Custom mobile app development in 2026: how to buy it well

Follow the citations in any "app development costs $X–$Y" article and you land, almost always, on a development agency's own blog — a seller publishing the anchor it would like buyers to arrive with. The figures are not audited, not from comparable scopes, and not falsifiable; they circulate from listicle to listicle until repetition looks like consensus. We wrote about this pattern at length in the cost guide, and the conclusion holds here: a price range without a scope attached is marketing, whoever publishes it — including us, which is why we don't.

What determines your real number is boringly concrete: scope (screens, integrations, roles, offline behaviour), the seniority and location of the people building, and how many rounds of change you'll actually request. Which means the only way to get a real number is to fix the scope first — and that is a document, not a phone call.

Agencies quote high when requirements are vague — not from malice, but because vagueness is risk and risk is priced. Invert it: write the spec before you talk to anyone.

A buyable spec is smaller than you think: the product in one paragraph; the user scenarios, concretely ("a returning user does X and sees Y"); the data you store; the integrations you need; what is explicitly out of scope for v1; and how you'll know it works. If that list looks familiar, it is because it is a PRD — and here is the 2026 shortcut most buyers don't know: you can have an AI agent draft it. Spec-driven tools produce exactly this document as their first artifact; even if an agency builds your app, an afternoon with such a tool produces the RFP that makes three quotes comparable, and a working prototype to attach to it — the single highest-leverage afternoon in the whole procurement.

Then send the same spec to every candidate and compare like for like. A vendor whose quote arrives without questions about your spec has not read it; that is also information.

1. Who owns the code — and when? Full ownership, assigned in the contract, from first commit — not "on final payment," which makes your codebase a hostage in any dispute.

2. Whose accounts hold the keys? Apple and Google developer accounts, signing keys, domains, databases: yours, from day one. Migrating them out of an agency's account later ranges from tedious to impossible.

3. What stack, and could someone else maintain it? Standard native (Swift/SwiftUI, Kotlin/Compose) or mainstream cross-platform means any competent developer can take over. A proprietary framework means the vendor is pricing your exit at zero.

4. What does a change cost after launch? Get the hourly/retainer terms for post-launch work in the same contract as the build. This number, not the build quote, is what you will live with.

5. What are the milestones, and what ships at each? Phased delivery with working software at every phase — not a nine-month silence ending in a reveal.

6. Who writes the tests, and where do they live? In the repo you own, runnable without the vendor. Untested code is a maintenance contract you didn't mean to sign.

7. What happens if you stop? The exit clause: notice period, handover obligations, and what you hold on the day it ends. Ask it while everyone is still friendly.

Now the part we sell, boundaries first.

You still want professional services when: the product carries regulatory weight (health data, finance, children's products); the backend is genuinely complex and lives inside enterprise systems; or the organisation buying needs a vendor to hold accountable more than it needs speed.

You probably don't when the job is validation, an MVP, or a focused product. The 2026 fact that changes this market: an AI agent can now produce exactly what the definition of "custom" above demands — your data model, your flows, native code you own. Modaal (ours) runs the same process an agency would, compressed: it writes the PRD, plans every feature as a reviewable spec before building it, and outputs native SwiftUI and Kotlin/Compose on your own disk. The free plan — one project, unlimited prompts — means the discovery-and-prototype phase that anchors an agency engagement now costs nothing but your attention; Pro (€9/month billed annually) adds both platforms and store distribution. Validate the idea first, always.

And the hybrid is often the honest optimum: build v1 with the AI route, and when the product earns it, hire engineers into a codebase rather than commissioning one — standard Swift and Kotlin means the handover is a pull request, not an archaeology project. Question 3 above works in your favour when you are the one who owns the repo from day one.