Capability Boundary
Honest about what it cannot do — before it tries
Before running, the system declares which parts of the task it can confidently handle and which it will refuse or substitute. Honest pre-flight disclosure beats confident hallucination — and offering a workaround beats a flat refusal.
Framing
The problem
AI systems prefer confident hallucination to honest refusal — users discover the limit by being misled.
The pattern
Surface a pre-flight capability sheet: what the system will do, what it will refuse, and how the user can supply the missing piece.
Why chat breaks here
Chat encourages a "yes-and" register that hides limits behind plausible-sounding answers.
Risks
Over-cautious capability lists become friction theatre and erode trust through false modesty.
Avoid when
The model is reliably within capability for the task — disclosure adds noise without value.
Use when
Limits are real and silent failure (confident hallucination) is costlier than honest refusal.
DOPE evaluation
- Directability
- Choose to proceed within the limits, supply the missing input, or abandon the task
- Observability
- The system declares its scope for this task before any tokens are spent
- Predictability
- Users know what part of the answer to trust and what to verify themselves
- Explainability
- Each refused capability comes with a reason and a workaround when one exists
In the wild
- Cursor 2.1 Plan Mode (Cursor) — Plan Mode emits an editable plan with file paths and a to-do list before any code runs — a pre-flight surface where the agent can flag what it cannot do safely. The closest in-product capability-boundary UX in 2026.
- ChatGPT Model Picker Disclaimers (OpenAI) — When a user selects a vision/voice/web-disabled model, the UI disables the relevant input (uploads/web) and surfaces a tooltip explaining why. Pre-flight scope disclosure for the lay user.
- Microsoft Copilot Studio (Microsoft) — Explicit "disambiguate intent" + capability declaration UX in agent setup — the clearest enterprise expression of the pattern as a configurable boundary.
FAQ
When should I use the Capability Boundary pattern?
Limits are real and silent failure (confident hallucination) is costlier than honest refusal.
When should I avoid the Capability Boundary pattern?
The model is reliably within capability for the task — disclosure adds noise without value.
What problem does Capability Boundary solve?
AI systems prefer confident hallucination to honest refusal — users discover the limit by being misled.
Why is chat the wrong fit for this?
Chat encourages a "yes-and" register that hides limits behind plausible-sounding answers.
Related patterns
- Often paired with: Disambiguation Branch — When the AI is unsure what you mean, branch. When the AI cannot do part of it at all, declare the boundary.
- Extends: Confidence Signals — Confidence per claim plus capability boundaries per task — both refuse to fake competence.
- Often paired with: Intent Preview — Preview the plan, then declare which steps the AI cannot perform. Honest preview includes honest limits.