Prototyping creates an early, simplified product or feature that makes behavior visible enough to test assumptions, gather feedback, and learn before full delivery. For AI products, the prototype should expose uncertainty rather than hide it behind polished screens. section 1.4 section 10.1.3
Use prototypes to shorten the distance between an idea and evidence:
- Start quickly when the opportunity is low-cost to test, the market rewards speed, or distinctive data creates an advantage. section 2.3.2
- Use real requests and realistic data where possible; fictional mockups can conceal problems that users will encounter in practice. section 2.3.2
- Treat the prototype as a learning instrument, not proof of product readiness: deploy to a limited audience, collect feedback, and iterate. section 1.5 section 2.3.2
- Simulate failures, personalization, and probabilistic outputs so testing reveals error tolerance, trust problems, and needed controls. section 10.1.3 section 10.4
Personalized playlists
A team builds a simple prototype that analyzes a small segment of user data and generates playlists for a limited group. Feedback shows that users want both familiar and novel songs, prompting a quick model adjustment. section 2.3.2
Table 10.1 Examples of errors you can simulate in your design artifacts
| Error type |
Example |
Observed user behavior |
Design adjustments needed |
| Hallucination |
AI falsely states CO2 emissions dropped 30% instead of 15%. |
Overtrusting users accept AI; skeptics reject AI entirely. |
Add confidence indicators and fact-checking prompts. |
| Regulatory misinterpretation |
AI incorrectly says Scope 3 reporting is optional. |
Legal experts catch it; others miss it. |
Include source citations and human validation steps. |
| Overconfidence in uncertain data |
AI estimates water consumption trends without full data. |
Some trust AI blindly; others demand citations. |
Use confidence scores and “Verify this data” prompts. |
| Formatting and structural problems |
AI-generated tables misalign financial figures. |
Users discard the entire report due to formatting flaws. |
Provide editable report structures for refinements. |
Simulated AI errors expose how users respond to hallucinations, regulatory misinterpretations, uncertain data, and formatting defects; resulting adjustments include citations, confidence indicators, verification prompts, and editable structures. section 10.1.3
