Google for Developers’ Post

Build production-ready, multi-agent applications with a graph-first workflow engine designed specifically for Go. The Agent Development Kit for Go 2.0 simplifies application mapping by running single agents and complex graphs on the exact same execution model. Manage complex states and runtime variations with built-in resilience features: 🛠️ Dynamic orchestration written in plain Go 🤖 Native human-in-the-loop primitives to safely pause and resume execution 📈 Built-in retry policies featuring exponential backoff and full jitter 🌲 Unified telemetry spans for absolute tracking visibility Upgrade your existing codebases safely with a highly additive framework that keeps your existing import paths intact: https://goo.gle/444xsMk ADK Docs: https://goo.gle/4gUqaCe SDK: https://goo.gle/441gfn1 Go pkg: https://goo.gle/4vHE9jF Code Samples: https://goo.gle/4ezMks8

  • The light-blue Go gopher mascot, wearing safety goggles and a tool belt, gestures toward the title "ADK Go v2.0" and a central icon. To the right, a code editor displays Go programming language imports and functions, while a diagram below illustrates "Graph based workflows" and "Human-in-the-loop review" with connected nodes. The background is a dark blue technical landscape with glowing circuitry and isometric cubes.

This is where AI is heading: from single prompts to systems that can reason, coordinate, recover, and operate reliably in production. Building agents is becoming easier. Building agents that can handle complexity, uncertainty, and real-world business constraints is where the challenge begins. The biggest advantage won't come from access to AI tools—it will come from the ability to design systems that consistently deliver outcomes when the stakes are real. The future belongs to builders who can prove they can make it work. If you're looking to identify and evaluate talent through real-world challenges instead of résumés alone, explore https://basestudy.app/ and discover how proof-based evaluation reveals capability before you make the hire.

Like
Reply

The decision to use the same execution model for both single agents and complex graphs stood out. That seems like it could make it much easier to start with a simple workflow and evolve it over time without reworking the application's architecture. How does ADK 2.0 handle versioning or migration when a production workflow graph changes while long-running executions are still in progress?

Like
Reply

Agent runtimes are becoming structured, observable, and resilient. But the code, workflow graphs, tools, and integrations that agents generate still need to conform to architectural intent.

Like
Reply

Looks interesting! Definitely going to check this out.

Like
Reply

Nice release. Keeping compatibility while upgrading the execution model is a smart approach for adoption

Like
Reply

Really solid release. Making multi‑agent workflows simpler, resilient, and production‑ready is a big step forward for builders. Love seeing tools that reduce complexity so teams can focus on creating real value.

With Gemini Pro 3.1 I created a new AI that is now running locally on a Pentium system that's too old for Windows 8 to recognize. It's even running Gemini in the browser right now. All the modern standards isn't required. A machine running in Mhz not Ghz

Like
Reply

Another great development from Google 👏

Like
Reply
See more comments

To view or add a comment, sign in

Explore content categories