Google ADK 2.14 Adds Teams Integration and Multi-Run Evals

Google released Agent Development Kit 2.14.0 on November 7, 2026, graduating 26 experimental modules to stable status and introducing Microsoft Teams integration for ADK agents. The release brings concurrent evaluation capabilities, semantic tool argument matching, and performance improvements that Google claims deliver 23% faster execution for long multi-turn workflows.

Release Date and Availability

Officially released on November 7, 2026, ADK 2.14.0 marks a significant maturity milestone for Google's agent development platform. The release transitions execution environments, A2A (Agent-to-Agent), evaluation classes, and core runtime modules from experimental to stable status. Sixteen toolset integrations also graduated alongside the 26 auth, environment, app, CLI, and cache modules, signaling Google's confidence in production readiness for enterprise deployments.

Microsoft Teams Integration and Multi-Run Evaluation

The headline feature is TeamsRunner, a new component that embeds ADK agents directly inside Microsoft Teams. Organizations can now deploy conversational agents within their existing collaboration environment without building custom integration layers. Alongside the Teams runner, Google introduced MultiRunConfig, which executes each evaluation case multiple times concurrently. This capability enables developers to test agent reliability across varied execution conditions and catch non-deterministic failures that single-run tests miss.

A new ToolArgsMatcher component adds flexible argument validation with four modes: exact matching, case-insensitive comparison, numeric tolerance checks, and LLM-backed semantic validation. The semantic mode allows developers to verify that tool arguments convey the correct intent even when phrasing differs, reducing false negatives in evaluation workflows.

Performance Claims and Benchmarks

Google states that a 30-node workflow executing 200 turns runs 23% faster with 9% lower peak memory consumption in ADK 2.14.0. The release notes provide no baseline hardware specification, no comparison against a named prior version, and no second benchmark to validate the claim across different workflow patterns. The performance improvement targets long multi-agent system interactions where agents chain multiple reasoning steps, though developers will need to validate gains on their own infrastructure.

Breaking Changes and Additional Features

Three breaking changes accompany the release. Enum tool arguments now arrive as enum members rather than raw strings, requiring type adjustments in existing agent code. VertexAiSessionService now raises ValueError when passed unrecognized keyword arguments, tightening input validation. RunSkillScriptTool no longer accepts a free-form command parameter, restricting execution to validated Python, shell, and Bash scripts from a designated skill directory to close a potential security gap.

The release also ships MongoDbMemoryService with vector-search retrieval and TTL expiration for agent memory management, plus an adk generate_agent_card CLI command that builds A2A agent cards from the command line. OpenAI Decisions third-party integration is included, though the release notes provide no detail on its functionality.

What This Means

ADK 2.14.0 represents Google's most mature agent development release to date, with 42 components and features moving from experimental to stable in a single update. The Microsoft Teams integration addresses enterprise demand for agents inside existing collaboration tools, while concurrent evaluation and semantic argument matching target reliability engineering for production deployments. Developers upgrading from earlier ADK versions will need to address the three breaking changes, particularly the RunSkillScriptTool security restriction. The claimed performance improvements are noteworthy but lack sufficient benchmark detail for independent validation, leaving teams to measure gains on their own workloads before committing to the upgrade.