AI-Assisted Prototyping at Draftly: How LLMs Shrunk Design Iterations by 40%

AI · 5 min read

Draftly, a collaborative writing app, experimented with generative AI to accelerate prototyping. Designers used LLM prompts to produce placeholder microcopy, onboarding variants, and even initial error states. The core idea was to prototype multiple interaction variants in hours rather than days, then validate with rapid unmoderated tests.

To keep outputs usable, Draftly created prompt templates and post-processing rules: tone presets, brand lexicon enforcement, and unit tests that check for personally identifiable information. Designers treated LLM outputs as a first draft; every piece of generated content passed through a designer+PM review before being included in prototypes used for research.

The impact was measurable. The team reported a 40% reduction in time between ideation and testable prototype and a 25% increase in the number of variants explored during concept validation. A/B tests showed that some LLM-created microcopy outperformed human-written controls, but several hallucinated or contextually awkward variants required edits.

Draftly's takeaway is pragmatic: LLMs are powerful accelerants for divergent exploration but not a replacement for human judgment. They institutionalized guardrails, a short human review loop, and a tracking table that links prompts to outcomes, so future teams can reproduce or avoid problematic prompts.