Assistance

Assistance is a configuration of agency where the locus is human-centric, coupling is tight, granularity is fine-grained, and dynamics are typically static. The human drives the interaction arc; the system intervenes selectively at each cognitive step — raising a signal, explaining something, narrowing the options, supporting execution, or helping the user look back at what just happened.

Assistance across interaction stages

Perceiving

Help the user notice what matters before they interpret or decide. Examples:

• Summarize situation: compress a large field of signals into a first-pass overview so the user can grasp the overall state before inspecting details.
• Guide the attention: highlight salient regions, anomalies, or cues so important signals stand out against distraction and clutter.
• Extend senses: translate otherwise hidden or imperceptible signals into forms the user can notice and use.
• Prime for possibilities: prepare the user to look for particular cues or conditions before scanning begins.

Knowing

The user has noticed something — a change, a cue, an anomaly. Help them understand what it means: state, process, stakes, context. Examples:

• Answer questions: give users a low-friction way to ask about the document, situation, or environment they are already in.
• Help to understand the state of things: explain labels, trends, workflow position, stakes, or likely consequences so the current situation makes sense.
• Help to recall: refresh relevant history, preferences, or prior context that the user has seen before but cannot fully hold in memory.
• Help to make sense of it all: support deeper conceptual understanding, so users build mental models they can reuse later.

Planning

The user understands the situation. Helps them choose what to do. Examples:

• Suggest next actions: narrow a large action space to plausible next moves without hiding the broader set of available actions.
• Generate options in open-ended domains: propose candidate approaches or stepwise plans where the space of possible actions is not fixed in advance.
• Visualize outcomes ahead of time: preview or simulate likely consequences so users can compare futures before they commit.
• Rank outcomes and support decisions: recommend or sort options while still exposing enough rationale, tradeoffs, and uncertainty for user judgment.

Performing

The user has decided. Help them carry out the action well. Examples:

• Help practice: create low-stakes rehearsal where users can build skill with coaching before the real stakes apply.
• Help keep on point: provide live cues during action so users stay within relevant boundaries while still doing the task themselves.
• Show best practices: place meaningful benchmarks inside the act so users can compare and regulate their performance in the moment.

Reflection

Help the user look back — comparing recent behaviour to goals and values, identifying drift, and deciding what to improve. Examples:

• Help review performance: bring the just-finished action back into view so the user can assess what happened rather than only move on.
• Reconnect behavior to goals or values: compare recent choices or habits against the standards, priorities, or commitments the user cares about.
• Decide what to improve next: turn hindsight into a concrete next focus for future attempts rather than leaving reflection at the level of summary.
• Structure reflection: lower the barrier to retrospective thinking by giving users a shape for the review.

Resources & references

• Noessel (2026) Designing Assistant Technology