How to design an AI feature
Design a good AI feature by defining the user task, stating what the AI can and cannot do, designing input/output affordances, adding fallback states, testing with real users, and iterating on confusion signals. Use this as a design review or PRD checklist.
Start here
How to spec it
Define the user task
Name the job-to-be-done in one sentence. What does the user need to accomplish, and what would success look like without AI?
Clarify capabilities and limits
State what the model can do, what it cannot do, and what data it uses. Surface limits before users hit them, not in a buried FAQ.
Design input and output affordances
Spec composer defaults, attachment flows, streaming, and output surfaces (text, artifacts, cards). Match affordances to task risk.
Add fallback and failure states
Design for wrong answers: graceful degradation, human handoff, retry, and honest failure disclosure, not only happy paths.
Test with real users
Run usability tests on mental models and trust calibration, not only task success. Watch where users over-trust or abandon.
Iterate on trust signals
Use correction rates, undo usage, and session return as feedback. Tighten citations, checkpoints, or autonomy as patterns demand.
Patterns to ship first
Streaming
Show replies token-by-token as they generate
Citations
Attach verifiable sources to generated claims
Human in the loop
Require human approval before AI acts
Failure Disclosure
Honest signaling of AI limitations
Prompt Starters
Example prompts on an empty chat screen
Regeneration Carousel
Swipe bot responses
Patterns by design step
Deeper build playbooks
Frequently asked questions
What is an AI feature?
An AI feature is any product capability where a model generates, transforms, or acts on user input: chat replies, summaries, agents, copilots, or generative UI, not static deterministic UI alone.
What should I spec first?
Start with the user task and failure modes. Capabilities, composer affordances, and trust patterns follow once you know what goes wrong when the model is wrong.
How do trust patterns fit in?
Trust patterns (citations, confidence, human checkpoints, failure disclosure) belong in the same spec as core UX. They are not a polish pass after launch.
How do I test AI UX?
Test whether users calibrate trust correctly: do they verify before acting? Do they know when to escalate? Task time alone misses over-reliance on wrong answers.
What are common AI feature design mistakes?
Hiding limits, auto-applying consequential actions, chat-only UI for structured tasks, and shipping without empty, loading, error, and low-confidence states.