Context-Rich Auto-Complete

Context-Rich Auto-Complete is a concept for dramatically improving AI suggestions by leveraging the user's full context across applications and activities.

The Evolution of AI Auto-Complete

• Started with code auto-completion that transformed software engineering
• Initially impressed users by suggesting code based on the whole project and corpus of existing code
• Improvements came through richer context - the more context provided, the better the suggestions
• Adding clipboard content to prompts was a breakthrough since it often contains what's in the user's mind
• Auto-completion gets better with more context, smarter models, and search capabilities

The Problem with Current Systems

• Current spell-checking and auto-complete systems lack awareness of recent context
• Example: When someone mentions "B zero" in Slack, then you type it in email, the system suggests "via" instead
• Context gets lost when moving between apps, breaking the sequential nature of human work
• Current systems don't exploit the full potential of next token prediction
• Every app switch results in lost context, even though users typically work sequentially

The Vision for Context-Rich Auto-Complete

• Create systems that understand what you're talking about, thinking about, and working on
• Connect the dots between different apps and activities
• Maintain context when switching between applications
• Leverage the sequence of everything you're doing and thinking about
• Recognize that work is sequential - reading an email about a problem likely leads to discussing that problem in another app

Implementation Approaches

• Enhance the operating system level auto-complete (non-trivial but powerful)
• Ingest the series of apps, integrations, and systems people use
• Put the right contextual information into AI prompts without changing the intelligence engine
• Create a system similar to "The Entire History of You" from Black Mirror - an external memory system
• Focus on connecting the dots between different applications and activities

Benefits

• Just-in-time expertise delivered inline while writing
• Suggestions that reflect your actual intent rather than generic corrections
• Reduced friction when moving between different applications
• Cognitive enhancement through better contextual understanding
• More natural and personalized assistance