Ship a differentiated capability navigator paused

Desired capability: decide what to build next in Marrow
Success: recommendations users find more useful than asking ChatGPT for a curriculum
Why Marrow thinks this
User has exposure to AI product development patterns, multi-agent systems, learning science, and first-principles thinking. Key gap: no evidence of systematic comparison methodology between Marrow's context-aware recommendations and baseline chatbot outputs. Missing framework for what makes a recommendation 'more useful' beyond measurement—needs decision criteria for which unlocks to surface when.
All unlocks (3)
foundation

Decision criteria for unlock prioritization

When Marrow identifies 10 possible concepts at the frontier, which 3 should it show first—and how does that decision logic differ from what ChatGPT would do?
Unlocks: Design a ranking algorithm that makes Marrow's output demonstrably different from and more useful than baseline chatbot responses
~45 min · 1 note(s)
 bridge

Cognee-style memory architecture for user capability state

How should Marrow represent not just what notes a user saved, but their evolving capability relationships—like 'understood concept X last month, applied it in project Y, now questioning assumption Z'?
Unlocks: Build a persistent capability graph that evolves as users explore, apply, and revise concepts—enabling truly longitudinal recommendation improvement
~60 min · 2 note(s)
 challenge

The reliability paradox in AI-native product development

If Marrow's core value is personalized recommendations, but LLM output is non-deterministic, how do you balance 'productionizing for reliability' with 'exploring frontier variations that might be more useful'?
Unlocks: Design a deployment architecture that maintains reliability for users while safely experimenting with recommendation improvements
~50 min · 3 note(s) · inferred
Generation details
Based on 24 relevant note(s) · claude-sonnet-4-5 · capability_frontier_v1