Disclaimer: Company name and branding have been changed for legal reasons.
Context
Software development requires synthesizing information from many sources, such as codebases, requirements, architecture, design decisions, meeting outcomes, and more.
Most AI assistants lack access to this context, resulting in incomplete or incorrect output. This shifts effort back to engineers for correction and reduces efficiency, trust, and adoption.
Challenge
Reimagine how engineers interact with an AI coding assistant once external context becomes accessible within the IDE.
My Design Footprint
Research, UX/UI design
Signs of Success
- 56% higher suggestion accuracyDuring the testing, engineers reported fewer irrelevant outputs and less time spent correcting code
- 2x faster task completionTesting showed engineers finishing tasks twice as quickly, reaching outcomes with less effort
- Improved interaction80% of engineers confirmed in-chat interaction sped up their coding process and made it smoother
Process
Research
I started this project by researching the current state of AI coding tools and interviewing developers and product managers who code with AI daily. I also drew on my own hands-on experience.
Key design principles
While reflecting on ongoing trends, I defined UX principles essential when designing for AI
Build Trust
Be transparent in actions and decisions and provide clear explanations
Right Level of Control
Give users control and let them choose between full automation and co-pilot mode
Context Is The King
Use only relevant, clean, and well-structured context
Pain points
Among all uncovered pain points, I prioritized three the most common
- Missing, excessive, or irrelevant context leads to incorrect or incomplete suggections
- Bring all the relevant context (docs, tickets, and chats etc.) into IDE is challenging
- Clarifying and explaining context slows the development process
Proposed solution
AI assistant connects to External Memory (codebase, project management tools, messengers, documentation, etc.).
The workflow is divided into three phases:
- Research. AI assistant searches connected sources, retrieves relevant context, and brings it into the chat
- Plan. AI assistant analyzes inputs and suggests a step-by-step implementation plan
- Implementation. AI assistant handles each step of the plan
At any point, users can add missing context, provide comments, or adjust guidance.
Some interactions are streamlined directly in chat to speed up collaboration.
Integration with Project management systems
Engineers can now integrate their project management platform (Jira, Monday, Asana, Linear, and more) and add issues directly to the context.
For quicker access, this action is also available in the Quick Start menu.
Context management
AI assistant scans connected memory sources to find relevant context
When Context Control is enabled, the AI assistant asks user to review the suggested context before continue
Users can unselect items, add extra context, or leave comments directly in the chat
Implementation plan
The Assistant suggests an implementation plan.
Users can review, unselect, edit, or add items before continuing.
User flow
