The Evolution of Multi-Tenancy: From Mainframes to AI Agents
The concept of multi-tenancy has been evolving for decades, with each technological advancement shrinking the size of the tenant. From mainframes to virtualization, and now with the rise of containers and Kubernetes, the tenant has transformed from organizations to teams, and further down to individual developers. But the latest disruption in this journey is the emergence of AI coding agents, which challenge our traditional understanding of tenancy.
The Agent Revolution
AI coding agents have revolutionized software development by enabling multiple concurrent workstreams for a single developer. What was once a linear process of one developer, one change, has now become a parallel symphony of multiple agents, each working on different changes simultaneously. This shift has profound implications for platform engineering.
The New Tenant: The Change
The traditional model of capacity planning based on the number of developers (seats) is no longer sufficient. With AI agents, the tenant is no longer the developer but the change itself. Each change, driven by an agent, requires its own isolated environment, data, and system version. This means that the demand for tenancy is now directly proportional to the number of changes in flight, not the headcount.
Scaling Challenges
The challenge lies in scaling multi-tenant platforms to accommodate this new reality. Traditional approaches, such as per-developer namespaces or shared staging environments, fall short as they are designed for human-paced changes. When changes are generated by AI agents, the demand for tenancy can skyrocket, leading to resource contention and inefficiencies.
Redefining Tenancy
To address this, we must redefine tenancy at the change level. The key principles are simple yet powerful: near-free tenant creation, isolation limited to the scope of the change, and a lifecycle bound to the change itself. This approach mirrors the best practices of multi-tenant production services, where tenants share resources while maintaining privacy and autonomy.
Applying the Multi-Tenancy Playbook
Interestingly, the multi-tenancy playbook used in production environments has rarely been applied to development platforms. By treating each change as a tenant, we can leverage the same principles to ensure efficient resource utilization and scalability. This means that a change owns only the services and data it modifies, sharing the rest with other tenants.
The Benefits of Change-Level Tenancy
Change-level tenancy offers several advantages. Firstly, it allows for rapid tenant creation, as the cost is minimal. Secondly, it ensures that isolation is precise, reducing the overhead of duplicating entire stacks. Lastly, automatic offboarding keeps the platform clean and accounting accurate, as tenants are tied to the lifecycle of changes.
Measuring and Planning for Change
For platform teams, the practical shift begins with measurement. Instead of counting developers, we must focus on changes in flight. This new metric helps us understand the demand and plan capacity accordingly. By pricing the marginal tenant, we can identify the parts of the platform that still operate on outdated assumptions and adjust them to accommodate the new reality of AI-driven development.
Enabling an Agent-Native SDLC
Redefining tenancy around changes is crucial for the software development lifecycle (SDLC) in the age of AI agents. With agents generating changes at machine speed, traditional person-sized tenancy becomes a bottleneck. By embracing change-level tenancy, we can convert agent throughput into merged work, unlocking the full potential of AI-assisted development.
Looking Ahead
As AI coding agents continue to gain traction, the software development landscape will undergo significant transformations. Organizations that adapt their platforms to accommodate change-level tenancy will be better positioned to harness the power of AI. This evolution in multi-tenancy is not just a technical shift but a strategic imperative for staying competitive in the era of AI-driven development.
