How we build

GDEV OS is the internal system our teams use to work with generative models. It applies our rules and checks to each task while keeping decisions and accountability with people.

LegalFinanceAnalyticsHRUASecurityComplianceLiveOpsNextersCubic GamesRoyal ArkGame GearsSPEEDSECURITYMEMORYOFFONSecurityLegalFinanceComplianceLiveOpsAnalyticsHRUANextersCubic GamesRoyal ArkGame GearsSPEEDSECURITYMEMORYOFFON

GDEV OS

Proprietary operational harness

Contents
  1. Where people make a difference
  2. The model is only part of the system
  3. Why we built GDEV OS
  4. How GDEV OS works
  5. How teams use it
  6. How the system improves
  7. Where people stay in control
  8. Talk to us

Where people make a difference

GDEV is in the business of entertainment. Our people should spend their time on the work that makes entertainment better: understanding players, exercising judgment, making creative decisions and building products people want to return to.

We use automation, including generative models, to take repetitive and mechanical work out of their way. Machines should do the work they can do better, so the teams can focus on the work where human talent really matters.

The model is only part of the system

Frontier models are widely available. What increasingly differentiates the result is the system around them: how a task is defined, what context and tools the model receives, how work is divided, and what checks happen before the output is used.

This surrounding system is often called a harness. GDEV OS is ours.

Why we built GDEV OS

A company of our size, with several products and teams running in parallel, needs shared rules for how models are used, controlled access to data and systems, consistent safety standards, and results that teams can trust.

So we built GDEV OS — our internal operating system for AI-assisted work.

How GDEV OS works

  • Context and rules

    Every task starts with the relevant operating principles, rules and conventions loaded into the system. Agents work within the same set of instructions, and teams do not need to rely on people remembering decisions or procedures from previous projects.

  • Executable procedures

    GDEV OS includes over 100 executable procedures covering workflows such as month-end close and release management.

    Each procedure defines the required steps and the criteria that must be met before the work is considered complete.

  • Task routing and review

    Complex tasks can be divided between 20+ specialized agents. Their outputs can then be combined and checked before use.

    The system routes different types of work to different models depending on the task. More demanding analysis can use larger models, while routine work can use lighter ones.

    Outputs intended for operational use or external distribution are reviewed by the person responsible for the work.

    System documentation is generated from the codebase so that it stays aligned with the current implementation.

  • Automated checks

    Up to 60 automated checks run before certain actions are taken. They are designed to catch issues such as credentials appearing in code, risky database operations or steps that bypass the release process.

    Actions with irreversible or external consequences, such as sending a message on someone's behalf or deleting data, require explicit confirmation from a person.

How teams use it

We train our teams to use generative models as part of their work and to understand the rules that govern their use.

GDEV OS handles much of the technical orchestration in the background. Employees access its capabilities through plugins and skills inside the tools they already use. They do not need to manage agent coordination, model selection or system checks themselves.

The same controls and safeguards are applied across teams by default.

How the system improves

GDEV OS is updated as teams use it.

When the same issue appears repeatedly across different workflows or teams, the pattern can be identified and turned into a proposed system change.

The team responsible for GDEV OS reviews proposed changes before they are introduced.

When a recurring mistake can be prevented with an automated check, we add one.

Where people stay in control

GDEV OS does not make business decisions independently. Judgment, approval and accountability remain with people, and each process has a named owner responsible for its results.

Our internal rules for AI use apply across GDEV:

  • Employees use only AI tools that have passed our software approval process.
  • New automations using generative models are approved before they are built. Each one is registered and follows the same change, access and incident-management processes as our other systems.
  • Confidential and personal data may only be used with AI tools approved for that type of data.
  • Model-generated output is reviewed by a person before it is relied on or shared outside the company. It is not treated as a substitute for qualified professional judgment.
  • Our rules prohibit the use of AI to infringe intellectual property rights or to manipulate people.

We review these rules as the technology and our use of it change.

Talk to us

If you write about GDEV, invest in it, or want to bring a studio to work with us, use the address that fits.

Back to GDEV main