AI Project Policy

Appearance

Dev AI Toolkit ​

Purpose ​

Dev provides tools and documentation to help employees explore AI.

These resources are for employees who want to:

  • test an AI idea;
  • build a prototype;
  • create an AI-assisted workflow;
  • write a product specification;
  • ask for help choosing the right approach.

You do not need to know the exact technical solution before you contact Dev. Bring the idea or problem through the official Dev request form. Dev can help choose the right next step.

Agent Development Kit ​

The Agent Development Kit is a company-supported toolkit for building AI agents and AI workflows.

An AI agent is a system where AI can do more than answer a single question. For example, it may read information, call a tool, search a source, ask for confirmation, or continue through several steps before producing a result.

ADK helps developers build these systems in a consistent way. It gives structure to the work so the prototype is easier to understand, test, change, and hand off if it becomes a production candidate. See What ADK is.

ADK can help with prototypes such as:

  • an AI assistant that uses company-approved tools;
  • a workflow that searches information and summarizes results;
  • a prototype that calls tools or services;
  • an AI flow that asks a person to approve an action;
  • an idea that may later need production ownership.

Employees do not need to decide whether ADK is the right tool before asking for help. Dev can help make that decision.

Working with coding assistants ​

Dev also provides an ADK MCP server for people using AI coding assistants.

MCP means Model Context Protocol. In this case, it lets an AI coding assistant read the correct ADK documentation while helping with ADK code.

This is useful when a coding assistant is helping build an ADK prototype. It gives the assistant better information about how ADK works, so it is less likely to guess.

This can support a more exploratory coding style. Employees can try an idea with help from a coding assistant while still using company-supported documentation and patterns.

Learning concepts ​

Employees who are new to AI agents can read How agents work.

That page explains the basic idea behind an agent:

  1. a person or system asks for something;
  2. the AI receives instructions and context;
  3. the AI may answer directly;
  4. the AI may ask to use a tool;
  5. the tool result is returned to the AI;
  6. the AI continues until it has a final answer or result.

The ADK glossary explains terms used in the ADK documentation. Use it when a word in the documentation has a specific technical meaning.

Product specifications are welcome ​

Employees do not need to know how to prototype every AI idea.

If you have a clear idea but are unsure how to build it, Dev can help turn the idea into a product specification. A specification gives Dev enough context to understand the idea, discuss options, and propose an approach.

A useful early specification can include:

  • the problem or opportunity;
  • who the idea helps;
  • what the AI system should do;
  • what a successful result looks like;
  • examples of inputs and outputs;
  • known limits or risks;
  • any systems, data, or teams that may be involved.

The specification does not need to be perfect. It only needs to explain the idea clearly enough for discussion.

Relationship to this policy ​

Dev toolkit supports experimentation and specification work.

A prototype built with ADK or an AI coding assistant can remain an experiment when it uses approved environments and appropriate data.

When the work needs production system access, sensitive data, real-world impact, long-term support, production monitoring, production credentials, or specialist ownership, start handoff.