From Prototype to Production: Engineering AI Agent Systems with Python on AWS

Every DevOps engineer has watched a Python prototype work perfectly in a notebook, then fall apart the moment it hits real traffic, real failures, and real cost constraints. AI agent systems make this gap even sharper, they're non-deterministic, hard to observe, and dangerous to run without guardrails. This talk closes that gap.

Using a production multi-agent DevOps system built with LangGraph, the Claude API, and FastAPI as a working case study, this session walks through what it actually takes to run autonomous AI agents reliably on AWS: policy enforcement with OPA so agents can't take unsafe actions, GitOps-based deployment for repeatable releases, and full observability with Prometheus, Grafana, and OpenTelemetry so failures are visible before they become incidents.

Attendees will leave with a concrete architecture pattern, real failure cases the guardrails caught, an AWS service map for running this affordably, and a practical roadmap, regardless of whether they're new to AWS or already running production workloads.

This is a talk for engineers who want to build the next generation of Python-powered AI systems in Africa, and ship them safely.