Best Practices

AI lab generation works best when you give the assistant a clear goal, good context, and steady feedback. These practices will help you get labs that are accurate, on-brand, and close to what you want on the first pass.

Start with strong research

Generation is only as good as the research behind it. The assistant picks up your organization’s published research on its own, so what you publish is what grounds every lab. Before you generate:

Publish accurate company research. This is what grounds the lab in your business, terminology, and brand style.
Research the relevant products. Published product research gives the assistant the real subject matter to build tasks and examples around.
Fix research first, not the lab. If generated content misrepresents your product, correct the underlying research and it will improve every future lab, not just this one.

Write a clear prompt

The prompt is your main instruction. Aim for a specific, outcome-focused description:

State the learning goal. Describe what a user should be able to do by the end, not just the topic. “Teach a developer to deploy a service and roll back a bad release” beats “a lab about deployments”.
Name the scenario. Mention the tools, product, or workflow the lab should center on.
Add any hard requirements. Call out things the assistant should include or avoid — specific commands, a particular flow, concepts to skip.
Keep it focused. One coherent objective per lab produces a tighter result than a prompt that tries to cover everything.

Choose settings that match your audience

Audience experience shapes depth and pacing. Pick Beginner for step-by-step fundamentals, Advanced when you can assume prior knowledge and move faster.
Duration guides scope. A 10-minute lab is a focused exercise; a 2-hour lab is a deeper, multi-chapter journey. Match it to how much you actually want to cover.
Autonomy sets the style. Lean Guided for precise, follow-along instructions; lean Autonomous when you want users to explore and problem-solve.

Review as you go

The plan-first, chapter-by-chapter flow exists so you can steer early — use it:

Review the roadmap before approving. It’s far cheaper to fix the shape of the lab at the plan stage than after chapters are built.
Check the first chapter carefully. It sets patterns the rest of the lab follows. Getting it right early prevents issues from compounding.
Give specific feedback in chat. When something’s off, say exactly what you want changed rather than regenerating and hoping. Precise guidance (“use our CLI instead of the web console”) produces better corrections. Send it once the turn ends — that’s when the assistant picks up your next instruction.
Answer the assistant’s questions directly. Clear answers to clarifying questions keep the generation on track.

Iterate after generation

A generated lab is a strong starting point, not a final artifact:

Toggle the docked assistant on a finished lab to make conversational edits as your product or training needs change — no need to regenerate.
Ask for one change at a time. A free-form edit is best-effort and scoped to what you asked for, so a specific request (“rename chapter 1 to Getting Started”) lands more reliably than a list of loosely related ones.
Make precise tweaks yourself in the lab editor when that’s faster than describing them — your edits and the assistant’s share the same working branch.
Always review before publishing. Treat generated content with the same care as hand-authored content — read it through and validate it before it reaches users.