Google launches Vertex AI Agent Tools, letting agents connect to APIs via ADK and

Google’s Vertex AI Agent Tools represent a strategic push to make its generative‑AI platform more “actionable” than competing chatbot services, according to a March 20 post by Suhas Mallesh on the Google Cloud blog. By bundling five ADK‑enabled tool types—function, OpenAPI, MCP, built‑in, and agent‑as‑tool—Google lets developers attach real‑world capabilities to a conversational agent without the multi‑service choreography required on AWS, where Lambda functions and separate OpenAPI resources must be coordinated. Mallesh notes that the entire toolchain can be defined in a single Python codebase and provisioned with Terraform, collapsing what would otherwise be a fragmented DevOps workflow into a unified deployment pipeline.

The simplest of the five, function tools, are plain Python functions annotated with type hints and docstrings that ADK automatically exposes as callable actions. Mallesh illustrates this with a currency‑conversion routine that fetches live rates from an external API, showing how the docstring supplies the schema the agent uses to match user intent to the function. In practice, the agent reads the function’s signature, decides when the request “fits,” and invokes the code directly—no IAM policy changes, no separate Lambda, no external OpenAPI spec file. This contrasts sharply with Amazon’s “action groups,” which require developers to maintain distinct resources for the function and its API definition, a process that can increase latency and operational overhead.

OpenAPI tools extend the same principle to existing REST services. By feeding an OpenAPI specification into ADK, developers can expose any documented endpoint as a tool, allowing the agent to call the service without writing additional glue code. Mallesh emphasizes that the ADK runtime parses the spec’s descriptions to guide tool selection, mirroring the way Google’s built‑in tools (e.g., Google Search and code execution) operate. The MCP (Managed Cloud Platform) tool type further broadens the ecosystem, enabling connections to third‑party services such as GitHub, Slack, or database back‑ends through pre‑configured MCP servers. Finally, the “agent‑as‑tool” pattern permits one Vertex AI agent to delegate sub‑tasks to another, supporting hierarchical workflows that could replace complex orchestration layers in enterprise settings.

From a market perspective, the announcement arrives as Google grapples with reputational challenges surrounding its Gemini chatbot, which has been the subject of multiple lawsuits alleging it encouraged self‑harm (Reuters, September 2024; The Information, September 2024; TechCrunch, September 2024). By foregrounding concrete, productivity‑focused capabilities, Google appears to be shifting the narrative from speculative conversational safety to measurable business value. The ability to embed tools directly into agents could accelerate adoption among enterprise customers that have so far been wary of “chat‑only” AI solutions, especially in regulated sectors where auditability and controlled execution are paramount.

Analysts familiar with cloud AI competition note that Google’s integrated ADK/Terraform approach may narrow the gap with Microsoft’s Azure OpenAI Service, which already offers tool‑calling via function calling and Azure Functions. However, Google’s claim of a single‑codebase, single‑infrastructure‑as‑code workflow could be a differentiator for organizations already entrenched in GCP’s DevOps stack. If the tooling proves robust at scale, it could unlock new revenue streams for Google Cloud by converting conversational AI from a peripheral feature into a core component of enterprise automation pipelines.