Serverless Compute for More Kinds of Workloads

Resources for Serverless Compute for More Kinds of Workloads

For ten years AWS Lambda had one shape: an event arrives, your function runs and returns. This talk shows how that shape has grown into three kinds of Lambda and two places to run them. Durable functions bring checkpoint and replay to long-running workflows and agent loops. Web Functions run a standard HTTP server with many requests per environment and CPU billed only while it works. MicroVMs give you a sandbox you control, for untrusted code, game servers, CI runners, and agent tools. Lambda Managed Instances run your functions on EC2 instances that Lambda manages in your account. The talk closes with the public Lambda roadmap and how to build all of this with an AI coding agent and the Agent Toolkit for AWS.

Lambda durable functions

Workflows up to a year, with checkpoint and replay

Write a workflow as ordinary sequential code. Lambda checkpoints each step, suspends during waits and callbacks at no cost, and replays from the last checkpoint after a failure. The talk's expense claim walkthrough uses map, parallel, a manager-approval callback and an idempotent pay-out step.

Lambda Managed Instances

Your functions, on EC2 instances Lambda manages

Run Event Functions, durable ones included, on EC2 instances in your account and VPC. Lambda launches, patches and terminates the instances. You pick the instance type, pay EC2 pricing with Savings Plans and Reserved Instances, and get a 90 minute timeout.

Build with an agent

Agent Toolkit for AWS

Your terminal or AI coding assistant plus the Agent Toolkit for AWS: current AWS documentation and tested skills that know the new Lambda form factors, so the agent can pick the shape that fits a workload and build it.

Lambda Web Functions

A standard HTTP server on Lambda

Your Node.js server listens on a port, with no adapter and no handler. Each execution environment serves many requests at once, CPU is billed only while your code runs, and one URL can serve several Regions. The demo deploys a streaming dashboard with a single command and measures 300 streaming requests across five environments.

Lambda MicroVMs

A sandbox you control

Launch, suspend, resume and terminate a microVM with its own kernel, built from an ordinary container image, with memory and disk kept across suspend for up to 8 hours. The demo uses the Lambda MicroVM Demo Playground for game servers, CI runners and an agent that builds an application from a prompt.

Help define the future of Lambda

The public Lambda roadmap

Everything in this talk started as an issue on the public roadmap. Upvote the issues you need and comment on what's missing.

Presented by:

Julian Wood

Julian Wood

Julian is a developer advocate at AWS who loves helping developers learn how serverless and AI technologies can transform the way they build and run applications in the cloud.

Follow on LinkedIn