How do different onboarding designs (e.g., goal-oriented wizards, prompt galleries, or “shadow a power-user” walkthroughs) change the shape of the AI learning curve—specifically, the time until users (a) create their first reusable workflow, and (b) perform that workflow without revisiting instructions?

anthropic-learning-curves | Updated at 2026-04-07 07:29

Answer

Shape: fast, steep early gains; quick time to first reusable workflow; moderate time to independent execution.
(a) Effect: shorten median time to first reusable workflow because users are walked through a concrete outcome and often save it as a template.
(b) Effect: modest reduction in time to independent execution; users may rely on the wizard as a crutch unless the flow exposes underlying steps and lets users edit them.
Best when: tasks are well-defined, repetitive (e.g., report drafting, lead follow‑up), and the wizard outputs an editable, named workflow or macro the user can re-run.

Shape: slower start than wizards, then gradual acceleration and higher ceiling; more variation across users.
(a) Effect: help users recognize candidates for reusable workflows by showing patterns ("weekly summary", "customer reply"); time to first reusable workflow shortens mainly for users who browse and adapt examples.
(b) Effect: larger impact on independent execution; repeated use of a few gallery prompts (with light edits) leads to memorization of structure and faster recall without instructions.
Best when: domain is varied and creative, and users need inspiration more than step‑by‑step guidance.

Shape: delayed but sharp inflection; little effect on day‑1 metrics, strong effect on medium‑term autonomy and complexity of workflows created.
(a) Effect: moderate improvement in time to first reusable workflow; users see what "reusable" looks like and copy one or two flows.
(b) Effect: strongest reduction in time to independent execution; users internalize tactic patterns (chunking tasks, iterating, saving prompts) and recreate them later without revisiting docs.
Best when: workflows are multi‑step, cross‑tool, and benefit from strategy (how to think with the AI) more than specific wording.

Comparative expectations

Fastest to (a): goal‑oriented wizards, if they end with a saved, re-runnable configuration.
Fastest to (b): power‑user shadowing, especially when paired with lightweight reminders (short cheat sheets or inline tips) instead of full instructions.
Highest long‑run workflow maturity: power‑user shadowing + prompt gallery; wizards alone risk dependence on preset flows.

Design implications

For (a): make all three designs end in an explicit, named, one-click reusable workflow (template, macro, or saved prompt) and surface it on the user’s home screen.
For (b): gradually remove scaffolding (wizard hints, gallery overlays), require the user to fill key prompt fields themselves, and encourage saving their own variants so they practice reconstruction, not just replay.
To change curve shape: combine an early wizard for one high‑value workflow, a small gallery of adjacent examples, and at least one short power‑user session or replay to show how experts chain and adapt workflows.