Case Study
Joy Studios: rebuilding the platform under a live games studio
Stakes
Joy Studios builds casual games for a global audience and runs them as live services: real players, real-time gameplay, and partners who expect the platform to be as dependable as their own. The studio had grown quickly, and the foundations it grew on needed rebuilding to carry what was coming next. Multi-tenant white-label hosting and a second region were already on the roadmap, with formal security assurance a condition of the deals behind them. The brief was to rebuild the platform properly, as code, while production kept serving players throughout.
Constraints
- The games stayed live: players and partners could not see the rebuild happening
- Two delivery toolchains ran side by side through the transition, with the outgoing one maintained until its last service moved
- Security assurance had to fall out of the platform as evidence, not be assembled afterwards
- The platform had to be operable by the team the studio already had, so simplicity of operation outranked sophistication
- Every design had to stand on its own in writing, reviewable without a walkthrough
Approach
Identity first
Before any infrastructure moved, the whole team came onto corporate single sign-on: one identity across the tools the studio runs on, with multi-factor authentication enforced once, centrally. Group membership became the source of truth for access, changes to it go through review, and off-boarding a leaver is a single action across every system.
Governed foundations, refactored in place
A new, isolated account structure houses the platform, with organisation-level guardrails on where workloads may run, how identity is bounded and how audit trails are protected. The existing production environment was brought under governance in place rather than rebuilt, so player balances and game records never moved. Shared security, backup and scanning services are administered centrally, so bringing a new environment under them is a small change rather than a project.
Delivery as a pipeline, not a person
Infrastructure is defined as code, and every change passes automated security, policy, vulnerability and cost review before a human approves it, with locking so two changes cannot collide. Third-party pipeline steps are pinned to verified versions and re-audited on a regular cadence. Networking, DNS and the studio's public website are managed the same way.
Cut over service by service
The rebuild followed a dependency-ordered roadmap with no calendar-driven big bang. The production foundation landed as reviewed layers, and the first production service moved onto it with its data migrations running as an ordered part of deployment. Backups and their copies were proven rather than assumed. Architecture decisions were written down with the alternatives considered, and handed to the company with the platform.
Deliverables
- Corporate single sign-on across every business tool, multi-factor authentication enforced centrally, and no long-lived credentials on the new environments
- Governed multi-account cloud foundation with organisation-level guardrails, centrally administered security and backup tooling, and audit trails throughout
- Infrastructure-as-code repository with change, drift and review workflows gated by security, policy, vulnerability and cost checks
- Application images built once and promoted unchanged, with data migrations run as ordered deployment steps
- Networking, DNS and the public website under code, with additional regions approved and policy-enforced ahead of demand
- First production service running on the new platform, cross-environment backups proven, and a written decision log handed over
Outcome
Joy Studios runs its first production service on a platform that is reproducible from version control, gated by review and policy, and legible to the whole team. Every access change, infrastructure change and cost change is a reviewed proposal, with the evidence auditors ask for produced as a by-product. Production served players throughout, and the remaining services follow the same path, one at a time, with no big-bang migration ahead of them.