PTN—05
Guided first run
A structured first session that produces a real result from the person’s own context instead of a tour.
The first run is where AI products squander the most goodwill. The classic mistake is the tour: five tooltips pointing at chrome, a video nobody watches, and then the same blank box as before, now with expectations raised. A guided first run takes the opposite bet — it structures the opening minutes around producing one genuine result, because a person who gets real value in session one has a reason to start session two.
The strongest first runs collect a small amount of context and spend it immediately. Connect a repo, upload a document, name your role — then use that context to do something visibly specific to this person: index the codebase and answer a question about it, summarize the actual document, draft in their stated domain. The setup cost is repaid on the spot, which is the only pricing users accept for onboarding friction.
Guidance must have an exit at every step. Power users arriving with a task in mind will resent any rail they cannot step off, and a first run that holds the product hostage until its checklist completes converts impatience into churn. Make every step skippable, keep the whole thing under a few minutes, and treat the person’s first self-authored prompt as the graduation signal — the moment guidance should get out of the way and stay out.
When to use
- The product needs setup — connections, uploads, permissions — before it can show its real value.
- First-session abandonment is high and analytics show users never send a second prompt.
- The mental model differs enough from chat-as-usual that unstructured exploration misleads.
- You can produce a personally relevant result within the first few minutes of use.
Design considerations
- 01Drive toward one real output in the first session — a summary of their file, an answer about their code — not a completed tour.
- 02Ask for context only when you will use it in the next step, and show the payoff immediately.
- 03Make every step skippable and the whole flow escapable without penalty or nagging.
- 04Set the contract during the run: one line on what the product is for and one on where it is weak.
- 05End at a live composer seeded with relevant starters, not at a dashboard of unopened features.
- 06Detect the first self-authored prompt and retire the guidance permanently — do not re-trigger it on updates.
Pitfalls
- ✕Tooltip tours that explain chrome instead of producing value, then dump the person at the same empty box.
- ✕Mandatory setup walls demanding connections and permissions before the product has demonstrated anything.
- ✕Demo-data first runs that impress in the abstract and leave the person still not knowing what to do with their work.
- ✕Onboarding checklists that resurrect themselves after every release, re-guiding users who graduated months ago.
In the wild
- Cursor
- First run opens on the person’s own codebase and indexes it, so the first AI interaction answers questions about their actual files rather than a demo project.
- Granola
- Onboarding is built around the first real meeting — connect the calendar, join a call, and the first artifact is notes from your own conversation.
- Perplexity
- Deliberately minimal first run: no setup, straight to the query box, with example queries doing the teaching — the guided path is simply a good first search.
- Linear
- Onboards new workspaces by having you create real projects and issues during setup, so the tool is populated with your actual work from minute one.