The Future of Serverless • Nick Coult • GOTO 2025

The core mission of serverless is to allow developers to focus purely on business logic, abstracting away everything else. This principle, articulated by AWS CTO Werner Vogels as “all the code you ever write is business logic,” remains the guiding star for the future of serverless. It’s not just about moving faster; it’s about enabling capabilities, like complex AI integrations, that were previously out of reach due to the overhead of infrastructure management.

The Serverless Abstraction Stack

Serverless can be visualized as the highest level of an abstraction stack that starts with physical data centers at the bottom, moves up through VMs and other compute abstractions, and culminates in serverless functions where you only provide the code. The evolution of this stack is about AWS managing the lower levels to provide ever-higher levels of abstraction for developers.

AI represents the next layer on top of this stack, moving from business logic to pure human intent. The key benefits that drive serverless adoption are:

• No Infrastructure to Manage: Developers are freed from managing servers, capacity planning, and scaling.
• Pay for Value: You only pay when your code is executing, eliminating costs for idle infrastructure. This allows for inexpensive prototyping and scales cost-effectively.
• Built-in Best Practices: By using serverless, you inherit AWS’s extensive experience in building available, resilient, and secure systems at massive scale.

Event-Driven Architecture: The Dominant Pattern

The most popular way to leverage serverless is through Event-Driven Architectures (EDA). An event is simply a change or an update, and arguably, the vast majority of software is fundamentally event-driven. AWS Lambda was designed for events from its inception.

EDA provides significant benefits:

• Technical Decoupling: Systems can scale independently. An event producer doesn’t need to know how consumers will process the event, allowing for more resilient and scalable applications.
• Organizational Decoupling: This is a crucial benefit for development velocity. Small, independent teams can ship features without coordinating with the entire organization, dramatically accelerating the development cycle.

The adoption of this model has been massive, with over a million active customers using AWS Lambda every month, resulting in over 10 trillion function invocations monthly.

The Future is Agentic AND Serverless

As we look to the future, the rise of AI Agents seems to be the next major shift. An agent is distinct from traditional software in that it is:

• Goal-oriented
• Uses context to make decisions
• Uses tools (including serverless functions) to perform actions
• Maintains memory of past interactions

It’s a common misconception that agentic systems will replace serverless. The reality is that as the world transitions to agents, it will require more serverless, not less. The fundamental benefits of serverless—no infrastructure management, automatic scaling, pay-for-value—are even more critical for the dynamic and non-deterministic nature of agents. You cannot perform capacity planning for an agent that might decide to use a tool unpredictably. Serverless functions as tools for agents solve this problem perfectly.

Two primary patterns for agentic systems are emerging:

1. Agentic AI: This is the “full-on” agent model where an LLM is in an iterative loop, planning and executing actions to achieve a goal. It is non-deterministic, and the LLM is “driving the bus.”
2. Agentic Workflows: In this pattern, a developer defines a structured workflow (e.g., using AWS Step Functions) with well-defined business logic. An LLM can be called as a tool within this workflow (e.g., to summarize a document), but it doesn’t control the overall process.

Both of these patterns are event-driven. An agent can be both a consumer of events (which might trigger its execution) and a producer of events (as it performs actions).

Building for the Agentic Future

AWS is actively investing in making the combination of serverless and agents even more powerful by focusing on:

• Harmonizing Payloads and Limits: Creating consistency across services like Lambda, SQS, and agents to simplify building complex applications.
• Workflows: Developing entirely new experiences with Step Functions and Lambda specifically designed for agentic use cases.
• Developer Experience: Improving the “inner loop” with better local development and debugging tools.
• Performance: Continuously investing in fundamentals like reducing cold starts, increasing throughput, and lowering latency.

For developers looking to build for the future, the advice is to:

• Focus on Outcomes: Think about the key business events you care about, not the specific models or languages.
• Identify Agent Opportunities: Determine where an agent could provide value that traditional software cannot.
• Leverage Existing Systems: Package existing functionality into tools that an agent can use to create new experiences.

The most important step is to start now. The full potential of serverless and agents can only be discovered through hands-on experimentation and iteration.

• Tags:
• AI Agents
• Agentic AI
• Feature
• AWS
• AWS Lambda
• Serverless
• Event-Driven Architecture
• Future of Compute