… the artifact holds a wide toolset serving both basic and specialised workflows, and onboarding has to bring newcomers into working use of it. This pattern instantiates progressive disclosure across sessions and developing competence rather than within a single view.
Problem
When an interface exposes its full set of functions from the first session, newcomers have to sort out which controls matter before they can accomplish basic work, and accidentally triggering advanced options can corrupt their work or put the artifact into states they do not know how to recover from.
Feature-rich tools serve both novices doing straightforward tasks and experts running complex workflows. A single static surface compromises both: showing every tool, mode, and configuration option overwhelms newcomers and invites early errors, while permanently stripping advanced capabilities leaves experienced actors without the tools they need.
A layered interface partitions the toolset into cumulative levels of complexity. The initial layer exposes only the controls needed to complete a basic version of the work from start to finish, while higher layers add specialised functions without removing or rearranging what the earlier layer taught. Because the first layer is a complete working environment rather than a restricted demo, newcomers can produce real work immediately without running into advanced states they cannot interpret.
Two tensions govern how actors move between layers. First, hiding advanced functions improves speed and confidence on basic tasks, but makes those hidden capabilities harder to know about; showing them as disabled placeholders advertises what exists, yet reintroduces the visual clutter the layer was meant to remove. Second, while newcomers benefit from starting in a designer-chosen basic layer, actors resent having the interface rearrange itself automatically around them. Keeping spatial positions stable and letting the actor control when to step up to a fuller layer preserves their sense of control.
Solution
Partition the interface into cumulative layers, starting with a small initial layer that is self-contained for basic work and making each subsequent layer a superset of the one before it. Start newcomers in the initial layer, make the existence of fuller layers and the switch to reach them visible, and keep every transition between layers under the actor’s explicit control.
Never trap an actor in a reduced layer or force them through a tutorial to unlock the full interface; experienced actors arriving on a new account or device need an immediate switch to the full toolset. Do not move controls to new positions when a higher layer is enabled, and if the system suggests moving up a layer, keep the suggestion confined to a separate prompt rather than silently mutating menus in place.
Each layer should add a coherent group of functions. One switch that hides unrelated options across the artifact does not work as a layer; when only a few scattered options need hiding, use progressive disclosure for them. Name a layer by what it adds. Avoid skill-level names such as “advanced”. Do not hide the switch in settings alone.
Continuation
For capabilities held back from the active layer, omit them from the surface as unavailable actions when keeping the view uncluttered matters most, or show them in a disabled state with a path to unlock them when the actor needs to see what higher layers hold. Keep the layer switch visible and store the choice in settings as a lasting preference so stepping up stays under the actor's control, and use feature discovery to offer the switch when the actor's work reaches a hidden capability. Treat outgrowing the reduced layers as a step toward mastery, keeping control positions stable as fuller layers open…
All connections
- Precedes
- Unavailable actions, disabled state — For capabilities held back from the active layer, omit them from the surface as unavailable actions when keeping the view uncluttered matters most, or show them in a disabled state with a path to unlock them when the actor needs to see what higher layers hold.
- Mastery — Treat outgrowing the reduced layers as a step toward mastery, keeping control positions stable as fuller layers open
- Enables
- Onboarding — The artifact holds a wide toolset serving both basic and specialised workflows, and onboarding has to bring newcomers into working use of it.
- Instantiates
- Progressive disclosure — This pattern instantiates progressive disclosure across sessions and developing competence rather than within a single view.
- Enacts
- Learnability — removes the functions where newcomers' early mistakes happen until the actor is ready for them
- Adaptability — the interface changes with the actor's stage, heavier scaffolding first and less as competence grows
- Complements
- Feature discovery — Keep the layer switch visible and store the choice in settings as a lasting preference so stepping up stays under the actor's control, and use feature discovery to offer the switch when the actor's work reaches a hidden capability.
- Appears in
- Learning: Shaping the first encounter, step 4
In practice
Designing the layer boundaries
Define the first layer around a complete, end-to-end basic workflow rather than a random subset of popular controls. An actor in the first layer should be able to create, edit, and finish a simple item without hitting a dead end that forces them to switch layers mid-task. Each higher layer should add coherent groups of capabilities—such as automation, batch operations, or advanced formatting—while leaving the first layer’s controls in the same places.
Signalling what lies in higher layers
Keep the reduced surface clean: advanced controls shown disabled crowd it almost as much as the full interface. Put the layer switch in the workspace chrome and label it by what the fuller layer adds. Settings can store the choice, but a switch that only lives there is easy to miss. When the actor searches for a hidden command or reaches the limit of the basic workflow, use feature discovery or the search result to say the capability exists and offer the switch.
Delegation as the first layer
In agent-assisted tools the first layer is often a conversation. The actor says what they want and the agent uses the full toolset for them, so the layer is complete for simple work while most functions stay out of sight. The next layer lets the actor work on the artifact directly, for example by opening the code behind an app built through chat. The actor decides when to make that move, and the conversation stays open beside the artifact so they can hand work back.
Research on this pattern
- Training-wheels interfaces and error recovery: Carroll and Carrithers (1984) blocked the advanced, error-prone functions of a word processor for novices and found that participants completed tasks faster, spent far less time recovering from errors, and scored higher on post-task comprehension than participants using the full interface.
- Multi-layer progression and user-controlled interfaces: Shneiderman (2003) formulated multi-layer interface design as a universal-usability strategy in which users begin at layer 1 and advance at their own pace. Comparing a user-chosen personal interface beside the full default against system-driven adaptive menus, McGrenere, Baecker and Booth (2002, 2007) and Findlater and McGrenere (2004) found that user-controlled layering gave participants greater perceived control, navigability, and speed than automatic adaptation.
- Findability vs awareness in reduced interfaces: in a lab study with 30 novice users, Findlater and McGrenere (2007) found that a minimal first layer improved speed on basic commands at the cost of awareness of advanced ones, whereas keeping advanced commands visible in a blocked state overwhelmed novices—nine of the ten participants in the visible-but-blocked condition preferred the full interface instead.
- Spatial stability in adaptive interfaces: Gajos, Czerwinski, Tan and Weld (2006) traced user frustration with adaptive interfaces to lost spatial stability when menus reordered themselves; the design participants accepted copied promoted commands into a separate, clearly marked area while leaving the standard layout untouched.
- Guided automation in feature-rich software: comparing a copilot that completed tasks itself with one that automated trivial steps and guided the actor through the rest, Khurana, Su, Wang and Chilana (2025) found that most participants preferred to keep control of complex tasks and used the guided version to learn the application, especially for unfamiliar tasks.
Resources & references
- Carroll, J. M. & Carrithers, C. (1984). Training wheels in a user interface. Communications of the ACM 27(8)
- Shneiderman, B. (2003). Promoting universal usability with multi-layer interface design. Conference on Universal Usability
- McGrenere, J., Baecker, R. & Booth, K. S. (2002). An evaluation of a multiple interface design solution for bloated software. CHI 2002
- McGrenere, J., Baecker, R. M. & Booth, K. S. (2007). A field evaluation of an adaptable two-interface design for feature-rich software. ACM TOCHI
- Findlater, L. & McGrenere, J. (2004). A comparison of static, adaptive, and adaptable menus. CHI 2004
- Findlater, L. & McGrenere, J. (2007). Evaluating reduced-functionality interfaces according to feature findability and awareness. INTERACT 2007
- Gajos, K. Z., Czerwinski, M., Tan, D. S. & Weld, D. S. (2006). Exploring the design space for adaptive graphical user interfaces. AVI 2006
- Khurana, A., Su, X., Wang, A. Y. & Chilana, P. K. (2025). Do it for me vs. do it with me: investigating user perceptions of different paradigms of automation in copilots for feature-rich software. CHI 2025
- Forsey et al. (2024). Designing for learnability: improvement through layered interfaces. Ergonomics in Design