A platform team proposes moving the inner loop into cloud development environments, so engineers edit locally and everything builds and runs remotely. What is gained, what is paid, and when does the bill arrive?

What is gained, quantified where it can be

• Onboarding time. The realistic before-and-after is days to under an hour for a new engineer's first running system, because environment setup becomes provisioning rather than a documented ritual with per-machine variation.
• Parity with production. The service runs on the same operating system and architecture as production, which removes a category of defects that appear only after merge.
• Access to dependencies. Systems needing twelve backing services or a large dataset become runnable at all, which for some estates is the whole argument: the local loop had stopped existing.
• Fleet-wide change. Toolchain upgrades apply centrally rather than through a wiki page asking everyone to update.

What is paid

• Latency on every keystroke-adjacent action. File sync, terminal round trips and language server responses now cross a network. 20 ms is invisible; 80 ms is felt continuously and is a common cause of quiet abandonment.
• A hard dependency on connectivity. Trains, conferences, bad hotel networks and the occasional cloud incident become total productivity stops rather than inconveniences.
• Continuous cost. A modest instance per engineer for working hours lands in the region of \(80 to \)200 a month each in 2024-2025 pricing, so 60 engineers is roughly \(60k to \)140k a year, which is real and small next to the engineer-years described above.
• A new tier-one service. The platform team now owns something whose failure stops all development, with an availability requirement most internal tools have never had to meet.

When the bill arrives

Cost arrives in month two, when idle environments are discovered running at night and at weekends. The fix is aggressive suspension on inactivity, and it must be built early, because retrofitting it after finance notices is a worse conversation.

The reliability bill arrives at the first platform incident, when 60 engineers are idle simultaneously. That is the moment the decision is judged, and the mitigation is a documented, tested local fallback for the highest-priority work, kept working rather than assumed.

How to keep the option to reverse

Do not let the local path rot. Keep the repository runnable locally for the core services, even if the full estate is only runnable remotely, and run that path in CI so it cannot silently break. Teams that delete the local path lose the ability to reverse the decision within about two quarters.

When not to do this at all

When the local loop works. A team whose services start in 30 seconds on a laptop gains nothing and pays latency and money. The trigger for this change is a measured inner loop that is already broken — a start-up sequence measured in tens of minutes, or a dependency set no laptop can hold — not an aesthetic preference for centralisation.