Serverless and Agentic Coding Are a Match Made in Heaven

The author of this article does not intend to rehash the typical arguments in favor of serverless architecture. They acknowledge that managed infrastructure, automatic scaling, and the absence of server patching are valuable aspects of serverless. However, the focus of this piece is on how serverless impacts the usefulness of agentic coding, which refers to AI-driven coding agents.

The key point is that serverless provides a more conducive environment for agentic coding. This is not due to a supposed increase in the intelligence of these AI agents, but rather because the serverless architecture becomes a more explicit, constrained, and inspectable system. This matters more than the author initially anticipated.

Traditionally, when building a web application, developers may have to configure various aspects such as the server, database, queue, and other components. This can be a valid setup, but when employing AI coding agents, a problem arises. The agent may understand the application code well, but not the surrounding environment. It might not be aware of the reverse proxy configuration, background worker setup, deployment scripts, environment variables, or even the knowledge residing in a README file.

These gaps can lead to the agent making incorrect architectural decisions when instructed to make changes.

Serverless addresses this issue by making the infrastructure explicit and visible. Functions, queues, buckets, schedules, event sources, permissions, and environment variables are part of the application definition. This visibility allows the agent to have better material to work with, as the context is less hidden and more accessible.

With a serverless.yml file, the agent can understand which functions exist, their triggers, resources, environment variables, and permissions. While this map is not perfect and may not represent the entire system, it serves as a useful representation.

Another advantage of serverless is the use of managed services instead of custom infrastructure code. Services like SQS, S3, EventBridge, and API Gateway eliminate the need for a developer to understand intricate, custom queue implementations or homegrown file storage layers. The managed services act as primitives for the agent, allowing it to connect known components in a standard manner.

This simplification is particularly helpful when making incremental changes. Each step has a clear purpose, making the review process more manageable.

In summary, the author argues that serverless architecture enhances the productivity of agentic coding by providing a more explicit, constrained, and inspectable system. By making the infrastructure visible and using managed services as primitives, the agent has a clearer understanding of the system's context, reducing the likelihood of making incorrect assumptions or introducing unwanted changes.