From Prompt to Production: AI Agents, MicroVMs & Heavy Workloads on AWS
#2 Containers and AWS Meetup 2026
Building an AI agent in a notebook takes an afternoon. Running one in production, with isolation, scaling, observability and a bill you can defend, is a different problem entirely. And once the agent works, where do you actually let it execute code? Join us for the next Containers and AWS Meetup in Zürich, where we go from prompt to production: a hands-on lab building real agents on Amazon EKS, a look at how AI is reshaping the daily work of an AWS engineer, the new isolation model behind Lambda MicroVMs, and a real-world scientific workload that shows what all this infrastructure looks like under serious load. As always, the event has two tracks. Join the hands-on lab, the evening talks, or both.
When and where
Tuesday, 15 September 2026, 15:45 to 20:30 CEST
AWS offices, Marsstrasse 2, 8002 Zurich
Registration is free via the Meetup event page.
Agenda
15:45 Arrival for the hands-on lab
16:00 Hands-on lab: AI Agents on Amazon EKS
17:30 Apéro and networking break
17:45 Arrival for the meetup
18:00 Welcome and introduction, Tobias Vonesch, copebit AG
18:05 Talk 1: The Augmented Engineer: From Idea to Production with Claude, Sebastian Hutter
18:40 Talk 2: Isolated Compute for Humans and Agents: Lambda MicroVMs, Benny Gaechter and Marcel Neidinger
19:05 Talk 3: Finding the Right Compute for Fortran in the Cloud, Jan van Thoor
19:45 Pizza, beer and networking
20:30 End
Hands-on lab:
AI Agents on Amazon EKS. Build and run real agents on Kubernetes, following the AWS Workshop Studio lab. Bring a laptop.
Talk 1:
The Augmented Engineer: From Idea to Production with Claude
AI can write code. But between "it works on my machine" and a feature running in production lies everything that actually makes engineering hard, and that is exactly where AI augmentation pays off. Aimed at solo developers and small teams, Sebastian Hutter, Cloud Software Engineer at copebit AG, places the "augmented engineer" on the AI maturity ladder, past vibe coding and well before autonomous agent fleets, then gets practical: setting up Claude for real daily work, sharing team conventions through a private plugin marketplace, and what changes once skills and MCP enter the workflow. Live finale: build a feature, ship it to AWS, break it in production, and debug it together with Claude.
Talk 2:
Isolated Compute for Humans and Agents: Lambda MicroVMs
Interactive dev environments, AI agent sandboxes and isolated CI/CD runners all need the same thing: per-session compute that starts fast, isolates strongly, and keeps state across pauses. AWS Lambda MicroVMs are Firecracker-based environments with VM-level isolation, snapshot-based startup, and hours of persistent memory and disk state. You launch, suspend, resume and terminate them explicitly, and each MicroVM gets its own HTTPS endpoint with built-in auth. No cluster, no load balancer, no ingress plumbing. Benny Gaechter and Marcel Neidinger from AWS cover the lifecycle and networking model, how MicroVMs differ from Lambda functions, Amazon EC2 and container platforms, and demo the use cases they have already put them to work in.
Talk 3:
Finding the Right Compute for Fortran in the Cloud
Take a pile of decades-old Fortran, terabytes of weather data a day, and a hard SLA, then find it a home in the cloud. Jan van Thoor, MeteoSwiss and copebit AG, follows a dispersion-modelling workload through four compute options on AWS: from AWS Batch, which worked but produced an expensive surprise over Christmas, through AWS Fargate and Amazon ECS on Amazon EC2 to Managed Instances. Expect real numbers, the I/O bottleneck behind every decision, and a migration that landed 50 to 65% cheaper.
Speakers
Sebastian Hutter, copebit AG
Benny Gaechter and Marcel Neidinger, AWS
Jan van Thoor, MeteoSwiss and copebit AG