Build vs. buy in 2026: AI broke the old math
Custom software used to lose the build-vs-buy spreadsheet by default. AI-assisted development changed the cost side so much that the honest answer is now 'it depends' — here is the new framework.
For twenty years the build-vs-buy conversation had a default answer: buy. Custom software meant six figures, six months, and a maintenance tail nobody budgeted for. SaaS meant a credit card. The spreadsheet wasn't close.
Between 2024 and 2026, the cost side of that spreadsheet collapsed. AI-assisted development — senior engineers directing code-generating tools rather than typing every line — cut delivery costs for well-scoped software by 60–80%, per most published estimates. Google's CEO said in 2025 that over 30% of the company's new code was AI-assisted; Gartner expects 90% of enterprise engineers to use AI assistants by 2028. The market noticed:
35%
of teams have already replaced at least one SaaS product with a custom build
Retool build-vs-buy research, 2026
78%
expect to build more custom internal software this year
Same survey
60–80%
typical reported cost reduction for AI-assisted custom builds
Range across 2025–2026 industry analyses
What actually changed (and what didn't)
Three shifts converged. AI coding tools made implementation dramatically faster. Foundation models made "smart" features — extraction, drafting, classification — something you rent by the token instead of building a data-science team for. And integration standards like MCP made connecting to your existing systems days of work instead of months.
What did not change: software still needs scoping, security, and maintenance. AI compressed the typing, not the thinking. Budget 10–20% of the build cost per year to keep anything alive — that number survived the revolution intact.
The 2026 decision framework
| Signal | Points to buy | Points to build |
|---|---|---|
| The workflow | Generic (payroll, email, accounting) | Specific to how you win business |
| Compliance stakes | Regulated, audited, certified | Ordinary business risk |
| SaaS pricing | Flat, reasonable | Per-seat fees scaling painfully with headcount |
| Fit | Tool covers 90%+ of the need | You run your business in the export-to-spreadsheet gap |
| Urgency | Needed live this week | Worth 4–8 weeks for exact fit |
| Ownership | No appetite to maintain anything | You want the asset and the data model |
The old rule was "build only your differentiator." The new rule is softer: buy the commodity substrate, build the workflow layer. Keep the accounting system; build the quoting tool that encodes how you actually price jobs.
A worked example (illustrative)
A 30-person services firm pays $79/seat/month for a field-ops SaaS it uses at maybe 40% — the rest is workarounds. That is roughly $28K per year, forever, rising with headcount.
An AI-assisted custom replacement scoped to the workflows they actually run: call it $30–45K to build, plus ~$6K/year maintenance. Break-even lands around 18 months, after which the custom tool is dramatically cheaper and fits perfectly — no per-seat tax on growth, no feature hostage-taking at renewal.
Five years ago that same build quoted at $120K+ and the spreadsheet said "don't." That is the entire shift in one paragraph. It is also why internal tools went from our slowest category to our fastest.
The Klarna caution
Build-vs-buy stories now come with a mandatory cautionary tale: Klarna publicly replaced Salesforce with an in-house AI-built system, then walked parts of it back when reality set in. The lesson is not "never build." It is that rip-and-replace of a mature, mission-critical platform is the hardest possible first project. The winning pattern in the 2026 data is the opposite: narrow point tools, one workflow at a time, replacing the SaaS you barely use rather than the one your revenue runs through.
How to run the decision
- Write the workflow down first. One page: triggers, steps, exceptions, systems touched. Most build-vs-buy debates dissolve here.
- Price the status quo honestly. SaaS fees + workaround hours + the errors the workaround causes.
- Get a fixed-scope build quote. If a vendor cannot scope your workflow into a 4–8 week first release, the workflow is not defined enough to build — or they are the wrong vendor.
- Compare at the 3-year mark, not year one. Include maintenance on the build side and seat growth plus renewal increases on the buy side.
- Whoever wins, demand an exit. Data export on the SaaS side; documentation and handover on the build side. Lock-in is the real cost either way.
The honest 2026 answer to build-vs-buy is that it is finally a real question again. Run the numbers with current costs — not the costs from the last time you checked.
