Task: Develop a unified task graph for managing a large task pool

Problem

AI dramatically reduces the cost of task assignment and decomposition. In a short time, dozens of reasonable tasks can be created, but new decompositions arise locally and often without considering the existing pool. As a result, not only does the backlog grow, but so does the cost of understanding the big picture.

Limiting the flow of ideas and experiments is undesirable: this destroys one of the main advantages of AI. We need a way to retain a large volume of potential work without forcing a human to keep it all in their head.

Research Direction

Treat tasks not as an independent list, but as a unified task graph. AI should be able to analyze the entire pool, find semantic and technical connections, combine local decompositions, and build representations at various scales.

Acceptance criterion: a human at the top level is able to see and understand the entire current picture. Details are revealed only when necessary.

When choosing the nearest execution point, the system should be able to show the local subgraph: which other tasks are connected to this point, which can be done concurrently, which are partially or fully closed by the same change, and which logically follow it.

What Needs to Be Addressed

• What types of relationships between tasks are truly useful: dependencies, shared system area, shared research or context, partial result overlap, joint implementation, succession, alternatives, etc.
• How to automatically discover connections between existing and new tasks.
• How to fold a large graph down to a human-comprehensible level without losing underlying details.
• How to determine the nearest sensible execution points without turning the system into a rigid automatic prioritizer.
• How to review the graph after system changes: obsolete tasks, effectively closed tasks, changed connections, and new opportunities for joint execution.
• How to use existing projects, tasks, WorkLogs, and haih-agent knowledge as the actual context of the graph.

Principle

Do not limit the number of ideas for the sake of manageability. Make the task space itself comprehensible and manageable.

AI is needed primarily for continuous analysis, linking, and folding of the task graph; the ultimate understanding of the picture and the choice of where to apply effort must remain accessible to the human.