What actually changed

AI-assisted development lowered the cost of writing software, but not the cost of owning it. Maintenance, security, support, and the opportunity cost of your attention are unchanged. That asymmetry is the whole modern build-vs-buy conversation.

Anyone arguing from build cost alone is answering the wrong question. Version one is the cheapest part of any system that lasts, and the gap between it and year three has if anything widened — the build got faster while the ownership did not.

1. Is this how you compete?

Build what customers would notice you doing differently; buy what they would never inspect. Payroll, email, accounting, ticketing — nobody chooses you because of them. The workflow that makes your business unusual is the one worth owning outright.

The test that cuts through the debate: would a customer ever see the difference between your version and a competitor’s? If the honest answer is no, you are considering building infrastructure, and infrastructure is what the market already sells you cheaply.

2. Are you the awkward shape?

Off-the-shelf products encode assumptions about how work is done. If you fit them, buying is nearly always right. If you find yourself running the business in spreadsheets alongside the tool, you are already paying for custom software — just in salaries and errors instead of engineering.

Look for the shadow processes before deciding. Every workaround spreadsheet, every field being used for something other than its name, and every step that begins “then we export it and…” is evidence that the product’s assumptions and yours have diverged.

The spreadsheet beside the SaaS tool is the specification for what you should have built.

3. What does the integration actually cost?

Bought software still has to be configured, integrated, migrated onto, and maintained against vendor changes. Compare total cost honestly and a surprising number of “obvious buys” turn out to be close calls, particularly where the tool sits in the middle of several systems.

Count the recurring half too: someone administers it, someone handles the upgrade when the vendor changes an interface, and someone owns the connection between it and everything else. Those hours are real and they are rarely in the comparison.

4. Who owns the data, and can you leave?

If the workflow generates data you will want to reason over — with AI or otherwise — check that you can get it out in full, continuously, in a usable shape. Vendors who make export slow or partial are charging you a cost that never appears on the invoice.

This question has become sharper as AI features move from novelty to leverage. Data you cannot reach is capability you cannot build, and the decision history a future AI feature needs is exactly the sort of thing exports tend to omit.

5. Will anyone maintain it in two years?

Custom software with no owner becomes a liability faster than most teams expect. If you cannot name the person or partner responsible for it eighteen months from now, buy — even where building would be better on every other axis.

A retained supplier counts as an answer, provided the arrangement is written down and the handover would let someone else pick it up. “The developer who built it will probably still be around” does not count.

The option most teams skip

Buy the product and build the thin layer that makes it fit — an integration, a screen that assembles what your team needs, a small service holding the logic the product can’t express. That captures most of the benefit of both and is far cheaper than either extreme.

It is skipped because it sounds unambitious in a meeting. It is usually the correct answer, and it is what consolidating information into one screen actually looks like in practice.

The common answer

Buy the commodity layers, build the thin slice that is genuinely yours, and connect them properly. Most companies get this backwards: they build an internal chat tool and buy the system their competitive advantage runs through.

If you take one habit from this, make it the eighteen-month question. Nearly every regretted build and every painful buy can be traced to nobody having asked who would be responsible for the thing once the excitement of choosing it had passed.