Four ways the system can offer to do part of the work: predictions, stored values, generated content, or recommended actions. The actor accepts, edits, or dismisses.
All four share the same basic loop (system offers, actor decides) but they vary in scope, timing, and how much trust the actor has to place in them. More system initiative means a greater need for transparency, easy dismissal, and earned trust.
The spectrum
Autocomplete narrows a known list as the actor types. One field, partial value, from a known list or index.
Autofill inserts stored values when a form loads or a field gains focus. Multiple fields, complete values, from the actor’s own data.
AI completion generates content inline while the actor writes.
Next-best action recommends the next step when context changes. Whole actions or decisions, from contextual reasoning or rules.
Autocomplete and autofill are operational: skilled actors barely notice them. AI completion and next-best action stay in conscious attention and require deliberate evaluation.
Resources & references
- Horvitz (1999) Principles of mixed-initiative user interfaces. CHI ‘99
- Parasuraman, Sheridan, Wickens (2000) A model for types and levels of human interaction with automation. IEEE Transactions on Systems, Man, and Cybernetics
Related patterns
Instantiates
- Suggestion — the broader pattern of system-generated recommendations; this spectrum focuses on the specific mechanisms
Enacts
- Agency — mirrors this spectrum; more system initiative demands more careful preservation of actor control
Complements
- AI completion — the spectrum this pattern belongs to
- Autocomplete — the spectrum autocomplete sits within
- Autofill — the spectrum this pattern belongs to
- Next-best action — the spectrum this pattern belongs to
Related
- Assistance
- Bot — powers AI completion and next-best action; the inline and ambient modes map directly to this group
- Transparent reasoning
Preceded by
- Good defaults — pre-population ends where the task needs the system to participate in the work itself — the assisted-completion spectrum picks up where a static guess stops sufficing