September 23, 2026 - 5 min read
When 'Temporary' Infrastructure Quietly Becomes Production
Cloud migrations often stall after phase one. Here's why the second phase disappears, and what that costs an organization that never notices.
The migration that never finishes
Most cloud migrations don't fail loudly. They stall quietly, right after the first environment starts working — and the stall gets mistaken for completion. Someone declares victory, the old provider gets decommissioned, and the team moves on to the next priority. Nobody schedules a follow-up, because as far as anyone can tell, the system is up.
"Up" and "production-grade" are not the same claim. We keep running into platforms sitting on infrastructure that was only ever meant to be a stabilization step — a place to get things working before the real production environment got built. That second phase quietly never happens. The interim environment inherits the production label by default, not by design, and nobody notices because nothing ever forced the question.
Why phase two disappears
This isn't a technical failure. It's an ownership failure. A two-phase migration plan assumes someone will still be accountable for phase two once phase one ships. In practice, the moment phase one works, the pressure that funded the migration evaporates. The team that built it moves on to the next deadline. The person who understood the plan changes roles. The old provider gets turned off — often for a good reason, like cost — and that decommissioning reads as the finish line, even though it was only ever supposed to close phase one.
We recently walked into exactly this pattern on a platform we'd been asked to document ahead of a planned expansion. The team already suspected the shape of the problem, which is itself telling: they described their own infrastructure's biggest structural flaw before we'd looked at a single resource. Years earlier, a migration from one cloud provider to another had been planned in two stages — stabilize an interim cluster, then build a genuinely separate production cluster, cut traffic over, and demote the interim one to development. Only the first stage happened. The old provider was decommissioned, the interim cluster kept running under its original name, and the second stage simply never got scheduled. Nobody decided to skip it. It just never got a date.
What we found once we looked read like the fingerprints of that gap:
- Nothing versioned the running state — no infrastructure-as-code, nothing reconciling the cluster against a source repository. Every resource existed because someone had, at some point, run a command against it.
- Every secret was encrypted under a single key that rotated monthly, with no owner tracking that rotation, and two secrets were already failing to decrypt.
- Pieces of a real production environment already existed, sitting unused: a registered domain nobody had pointed traffic at, a separate database project provisioned and never connected.
That last one is the detail worth sitting with. Someone had started phase two. The scaffolding was there. It just never got finished, because finishing it wasn't anyone's job by the time there was time to do it.
None of this was visible from outside. The platform worked. Requests were served, dashboards stayed green, nothing was on fire. That's exactly what makes a skipped phase two dangerous — it doesn't announce itself. It waits.
Debt with no name is debt nobody manages
A promoted interim environment isn't a finished migration. It's a skipped one wearing the finished migration's name. And because it's never formally logged as a gap, it doesn't get a line on a roadmap, doesn't get a budget, doesn't get assigned to anyone. It just accumulates — one more manual change applied straight to the cluster, one more secret nobody remembers the origin of, one more service that was supposed to be temporary and has now been load-bearing for years.
The fix starts before any engineering work happens: naming the thing. A promoted dev environment that's actually carrying production traffic needs to be written down as exactly that — a specific, owned item of technical debt, with a description of what phase two was meant to look like and why it stopped. Once it has a name, it can go on a roadmap. Once it's on a roadmap, someone can be made accountable for it. Until then, it's invisible by construction, and invisible debt doesn't get scheduled — it gets discovered during an outage.
What this means if you greenlit a migration
If you approved a cloud migration and assumed it finished because the system started working, the question worth asking isn't "is it stable" — stability is exactly what lets this failure mode hide. Ask instead: was there ever supposed to be a phase two, and did anyone explicitly decide to cancel it, or did it just quietly stop happening? If nobody in your organization can answer that clearly, you likely have an environment carrying more responsibility than it was ever designed for, with no one currently accountable for closing that gap.
That's not a reason to panic. It's a reason to name the debt, put someone's name next to it, and decide on purpose whether to finish the migration or formally accept the interim environment as permanent — which is a legitimate choice, as long as it's made deliberately instead of by default.
We like this kind of work precisely because the fix is rarely the hard part once the gap is named and understood. The hard part is that nobody had gone looking. If that description sounds close to home, we'd rather help you find out on purpose than have an outage tell you first.