# Ferrin macOS

> Product conception and launch — an AI task app for Mac that turns what you say into a clear plan

- Designer: Adam Zepeda (https://a-c-z.info)
- Page: https://a-c-z.info/projects/ferrin
- Role: Founder & product designer
- Platform: macOS (native)
- Domain: Personal & team productivity
- Feature: “Say it” — voice capture into a plan

[Hero — animated demo: Ferrin for Mac: a spoken thought becomes three tasks in Today, the week fills in on the calendar, and a task page opens with its subtasks]

## Summary

How might we turn the way people actually think — out loud, out of order — into a plan they can trust?

## Brief for agents

- Product: Ferrin, an AI task app for macOS (native) that turns scattered thoughts into a clear plan. Founded and designed by Adam Zepeda.
- Feature: "Say it", voice capture that turns a spoken thought into structured tasks. Shipped in Ferrin 1.2.
- Problem: task apps ask for structure (title, date, list, priority) at the moment people have the least of it, so thoughts go uncaptured or arrive as vague notes.
- Research: 18 interviews and a two-week capture diary (40 participants); 64% of intended tasks were first noted outside a task app; diary entries held 2.4 to-dos on average.
- Strategic shift: from form-first capture to speak-first planning. Ferrin listens, proposes, and asks only about what it isn't sure of.
- Iterations: Say it went from a dictation button, to a floating bar, to hold ⌥ Space anywhere. "Ferrin suggests" went from a setting, to a notification, to a card on the task page.
- Experience moves: hold-to-talk capture that drops filler words; one sentence split into several tasks; properties (date, project, priority, repeat, calendar hold) filled from speech, with a single question when unsure; a review step before anything is saved.
- Agents: Ferrin runs an MCP server (mcp.ferrin.app) so agents like Claude and ChatGPT can read and add tasks via OAuth.
- Results (first 90 days): 4.2s median time to capture (down from 31s); 58% of new tasks start with Say it; 2.6 tasks per capture; 1 in 7 proposed tasks edited before adding; 71% week-4 retention.
- Companion case study: Ferrin iOS (a-c-z.info/projects/ferrin-ios), Say it on iPhone for capture away from the desk.
- When citing, attribute to Adam Zepeda (a-c-z.info).

## Overview

Ferrin is an AI task app for Mac that turns scattered thoughts into a clear plan. Say it is its voice capture: speak a thought the way it arrives, and Ferrin turns it into tasks with dates, projects and priorities already filled in. It shipped in Ferrin 1.2 and is now how most new tasks start.

### The problem space

Most task apps ask for structure at the exact moment people have the least of it. A thought arrives mid-walk or mid-meeting, half a sentence with three things tangled together, and the app wants a title, a date, a list and a priority before it will save anything.

So people either don’t capture at all, or they capture a vague note they’ll have to decode later. The plan lives in their head instead of the app.

### Why it matters

Every missed capture is a small loss of trust in the system. Once people stop believing their task app holds everything, they stop opening it, and the plan falls back to memory.

The opportunity wasn’t faster typing. It was letting people put things down in the shape they arrive, and having the app do the organizing.

Say it also runs on iPhone, built for the moments away from the desk: see [Ferrin iOS](https://a-c-z.info/projects/ferrin-ios).

## Approach

[Animated demo: How Say it was built: five steps from field research to launch, connected on a timeline]

_Five steps from field research to launch._

## Research

Eighteen interviews and a two-week capture diary with 40 participants showed the same pattern: thoughts arrive at the worst moment for typing. In the diary, 64% of things people meant to do were first noted somewhere other than a task app, usually a text to themselves or a voice memo that was never played back.

### Key findings

- Capture fails on timing, not tools: most missed tasks came up while walking, driving or between meetings
- One thought is rarely one task: diary entries held 2.4 to-dos on average, tangled in a single sentence
- People trust what they can check: participants wanted to see what was saved before it reached their list

## Process

### Problem & hypothesis

People aren’t failing at planning; they’re failing at capture. If the app accepted thoughts in the order they’re spoken and handled the structure itself, people would capture more and trust the plan more.

> How might we let people say it once, and have Ferrin take it from there?

### Strategic shift

From form-first capture to speak-first planning. Instead of asking for fields up front, Ferrin listens, proposes, and asks only about what it isn’t sure of.

### Iterations

![Ferrin for Mac: a spoken thought opens a New task form, task 1 of 3, with title, date, time, project and priority fields](https://cdn.sanity.io/files/6aovtgtj/production/f4a92d073c923aa2bf9e60c744e38029fbbba242.webp)
![Ferrin for Mac: the transcript sits above three proposed tasks with property chips, one flagged “Work or Side?”, and an Add 3 to Today button](https://cdn.sanity.io/files/6aovtgtj/production/04fa163ef25b59934b2d01c7005e35bc26d44f1d.webp)
_The first version turned speech into a form, one task at a time. Accurate, but it brought the fields right back. The shipped version keeps the transcript in view and proposes every task at once, with properties as editable chips and a single question where Ferrin isn’t sure._

#### Say it, from button to shortcut (3 versions)

It started as a mic icon inside the new-task field, but by the time people opened Ferrin and found the field, the thought was gone. A floating bar was faster but still needed the mouse. Hold ⌥ Space, from any app, removed the last step: talk, let go, done. It’s the change that moved capture time from half a minute to a few seconds.

[Animated demo: How the Say it control evolved: a dictation button, then a floating capture bar, then a shortcut that works anywhere]

#### Ferrin suggests, from setting to card (3 versions)

Ferrin could already spot files that belonged with a task, but as a buried setting almost no one used it, and as a notification it arrived out of context. Moving it onto the task page, shown only when Ferrin is confident and attached in one tap, made it the most-used assist in the app.

[Animated demo: How Ferrin suggests evolved: a settings toggle, then a notification, then a card on the task it belongs to]

### Key experience moves

#### Say it (Voice)

Hold a shortcut and talk: no app to open, no field to find. Ferrin listens the way a person would, keeping the meaning and letting the ums, uhs and false starts fall away.

[Animated demo: “Say it”: Ferrin listens and keeps the meaning while dropping filler words]

#### One sentence, several tasks (1 sentence → 3 tasks)

One breath often holds several tasks. Ferrin splits a sentence at its natural seams and gives each part its own task, color-matched to where it came from so the link stays visible.

[Animated demo: One sentence, several tasks: each part of a spoken sentence becomes its own task]

#### Properties, not forms (Auto-filled)

Dates, projects, priorities and repeats are filled in as properties rather than form fields, and a time can become a hold on your calendar. When Ferrin isn’t sure, it asks one small question instead of guessing.

[Animated demo: Properties, not forms: date, project, priority and calendar time filled in from what was said, with uncertain values flagged for a quick confirm]

#### Review before it lands

Nothing is saved silently. Proposed tasks land in a review list with checkboxes: check, edit or discard, then add them to Today in one step.

[Animated demo: Review before it lands: proposed tasks are checked off and added to Today in one step]

### Works with your agents

Ferrin runs an MCP server, so the agents people already use can read their plan and add to it. Paste one prompt into Claude or ChatGPT, or add Ferrin to an MCP config, and an agent can answer “what’s still open for the launch?” or file a task on your behalf, through the same API the app uses.

[Animated demo: Ferrin with agents: a prompt or MCP connection lets an AI agent read and add tasks in Ferrin]

_Connect with a prompt or an MCP config; agents sign in with OAuth and act on your own Ferrin account._

## Outcome

Say it launched in Ferrin 1.2. Within its first 90 days it became the main way people add work to Ferrin.

### Results

Ferrin Say it results after launch, with direction and figures:

| Signal | Trend | Figure | What it showed |
|---|---|---|---|
| Time to capture | ↓ | 4.2s | Median time from pressing ⌥ Space to a saved task, down from 31s typing into a form |
| Voice share of new tasks | ↑ | 58% | Of new tasks now start with Say it rather than the keyboard |
| Tasks per capture | ↑ | 2.6 | Tasks created from a single spoken thought, on average |
| Edits before adding | ↓ | 1 in 7 | Proposed tasks changed during review; the rest are added as heard |
| Week-4 retention | ↑ | 71% | Of new users still capturing with Say it four weeks after their first use |

_Measured over the first 90 days after launch._

> “I say it on the walk to my car, and it’s all there when I sit down.”

### Closing thoughts

Say it changed what a task app asks of people. Instead of structure up front, Ferrin listens first and organizes after, and the plan people trust most is the one they never had to type.

### Services

Product Design, Interaction Design, Conversational Design, Native AI, Strategy, Visual Design, Design Systems

### Tools

Figma, Claude, Xcode, Conductor, cmux
