AI Design Process
Role: Senior Product Designer
Duration: 2 months
Organization: Tracfone
Problem
Leadership encouraged AI adoption to enhance design productivity and efficiency.
Approach
Develop AI-assisted design processes that is easy to implement across a large design team.
Our Approach
- Compile Alignment Content: Created a comprehensive knowledge base to train and guide AI-assisted design processes. This content updates automatically as new features launch, keeping the AI aligned with the latest product work.
- NotebookLM as Master Brain: Implemented NotebookLM to centralize and organize all design knowledge and insights. This enables the AI to provide accurate, contextually relevant guidance that remains consistent across the design team.
- Prompt Engineering: Developed a library of pre-designed prompts to reduce unnecessary AI-generated content and token usage. Includes a simple process for updating and adding new prompts as needs evolve.
- Specialized AI Agents (Gems): Created BMAD agents specialized in particular roles or tasks, enabling more focused and efficient AI assistance across different design workflows.
Initial Findings
After spending 4 weeks updating the alignment content and keeping the master brain current, we observed several key challenges.
- BORING: I feel like spent 4-5 hours per day just looking at Gemini waiting for responses.
- Low Output Quality: Generated content frequently contained errors and inconsistencies that required extensive manual correction, during iteration it would quickly hallucinate and start producing irrelevant or incorrect information.
- High Refinement Effort: Simple issues required disproportionate effort to resolve, often exceeding the time to create from scratch
- Organizing alignment content: Keeping the alignment content organized and up-to-date was challenging and time-consuming
- Quality Control Overhead: Significant review cycles needed to meet professional standards
- Product details: Relying on AI to do some of the heavy lifting in the design process reduced the teams knowledge on the product feature.
- Developer input: The introduction of these AI tools meant that anyone could do these prompts and often work was done by committee, with multiple people making suggestions to the prompt.
Ideation
After using the AI tools for ideation, we observed several key insights.
- Takes a long time: it could take several hours to generate and refine ideas to get satisfactory results.
- Simple examples: Seemed to work better with straightforward requests.
- Iteration was challenging: Refining ideas often required multiple rounds of feedback and adjustments but would often start to diverge from the original concept after a few iterations.
Overall, while AI-assisted design processes showed promise for certain tasks, the current state of the technology presented significant challenges that limited its effectiveness and efficiency in a professional design context.