task scheduler

a better version of Morgen (or any AI scheduling tool) that solves two problems the current crop misses: auto-planning and wait-time awareness. auto-planning means when you add a task, the scheduler doesn't just block time — it reasons about what the task actually requires, breaks it into steps, and slots them intelligently given your current calendar and energy patterns. wait-time awareness is the more unusual insight: many tasks have external dependencies that block you mid-execution (waiting for a build, waiting for a reply, waiting for an API to respond), and a good scheduler would interleave work around those gaps rather than treating each task as a monolithic block.

the scheduling problem is harder than it looks because it requires a model of the task itself, not just its time cost. a task that takes 2 hours is very different from one that takes 20 minutes active + 100 minutes waiting. the system would need to learn, over time, which of your tasks have wait-time signatures and start predicting them. the auto-planning component would decompose a high-level task like "prep for meeting with X" into component steps, estimate each, and slot them in reverse from the deadline. this is essentially a planning agent with calendar access.