those are the build-vs-buy candidates.
- Keep the systems of record. Do not rebuild your ERP, accounting or CRM. Build only the glue and judgment layer that sits on top of them.
- Scope one bounded agent, not a platform. Pick a single process with a clear start and finish, and write the success metric in hours saved or errors prevented before anything is built - the approach in our implementation guide.
- Keep humans at the gates. Read-only by default, approval on writes, a full audit log. Then measure for a few weeks and expand only on what the numbers justify, as in our small-business playbook.
Frequently asked questions
What does the build vs buy shift in AI mean for 2026?
AI coding agents lowered the cost of building custom software enough that companies now weigh building against buying for capabilities they would have simply bought a year ago. McKinsey found 32% of organizations skipped at least one software purchase to build it internally instead.
Does a small business need developers to build instead of buy?
No. For a company without R&D, build means replacing a narrow SaaS feature with a bounded agent built on your existing systems by a partner - you own the workflow and pay no recurring seat, without writing code yourself.
Should a small business stop buying SaaS?
No. Keep buying your systems of record - ERP, accounting, CRM. The build decision applies only to the thin workflow layer on top, where a custom agent fits your process exactly and often costs less over time.
Is a custom agent safer than another SaaS subscription?
It can be, when it runs inside your environment with your keys, read-only by default and approval on writes. Your data never leaves your perimeter, and there is no extra vendor holding a copy of it.
References
Wondering which of your subscriptions is really a build-not-buy candidate? Happy to map it together on a short call.
Book a call