AI-assisted task planner

Problem and motivation

Most to-do apps make you decide what's urgent yourself: you set priority flags, guess, and re-plan by hand whenever a deadline moves or a day fills up. Those guesses go out of date within a day, and they only look at the deadline, not at how much free time you actually have before it.

Approach

Each app keeps doing the job it's already good at. Todoist holds my assignments, and all I enter is a due date and a rough guess of how long each will take; I never set priorities myself. Google Calendar holds things that can't move, like classes, practice, exams, and meals. Claude, an AI assistant, reads both every time it plans and builds the day's schedule following rules I wrote for it:

• Splitting big tasks: if a task doesn't fit today, it schedules the part that fits and plans the rest later.
• Breaking ties: when several tasks aren't urgent yet, shorter ones go first.
• Not neglecting routine work: optional recurring tasks, like project time, move up each day they get skipped, so after about three skips they beat everything except truly urgent deadlines.
• Learning my time estimates: it compares how long tasks were expected to take with how long they actually took, and adjusts my future guesses. Only the planner makes this adjustment and my original guesses stay unchanged, so the correction never gets applied twice.

Outcome and results

The result is a personal planner that turns my to-do list and calendar into a realistic daily schedule. Adding a task is designed to take one line, just a due date and a time guess, and each day's plan shows the adjusted time estimates so I can double-check them. I'm now expanding it into a fuller scheduler.

So far, every missed assignment has come from what I entered, not from how it was scheduled: tasks I forgot to add, and the one described below.

Human error still counts

I missed one class assignment because I assumed it was due at midnight when it was actually due at 8 a.m. The planner can only work with the deadline it's given; it can't catch a due time entered wrong or a task that never got entered. A tool like this can handle the planning, but human error still matters, whatever system you use.

Skills practiced

• Connecting separate apps so they share information
• Designing how an AI assistant makes decisions, and the scheduling rules it follows
• Saving information between sessions so the planner remembers past results
• Designing a self-correcting system that doesn't over-correct

Challenges and what I'd do next

Some tools I needed weren't available. I couldn't edit individual spreadsheet cells, only whole files, so the planner's learned adjustments and skip counts are stored in a small web page instead. Claude can't read that page directly, so a “sync to chat” button passes the information along each session. I also tried existing AI productivity add-ons and dropped them: none compared deadlines against actual free time, checked my calendar, or learned from my time estimates.

Next, the fuller scheduler will estimate how long tasks take on its own, with no guesses from me, which needs more data than I have right now. It will also pull assignments straight from a course calendar or syllabus, which should cut down on entry mistakes like the 8 a.m. deadline above.

Tools

• Claude (AI assistant)
• Todoist
• Google Calendar
• A small web page that saves data
• Spreadsheet