Month Four: What Maintaining a Shipped AI-Built App Actually Involves
A build quote ends the day you launch. The stack underneath it does not. Four dated changes between 23 July and 21 October 2026 show what month four actually asks of you.
If your app keeps breaking, start with our guide on how to fix an AI-built app without rebuilding it.
The quote ends at launch. The dependency tree does not.
Every fixed-price build package answers the same question: what does it cost to get to launch? That is a fair question and it has a clean answer. The question nobody prices is the second one, which is what it costs to still be running in month four. That one has no clean answer, which is exactly why it tends to go unmentioned. A package that says “production-ready” is describing the state of the code on handoff day. It is not a claim about the state of the code ninety days later, because nobody can make that claim: your code will not have changed, and everything underneath it will have.
Here is what “everything underneath it” did in the last eight weeks.
Four dates, two platforms, one already-shipped app
| Date | What changed | How it announces itself | Work it creates |
|---|---|---|---|
| 22 Jul 2026 | Supabase deprecated explicit version clauses in CREATE/ALTER EXTENSION; they are now ignored | Silently | Re-check any extension you pinned on purpose. |
| 25 Aug 2026 | Next.js patched two critical-severity remote code execution issues in 16.3.3 and 15.5.24; the patch disables AVIF image optimisation until the upstream fix propagates | Silently, because images keep working and only the format changes | Patch, then re-check your image pipeline and payload weights. |
| 23 Sep 2026 | Supabase removes the Management API logs.all endpoint; the ClickHouse-backed replacement accepts ClickHouse SQL only | Loudly, for anything that calls it | Rewrite log queries and any monitoring built on them. |
| 21 Oct 2026 | Next.js 15.x leaves Maintenance LTS, two years after its release | Not at all | Plan the 16 upgrade before the next advisory, not after. |
None of those four is a crisis, and that is the point. Three of them fail quietly, and the fourth arrives on a date that will never show up in an alert.
Two kinds of upkeep, and only one of them sends email
Loud upkeep is the kind founders already budget for. Something returns a 500, a customer writes in, you fix it. It is unpleasant, but it is self-announcing, and a support retainer covers it well.
Quiet upkeep is the other kind, and the August Next.js patch is the clean example. Applying it is a one-line version bump. What it actually does is disable AVIF optimisation until the upstream libheif fix lands, so your images keep rendering and quietly get heavier. No error, no alert, no ticket. You find it when somebody finally looks at Core Web Vitals, or you never find it at all. The extension-pinning change has the same shape: a clause you wrote deliberately is now ignored, and nothing tells you.
Quiet upkeep is what a launch-day quote structurally cannot cover, because catching it requires somebody to be reading changelogs in month four for an app that is, by every visible measure, fine.
What “just bump the version” hides
The Next.js 16 upgrade is the one most Lovable-derived apps are heading toward, so it is worth knowing what it moves before you schedule it. Turbopack is now the default bundler, and a project carrying a custom webpack config will fail the build rather than quietly ignore it. The middleware.ts convention is deprecated in favour of proxy.ts, which runs on the Node.js runtime and does not support the edge runtime. Synchronous access to params, searchParams, cookies(), headers() and draftMode() is gone, because the Next.js 15 compatibility period has ended. revalidateTag() now takes a cacheLife profile as a second argument. Every parallel route slot requires an explicit default.js file or the build fails.
Then there are the defaults that shift underneath you without a single code edit. images.minimumCacheTTL moves from 60 seconds to 4 hours. images.qualities goes from accepting anything to accepting only [75], and any other quality value is coerced to the nearest allowed one. Neither of those breaks a build. Both change what your users download.
There is a codemod for the mechanical part of the upgrade. There is no codemod for deciding whether the new defaults are the ones you wanted.
Where the Lovable rendering line sits
One more dated boundary, because it decides which conversation you are even in. Lovable’s own documentation draws the line at 13 May 2026. Apps created on or after that date run on TanStack Start with server-side rendering, so every request returns fully rendered HTML. Apps created before it run React and Vite with on-request pre-rendering on deployed public URLs, and that pre-rendered HTML is served only to verified crawlers. Human visitors and unverified agents get the SPA shell.
The practical consequence is worth internalising before you panic at a report. Paste an older Lovable app into a third-party SEO scanner and the scanner sees the shell, not the pre-rendered output, and tells you the page is empty. That result is documented behaviour, not a bug, and it is not evidence that Google cannot see you either. Knowing which side of 13 May your app was built on is what tells you whether an SEO report describes a real problem or a measurement artefact, and therefore whether the right next step is a migration or nothing at all.
Three questions worth asking of any build quote
We are not arguing that fixed-price builds are dishonest. Ours are fixed-price. We are arguing that the honest version of a fixed price states where the line falls instead of leaving it to be discovered in month four. So, three questions to put to any quote, including ours.
First: what is the support window in days, and does it cover only bugs in what was shipped, or upstream changes too? Second: after that window closes, who is reading the changelogs of the platforms this app depends on, and if the answer is nobody, is that a decision you made or one that was made for you? Third: when a framework support date arrives, is the upgrade treated as a continuation of the relationship or as a new project at a new price?
Any of those can have an answer you are happy with. The problem is only ever the unasked one.
Where we land on it
Our own numbers, for the record: 80 jobs on Upwork, 100% Job Success and a 4.9 rating from 30 reviews. Both build tiers, Founding Client at $2,999 and Publishable MVP at $4,999, include 90-day support until your first paying customer, and the revenue guarantee is first paying customers within 90 days of launch or a full refund. The Founding Client tier carries a 14-day build guarantee with a full refund if the deadline is missed. A Strategy Audit is $499 and is credited toward any build.
Ninety days is ninety days. It is a real window, and it is also a finite one, and the four dates above will keep arriving after it closes. That is what our ongoing maintenance and support work is for, and we would rather keep it a separate, explicit conversation than fold an unbounded promise into a fixed price and hope the month-four bill turns out to be somebody else’s.