Meet Strands harness: Build State-of-the-Art Agents at 28% Lower Token Cost

Building an AI agent that works reliably without burning through your token budget takes a lot more engineering than most demos make obvious. In this video, we're breaking down Strands harness, a new general-purpose agent harness from AWS that packages context management, memory, tools, skills, prompt caching, and more into tuned defaults you can use out of the box. The launch benchmarks report state-of-the-art agent performance with 28% lower token cost. We'll also show you how to use Strands harness and GitHub Actions to build a real documentation maintainer bot that automatically updates documentation when code changes merge, remembers context across runs, and lets us send it back to work with feedback. Strands harness: https://go.aws/47c8JHh Read the blog: https://go.aws/4ySbzgx Docs bot code: https://go.aws/4yVmtlS Getting started: https://go.aws/4ybOVjt Launch benchmarks: https://go.aws/4hpHIVD Follow AWS Developers! 🆇 X: https://go.aws/46F6I6j 💼 LinkedIn: https://go.aws/4xPjYRa 0:00 Don't Build a Harness From Scratch 0:41 You Supply What Makes Your Agent Different 1:37 Model Providers & Custom Instructions 2:21 Memory, Session IDs & Skills 3:02 Watch It Run: Docs Agent in GitHub Actions 3:50 The PR + Leaving Feedback 4:56 Long-Term Memory 5:34 What You Get: Tools, Context Management, Caching 6:18 Programmatic Tool Calling & Sub-Agents 6:59 Built on Strands + CLI Export #AWS #strandsharness #StrandsAgents