# AI Quest Challenges This document lists every challenge in the Deere AI Quest. Upload it into the Deere AI Assistant so it has context on what you're trying to build before you ask for help. ## Onboarding ### Console Account Confirmed ### Join AI Community ## Discover ### Take the Guided Tour Category: Discover · Difficulty: Beginner · Estimated time: 10 minutes Select Open in Console to enter the Deere AI Console, then complete the Show Me Around tour. The tour may start automatically or may require you to select it from the left navigation. Steps: 1. Select Open in Console below and sign in with Deere SSO if prompted. 2. If Show Me Around starts automatically, follow it. Otherwise, select Show Me Around from the left navigation. 3. Complete the tour, then note one Console area you expect to revisit. Why this matters: The guided tour gives new users a shared vocabulary and a mental map of the platform. Whether it starts automatically or manually, completing it makes every later challenge easier to follow. ### Run Your First Prompt Category: Discover · Difficulty: Beginner · Estimated time: 10 minutes Open an agent in Try It Out and send a safe, useful prompt, then recognize the session and response signals that show a real agent interaction occurred. Steps: 1. Select Open in Console to try the suggested agent, or choose another agent from the Registry. 2. Send a harmless question and read the response. 3. Return to AI Quest and sync progress; a session is the completion signal. Why this matters: A first prompt turns the platform from something you read about into something you can test. The same prompt-and-observe loop is how you validate agents throughout the quest. ### Discover the Platform Category: Discover · Difficulty: Beginner · Estimated time: 15 minutes Compare registered agents, tools, and models, inspect how each asset is described and owned, and identify one capability you could reuse instead of rebuilding. Steps: 1. Open Registry and visit Agents, Tools, and Models. 2. Open one asset profile and review its description and ownership. 3. Record one asset that could help a future use case. Why this matters: The Registry is the platform's reuse layer. Finding an existing, owned capability can save implementation time, reduce duplication, and connect you with the team that already supports it. ## Build ### Build Your First Agent Category: Build · Difficulty: Beginner · Estimated time: 90 minutes Build a minimal production HostedAgent from its required identity, Gateway, authorization, runtime, prompt, and model settings, then validate, deploy, and test it through Anvil GitOps. Steps: 1. Create a HostedAgent under the repository's .deere/prod tree. 2. Open a pull request and confirm the Anvil dry run succeeds. 3. Merge, confirm reconciliation, and send a safe test message in Console. Why this matters: A production agent is more than a prompt. This challenge teaches the complete delivery chain, from governed configuration and schema validation to reconciliation and a verified runtime response. ### Create an API Tool Category: Build · Difficulty: Intermediate · Estimated time: 60 minutes Create a reusable API Tool that exposes selected, safe REST or GraphQL operations to agents without requiring your team to build and operate an MCP server. Steps: 1. Open Registry -> Tools -> Create Tool and choose API Aggregation. 2. Name the Tool appropriately, then import an OpenAPI 3.x spec, introspect GraphQL, or add a small operation by hand. If you do not have an API, use https://petstore3.swagger.io/api/v3/openapi.json. 3. Validate a read-only operation and create the Tool resource. Why this matters: API Aggregation is the shortest path from an existing service to an agent capability. Validating read-only operations first also gives you a safer foundation before adding higher-impact actions. ### Create a Skill Category: Build · Difficulty: Intermediate · Estimated time: 30 minutes Package focused, reusable knowledge or procedures as a Skill that an agent can discover and load when the topic is relevant, then prepare it to be bound to an agent. Steps: 1. Open Skills -> Create Skill. 2. Write focused instructions for one reusable capability or workflow. 3. Save the Skill and confirm its profile appears in the Registry. Why this matters: Skills keep specialist knowledge reusable and focused. Separating it from the agent's permanent identity makes agents easier to maintain, extend, and govern. ### Build with the Deere AI Assistant Category: Build · Difficulty: Beginner · Estimated time: 20 minutes Open the Deere AI Assistant drawer in the Console and let it coach you through creating an Agent, Tool, or Skill end to end. Console only: the Assistant is a Console feature with no repository equivalent. Steps: 1. Open the Deere AI Assistant drawer from any page in the Console. 2. Tell it which kind you want help with: an Agent, a Tool, or a Skill. 3. Follow its guidance to complete the creation flow, then confirm the new resource appears in the Registry. Why this matters: The Deere AI Assistant lowers the barrier to a first Agent, Tool, or Skill by coaching newcomers through the same decisions an experienced builder would make, without requiring prior platform knowledge. ## Connect ### Wire Up a Deere MCP Server Category: Connect · Difficulty: Intermediate · Estimated time: 30 minutes Connect an MCP-compatible IDE or assistant to the Deere AI Console MCP server, complete Deere SSO, and verify that the client can discover an allowed platform capability. Steps: 1. Configure an MCP-compatible client with console-api.deere.com/v1/mcp. 2. Complete Deere SSO authorization when prompted. 3. Confirm the client can discover and invoke an allowed Console capability. Why this matters: MCP brings the platform's registry, Help content, and operational knowledge into the tools where technical work already happens, reducing context switching and repeated setup. ### Configure Downstream-Tool Auth Category: Connect · Difficulty: Intermediate · Estimated time: 45 minutes Intermediate challenge: choose and configure the correct downstream authentication pattern so an agent can call a protected tool with the intended user or service identity without exposing credentials. You need this when a Tool reaches a system that requires sign-in, a service identity, or a token exchange instead of anonymous access. Steps: 1. Identify the downstream tool and the credential it requires. 2. Configure the appropriate provider connection or run credential. 3. Test the agent path and confirm the downstream call is authorized without exposing a secret. Why this matters: An agent that can reason but cannot securely reach the systems it needs is not useful in production. This challenge builds the identity boundary that makes downstream automation trustworthy. ### Create a Provider + Connection Category: Connect · Difficulty: Intermediate · Estimated time: 30 minutes Intermediate challenge: create and authorize a provider connection for an agent that needs access to a protected system. You need this when the agent or one of its Tools must act with a user's identity or a configured service identity; complete the consent or secret exchange, then retry the request with the connected identity. Steps: 1. Open the agent and identify the protected Tool or operation that needs authentication. 2. Create or select the reusable Provider definition and choose its authentication kind: three-legged OAuth, two-legged OAuth, or API key. 3. Create a Connection from that Provider for the intended user or service principal; keep credentials out of prompts and source code. 4. Attach the Provider requirement to the agent or Tool, then authorize the connection when the runtime requests it. 5. Retry a safe request and confirm the protected operation succeeds with the connected identity. Why this matters: Connections let each user authorize access explicitly and within scope. That keeps protected systems protected while still allowing an agent to act on the user's behalf. ### Connect Tools and Skills to Your Agent Category: Connect · Difficulty: Intermediate · Estimated time: 30 minutes Connect the registered API or MCP Tool and Skill from Build to a HostedAgent, deploy both bindings, and prove that the agent can use the Tool with the Skill's instructions. Steps: 1. Open the managed agent's Configure tab. 2. Add a ToolBinding for the registered API or MCP Tool and a SkillBinding for the Skill. 3. Save through the repository PR flow, then test the agent with a request that uses both capabilities. Why this matters: Tool creation and ToolBinding solve different problems: registration makes a capability reusable, while binding gives one agent permission and context to use it. This is the step that turns an asset into agent behavior. ## Innovate ### Configure Subagents Category: Innovate · Difficulty: Advanced · Estimated time: 90 minutes Configure specialist subagents with explicit responsibilities and bindings, then orchestrate a bounded scenario that delegates work and combines the specialist results. Steps: 1. Define the specialist responsibilities and the handoff contract. 2. Configure the subagent tool bindings for the orchestrator. 3. Run a scenario that demonstrates parallel or delegated specialist work. Why this matters: Subagents make complex work easier to divide and evolve, but delegation without clear contracts creates confusion. This challenge practices the boundaries and aggregation needed for reliable multi-agent systems. ### Bring Your Own Remote Agent (A2A) Category: Innovate · Difficulty: Advanced · Estimated time: 120 minutes Connect a remote agent through the A2A client pattern, establish a clear responsibility boundary, and demonstrate a traceable handoff that returns a useful result. Steps: 1. Choose a remote agent with a documented Agent Card. 2. Configure the A2A client integration and required authorization. 3. Run a scenario that proves the remote agent returns a useful result. Why this matters: A2A lets teams compose capabilities across runtime boundaries. The real skill is designing a handoff that preserves authorization, context, failure behavior, and ownership. ### Measure Your Agent with Live Evaluations Category: Innovate · Difficulty: Intermediate · Estimated time: 45 minutes Attach an evaluation to an agent you manage so real traffic is scored continuously, then read the resulting pass rate to see whether the agent behaves the way you intended in production. Steps: 1. Open the Quality surface of an agent you manage. 2. Configure a live evaluation with the assertions that matter for your use case. 3. Let real traffic score, then review the pass rate and note what it tells you. Why this matters: Shipping an agent is not the same as knowing it works. A live evaluation turns every real interaction into evidence, so regressions show up as a falling pass rate instead of as a surprised user. ### Create and Run a Golden Evaluation Category: Innovate · Difficulty: Advanced · Estimated time: 60 minutes Create a golden dataset with repeatable scenarios and assertions, run it against a production HostedAgent, and use the results to identify an evidence-based improvement. Steps: 1. Create representative scenarios with expected behavior. 2. Run the golden evaluation against the agent configuration. 3. Inspect the results and document one change or follow-up based on evidence. Why this matters: Evaluation turns “it seems to work” into a repeatable quality signal. Golden scenarios give teams a shared way to detect regressions before expanding an agent's scope. ### Publish to the Registry and Get Reused Category: Innovate · Difficulty: Advanced · Estimated time: 60 minutes Prepare a Tool, Skill, or agent capability with clear ownership, audience, and usage guidance, publish it through the managed workflow, and prove that another consumer can reuse it. Steps: 1. Choose a Tool, Skill, or agent capability that is safe to reuse. 2. Complete its description, ownership, and usage guidance before publishing. 3. Bind or recommend the asset to another agent and verify a real consumer uses it. Why this matters: Publishing is successful only when someone else can find, understand, trust, and use the asset. Reuse turns one team's work into shared platform capability instead of isolated project code.