Jack Wotherspoon

Jack 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.

From database connections to agent tools

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.

Reusable workflows in Gemini CLI

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:

• Turn recurring instructions into shared commands. Their July 2025 introduction to custom slash commands demonstrates reusable prompts for pull-request review and planning. The review example gathers a pull request’s description and changes before preparing feedback; the planning example asks the agent to investigate before implementing. Project-scoped command files can be committed to a repository, so teammates can use and revise the same instructions alongside their code.

• Give delegated work its own context. Their April 2026 guide to Gemini CLI subagents describes specialists with separate instructions, tools, and context windows. A subagent can perform searches or other intermediate work and return a consolidated result, keeping that detail out of the main conversation. The article also explains the tradeoffs: parallel agents reach usage limits faster, and simultaneous edits can overwrite one another. Custom subagent definitions can be shared through project repositories or extensions.

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.