Agent Development Kit at a glance
Google's Agent Development Kit 2.14.0 graduates A2A, sandboxed execution environments and evaluation from experimental, adds multi-run evals and a Microsoft Teams runner, and runs long workflows 23% faster in Google's own test.
What changed with Agent Development Kit
On November 7, 2026, Google published version 2.14.0 of the Agent Development Kit (ADK) for Python, the release that graduates the framework's core pieces from experimental to stable: A2A agent-to-agent communication, execution environments and sandboxes (local, E2B, and Daytona), evaluation, and core runtime modules. The release notes are headed November 4, 2026; the release itself was published on GitHub on November 7.
New in 2.14: MultiRunConfig runs each evaluation case several times, in parallel, and combines scores or pass rates across runs, while ToolArgsMatcher checks tool-call arguments with exact, case-insensitive, numeric-tolerance, or LLM-judged semantic matching. A Microsoft Teams runner and an OpenAI Decisions integration were added, and ContainerCodeExecutor gained input and output file support. One change is security-relevant: RunSkillScriptTool now runs skill scripts from validated paths and arguments instead of a free-form shell command string, closing a prompt-injection path. Google says a 30-node workflow over 200 turns runs 23% faster with 9% lower peak memory.
Agent Development Kit in numbers
Values are copied from the linked source on the date shown. They are not IA Compa measurements.
- Long-workflow speed gain (30-node workflow, 200 turns)as of · Source for Long-workflow speed gain (30-node workflow, 200 turns): github.com (opens in a new tab)
- 23% faster
Where to compare Agent Development Kit
Independent leaderboards and official resources for comparing this model yourself. Rankings change often; each site shows its own current data.
Why Agent Development Kit matters for businesses
Why it matters: ADK is one of the main open frameworks teams use to build agents that talk to each other and run code on a user's behalf. A2A and sandboxed execution leaving 'experimental' is the signal that multi-agent setups are ready for production work, and the script-execution tightening is exactly the kind of guardrail any agency running client agents should copy into its own stack.
Get AI News by email
Verified AI stories, each with its official source, on Monday and Thursday mornings. Add breaking news to hear about the biggest ones as soon as we publish them.
Every story links its official source
Free, in English or Spanish
Unsubscribe with one click, any time
Sources for this story
- ADK Python v2.14.0 release notes (opens in a new tab)Official
Google · github.com
- Agent Development Kit documentation (opens in a new tab)Official
Google · adk.dev
Last verified: . IA Compa summarizes public announcements for business readers. We are not affiliated with the companies mentioned. Always confirm details, pricing, and availability with the official source.
- #adk
- #ai-agents
- #a2a
- #open-source
- #evaluation