Use a Thinking Path

Once you've built a Thinking Path with steps connected and an Entry Step set, attach it to an assistant and deploy before learners can use it.

Examples use the Software Decision Walkthrough Thinking Path from Getting Started with Thinking Paths.

Before you start

• Thinking Path created with at least one step → Getting Started with Thinking Paths
• Steps connected in the graph view and Entry Step set
• The Framework(s) linked to your steps must be synced → Getting Started with Frameworks

Step 1 — Create an assistant

Navigate to your course → Assistants tab → Create Assistant.

Choose Thinking Path as the assistant type. This is the highest-structure option — the assistant walks learners through your path step by step instead of answering freeform questions.

Select the Thinking Path to attach. The Thinking Path defines the assistant's behaviour — no system prompt needed.

In our example: Select Software Decision Walkthrough. Name the assistant Build vs. Buy Advisor.

Step 2 — Configure settings

Access Level

Controls who can use this assistant:

The assistant's access level cannot be more permissive than the course it belongs to.

Conversation Starters

Suggested starter prompts shown to learners when they first open the assistant. These reduce the "blank page" friction of starting a conversation.

In our example:

• "I have a software decision to work through — can you guide me?"
• "We're deciding whether to build our own tool or buy something off the shelf."
• "I need to write a Build vs. Buy recommendation for my team."

Up to 5 conversation starters. Keep them specific enough to be useful.

AI Model

Choose which underlying language model powers the assistant. The default recommendation is Gemini 2.5 Flash.

For analytical assistants guiding structured workflows, keep Temperature low (0.0–0.3). Higher values make responses less consistent.

Step 3 — Test before deploying

Use the test assistant to check behaviour before deploying. The test assistant runs the same configuration but is only accessible to you.

What to test:

• Does the assistant start at the correct Entry Step?
• Does each step advance when the checklist is satisfied?
• Does the linked Framework apply at the right moment (Step 2 in our example)?
• Does the assistant enforce the Rules from the Guardrails tab?
• Does the Final Output check work at the end?
• Does the Tutoring Approach come through in the assistant's tone and coaching style?

Test with realistic learner inputs. Try giving vague answers, skipping ahead, and asking for the answer — these edge cases reveal gaps in your path.

Step 4 — Deploy

When you're satisfied with the test, click Deploy from the assistant's detail view.

Deployment publishes the current version. Learners will see the deployed version immediately.

If you continue editing the Thinking Path or its steps after deploying, your changes are in draft. You need to deploy the assistant again to push those changes live.

What learners experience

Learners see this assistant as a Nudge — a guided experience, not a freeform Q&A.

In our example — what a learner sees when they open Build vs. Buy Advisor:

1. The assistant opens with the conversation starters you configured
2. Once the learner starts, the assistant begins at Define the problem — asking them to describe the technology decision they're facing, what success looks like, and what constraints exist
3. When the learner has written a clear problem statement with success criteria (the step's checklist), the assistant advances to Evaluate the options
4. At this step, the assistant applies the Build vs. Buy Software Decision framework — collecting internal build capability (yes/no) and estimated build cost, then guiding the comparison
5. When the evaluation checklist is satisfied, the assistant advances to Make a recommendation
6. The learner writes their recommendation. The assistant checks it against the checklist and the Final Output definition
7. When both are met, the path is complete — the learner has a documented Build vs. Buy recommendation

Throughout the workflow, the assistant enforces the Rules, uses the Reasoning Questions internally, and coaches according to the Tutoring Approach.

Update or delete

To edit the path, its steps, or the assistant configuration, see Update a Thinking Path.

Next steps

For full details on assistant configuration, access levels, cohorts, and lifecycle management, see Getting Started with Assistants and Deployment and Versioning.