Why Was OverDue Created

OverDue was created because a single overloaded Tuesday made the problem unignorable. Nine hours on calls: five sales, two team check-ins, a vendor sync, a partner intro. Each one ended with action items. By 7 PM there were 23 things to do across the meetings just held, and zero of them were on the calendar. They were sitting in Fathom, in Fireflies, in two different Notion pages, and in a notes app where three things had been quickly typed during a call so they would not be forgotten.

None of those 23 things got done that night. Most of them did not get done the next day either. They lived in five different places, and the cost of looking through all five places was higher than the cost of just forgetting them, so they were forgotten. By Friday, four had become urgent (because someone else followed up), eleven had quietly expired, and the rest were still where they had been left.

This is not a personal discipline problem. This is a tooling problem. And most of the tools had already been tried.

What was tried first

Motion

Motion is genuinely good at what it does. It auto-schedules tasks given to it, rebuilds the week when deadlines change, and the deadline-reshuffle feature is unique. Three months of Motion revealed the same gap: Motion expects typed tasks. Knowledge workers who run on calls generate tasks by talking, not by typing. By the time every action item from every call had been transcribed into Motion, the deep work Motion was supposed to schedule had already been eaten.

Reclaim

Reclaim solved a different problem. It defends a calendar from meetings, which is great if the problem is too many meetings. For workers whose meetings are the work, the meetings need to stay. The problem was the tail of action items every meeting generated, with nothing capturing them. Reclaim was not built for that. It was built to keep afternoons clear, not to figure out what was supposed to happen in those afternoons.

Notion + Calendar

A Notion database connected to a calendar via a hand-built workflow lasted about two weeks. It broke in the third week when three tasks went unentered during a busy stretch and the system was only as good as the manual input. Every tool that depends on a human being a reliable data-entry layer eventually breaks, because nobody is a reliable data-entry layer when they are tired.

Apple Reminders, Todoist, Things

List apps. They are great at lists. They are bad at scheduling. They are worst at scheduling around real meetings. A task typed into Things gets forgotten for a week, and a glance at the list afterward gives no clue which of the 60 items should happen today. Lists do not tell you when to do things; they just tell you that you have not done them.

The first insight: tasks come from calls

Once the pattern was named, it was hard to unsee. For knowledge workers who run on conversations, roughly 80% of the work on any given day was decided in a recent meeting. The recordings were right there, in Fathom and Fireflies. The action items were already extracted. The only missing piece was a layer that took those action items and turned them into placements on a calendar.

That gap, between transcript and calendar, was the entire product. Not a better list. Not a smarter scheduler in isolation. The bridge. The thing that did the boring work of moving an action item from "captured by AI in a transcript" to "scheduled in a specific slot on Wednesday afternoon."

The second insight: not all minutes are equal

Designing the scheduler surfaced another quiet mistake in how most calendar apps treat time. A calendar is usually treated as a uniform substrate where any 30 minutes is as good as any other 30 minutes. It is not. A peak window (mornings for most people) is worth two of an afternoon for the kind of deep work that pays the bills.

So the scheduler had to be aware of that. Each time window carries information about what kind of work it is good for: deep, shallow, admin, or mixed. Each task carries the same information. The match between the two became one of the scoring axes. The system stopped doing things like booking a strategic-review task at 4 PM and writing a quarterly plan at the same time most brains are wanting a snack.

The third insight: money matters

Looking at the action items piling up across a typical week of calls, a lot of them are not equal in value either. "Send the contract redline to Sarah" is a direct-revenue task. "Update the team wiki" is important, but not in the same way. Most schedulers do not know the difference. They will happily put the wiki update at 9 AM and the contract redline at 4 PM, because priority is the only thing they look at, and both got marked P2.

So the scheduler had to be aware of dollar impact too. Direct, indirect, none. When two tasks are otherwise tied, the one with direct revenue impact wins. This is obvious if you say it out loud. It is missing from every other productivity tool on the market.

The fourth insight: boundaries are personal

Every scheduler tried before made implicit assumptions about when work happens. Motion defaulted to 9 to 5. Reclaim assumed evenings off. Sunsama assumed planning happens the night before. The assumption was almost never wrong, but the moment it was wrong (because of travel, a late-night project, an unusual shift, or a parent working pockets after kids sleep), the tool stopped fitting.

So OverDue does not assume anything about when work happens. The user defines the windows. The scheduler places work inside them and refuses to place work outside them, full stop. If a deep window at 10 PM fits, OverDue is fine with that. If admin only on Sunday mornings fits, OverDue is fine with that too. The boundary is the user's.

Where the product is now

The auto-scheduler runs every 15 minutes autonomously. Action items flow in from Fathom, Fireflies, Grain, ClickUp, Granola, MeetGeek, Plaud, and others. The triage queue handles the human decision (accept, edit, skip) and everything else is automatic. Every placement carries a confidence score, and anything below the threshold gets surfaced for review. The five-axis scoring (energy, due-date, slot-fit, buffer, no-conflict) is honest math, not a vibe. The numbers are persisted so the UI can show exactly why a slot was picked.

Beyond that, there is a developer API for the people who want to plug their own agents in, a public webhook surface for outbound events, and an ergonomic mobile PWA so the same product is in your pocket. There is dark mode for those who work in the dark. There is a small AI agent called Odee on the way that will surface suggestions for tuning windows and right-sizing priorities based on what actually gets completed.

Why now

AI meeting recording crossed a threshold in the last 18 months. Fathom, Fireflies, Grain, and the rest are now genuinely good at capturing what was said and what needs doing. Millions of knowledge workers use one of them every week. The transcripts are sitting there. The action items are extracted. The bridge to a real schedule did not exist, and the marginal cost of building it (now that the upstream AI has been figured out by someone else) was small enough that a small team could do it.

OverDue is Due soon. If a scheduler that starts where tasks actually come from sounds like what was always missing, this is it. Founder pricing is locked at $15 a month for life for the first 100 customers. After that, the regular Pro tier is $29 a month and the Personal tier is $19 a month, both with a meaningful annual discount.