How Flowo Lowers the Cost of Starting a Task

What we mean by initiation friction

Initiation friction, as we use the term, is everything an app asks for before a person can begin. We cannot observe what happens inside someone at the moment of starting. We can count the fields and screens an app puts in the way, and those are in our control.

This article describes design decisions and what the code does. Flowo is a planning tool, and nothing here is a claim about results for any person.

Capture asks for text and nothing else

A Braindump note needs one thing to exist, which is non-empty text. There is no title field, no group to pick and no date to set. The editor saves 500 milliseconds after the last keystroke and again when the screen closes, so there is no save step to forget.

Sorting happens after capture, on the phone. An analyzer reads the saved text and fills in metadata the person did not have to enter.

• Links in the text are extracted and deduplicated.
• Text that starts with checkbox syntax or matches a list of action verbs is flagged as a to-do.
• A date or time phrase becomes a reminder only when the parser's confidence is 0.75 or higher. Below that, the note is saved without one.

Text shared from another app is saved on arrival. We do it then because a share is an intent to keep something, and the note should exist even if the person dismisses the screen that follows. A share that arrives before storage is ready is held and written as soon as storage opens.

Three actions are reachable from outside the app

Flowo exposes three actions to the operating system: capture a thought, create a task and start focus. On iOS they are Home Screen quick actions and Siri App Intents. On Android they arrive from launcher shortcuts, a quick access widget and Google Assistant.

Each action opens its own destination as a full-screen page over the app. Capture lands in the note editor, and start focus lands on the focus screen.

Only the share sheet may carry content. A deep link or an assistant phrase carries an intent and no text, so an external entry point cannot write into someone's notes.

Two more Siri intents work on the current schedule. One answers what is next with a start time and a length in minutes, without opening the app and without speaking a task title. The other marks the current block done and hands the change to the same code path a tap in the app uses.

A task needs a title, and the form fills in the rest

The task form requires a title, and that is the only thing the person must supply. A default duration is filled in, and a task saved without a category is stored as uncategorized. Duration and priority are the two values the form goes on to propose.

When the title is longer than three characters and typing has paused for 1.5 seconds, the form asks the cloud model for an estimate in minutes and a priority from 1 to 10. Each new keystroke resets the timer, and a reply to an older request is discarded. The proposed values stay editable, and a switch turns the automatic estimate off. If the request fails, the form says the estimate is unavailable and the person sets the values by hand.

A breakdown is a proposal until the person accepts it

Breaking a task into steps writes nothing until the person confirms the draft. The sheet first asks how many steps to make. The default follows the estimate: two steps for up to 30 minutes, three for up to an hour, four for up to two hours, and seven at most by default. A slider allows from two steps up to a ceiling of one step per five minutes of estimate, capped at twelve.

The request holds the task title, its estimate, its notes, the step count and the app language. The backend forces the model to answer through a structured tool call and then corrects the answer in code. If the model returns too many steps, the two neighbours with the smallest combined time are merged. If it returns too few, the longest step is split in half. Times are rescaled so that they add up to the parent's estimate, and the last step absorbs the rounding.

The person can edit titles and times in the draft. Confirming it runs one write path that does four things.

1. 1.Deletes any earlier subtasks of the task.
2. 2.Releases the parent's future schedule blocks. Confirmed past work stays on the record.
3. 3.Creates the new subtasks, which inherit the parent's priority, deadline, category and reminder settings.
4. 4.Asks the planner to plan the day again.

Step two exists because a task with children stops being schedulable work. Its old blocks would otherwise keep occupying the day while no screen showed them.

Only a step without children can be started

In Flowo a task can be started only when it has no subtasks. A parent is a container, and the question of where to begin always resolves to one concrete step. When the last open subtask is completed, the parent is completed too, and the same check runs on the parent's own parent.

Nesting has a limit on the free plan. A task can hold steps and those steps can hold sub-steps, and deeper levels belong to Flowo Pro. The depth check counts ancestors only as far as the limit, which also ends the walk if stored data ever contained a parent cycle. The focus screen that a started step leads to is described in our article on Flowo's focus screen.