▶ Watch ↗19:43
Conference affiliation: Google DeepMind
Public sample repository owned by Wotherspoon’s verified GitHub account. It demonstrates currency conversion with an MCP exchange-rate tool backed by the Frankfurter API, ADK conversation and tool orchestration, and an A2A server and client for agent discovery and invocation. Repository ownership does not establish sole authorship or authorship of the underlying protocols and frameworks.
github.comAuthored by Wotherspoon, published August 24, 2022, with a Developer Advocate byline. Explains Cloud SQL Python Connector authorization through IAM, encrypted and identity-verified connections, reduced certificate-management work, and optional automatic IAM database authentication. Links interactive demonstration notebooks for PostgreSQL, MySQL, and SQL Server. The article establishes authorship of this guidance, not creation of the connector.
cloud.google.comCoauthored with Megan O’Keefe, published June 26, 2025. Builds a software-bug assistant for fictional coffee-machine company QuantumRoast using function tools, Google Search, a StackExchange integration, selected GitHub MCP tools, and an internal PostgreSQL ticket database exposed through MCP Toolbox. Separates model-generated tool requests from framework execution, restricts GitHub tools and token permissions, and links sample code and deployment guidance.
cloud.google.comCoauthored with Abhi Patel, published July 30, 2025. Introduces reusable prompts for Gemini CLI, an open-source terminal-based AI agent. Demonstrates pull-request review and investigation-before-implementation workflows, TOML command definitions, arguments, shell-derived context, namespaced commands, user and project scopes, and MCP prompt integration. Recommends sharing project-scoped commands through Git.
cloud.google.comCoauthored with Denise Kwan, published January 21, 2026. Announces a free Gemini CLI course developed with DeepLearning.AI. Describes lessons on setup, GEMINI.md and memory, MCP tool connections, extensions, software development, data analysis, content creation, and learning.
cloud.google.comEducational course announced by Wotherspoon and Denise Kwan as a collaboration with DeepLearning.AI. Its nine lessons connect project context, MCP integrations, and extensions to follow-along coding and non-coding workflows. The retained announcement establishes Wotherspoon’s participation in introducing the course, but does not explicitly establish sole course authorship or an instructor credit.
goo.gle · Source ↗Coauthored with Abhi Patel, published April 15, 2026. Explains specialized subagents with separate context windows, instructions, tools, and MCP servers that return consolidated results to the primary agent. Covers custom Markdown definitions, repository and extension distribution, explicit delegation, and parallel execution, while warning about faster usage-limit consumption and conflicting simultaneous edits.
developers.googleblog.comJack Wotherspoon works in developer experience at Google DeepMind, helping developers build with Gemini and Google AI Studio. His technical writing and open-source examples explain how to connect applications to databases, give AI agents useful tools, and turn recurring prompts into shared project workflows.
He joined Logan Kilpatrick’s team at DeepMind after five years at Google Cloud, where he worked as a developer advocate. In his announcement of the move, he described his new focus as making Gemini and AI Studio better to build with.
Wotherspoon’s earlier technical writing addressed a practical infrastructure problem: connecting Python applications to managed databases without making developers distribute SSL certificates or handle connection security themselves. His August 2022 guide to the Cloud SQL Python Connector explained how the library uses IAM permissions to authorize connections and encryption to protect them. It covered PostgreSQL, MySQL, and SQL Server, introduced optional automatic IAM database authentication, and linked interactive notebooks for trying the connector with each database engine.
By 2025, his examples also connected AI agents to the services needed to complete a job. With Megan O’Keefe, he coauthored a bug-assistant tutorial using Google’s Agent Development Kit (ADK). Their fictional coffee-machine company gives the example a concrete task: investigate software problems by searching the web, finding related GitHub issues, and searching or updating an internal ticket database.
The tutorial makes an important distinction between choosing an action and executing it. The model selects a tool and prepares its arguments; the framework makes the call. The model then interprets the result and decides whether to continue. The authors also restrict the GitHub integration to selected read-only tools and scope its access token to public repository data. Developers can see both how an agent takes action and how to narrow the actions available to it.
Wotherspoon’s open-source Currency Agent uses a smaller task to explain how three agent technologies fit together. A Model Context Protocol (MCP) server exposes an exchange-rate tool. ADK manages the conversation and invokes that tool when needed. An Agent2Agent (A2A) server makes the agent discoverable and callable by a separate client. Currency conversion keeps the example small enough to follow while showing the different jobs of tool access, agent orchestration, and agent-to-agent communication.
Wotherspoon’s work on Gemini CLI explains how developers can customize an agent’s behavior and share those choices with a team. Two articles coauthored with software engineer Abhi Patel address different parts of that workflow:
Wotherspoon’s teaching also reaches people using agents outside software development. In January 2026, he and Denise Kwan introduced a free Gemini CLI course developed with DeepLearning.AI. Its lessons cover project context through GEMINI.md, connections to external tools through MCP, and customization through extensions. Follow-along workflows apply those capabilities to coding, data analysis, content creation, and learning—showing how context and tool access shape what an agent can usefully do.
▶ Watch ↗19:43
Loading more talks and highlights…