Senses · 03: The Interface

The Cortex has no single front door, and forcing one is a mistake. The right surface depends on the job: the terminal for building, the desktop or phone app when you’re away, voice notes for raw capture, a dashboard for the bird’s-eye view. The skill isn’t installing the surfaces. It’s knowing which one to reach for, so the right tool never has friction.

Why “the interface” is plural

Most apps have one screen. An operating system has many: a place to type commands, a file browser, a settings panel, a phone widget. Different jobs want different surfaces. The Cortex is the same. The mistake is picking one and forcing everything through it: doing a strategy review in a text window, or trying to build an automation on your phone. Each is painful, so the habit dies.

The fix is to match the surface to the job. The same Cortex underneath, the same Memory and the same connectors, is reachable through a few surfaces, each tuned to a kind of work:

Deep work and building want the full tool: Claude Code in the terminal (the text window where you type commands) or a code editor.
Away from your desk wants the desktop or phone app: lighter, still connected.
Raw capture wants the lowest-friction input there is: your voice, recorded and routed.
The bird’s-eye view wants a read-only dashboard, not a chat at all.

Set them all up, then let the rubric below decide which you’re in.

The surfaces

Surface	Best for	Why this one
Claude Code (terminal or editor)	Building, editing your files, running skills, multi-step work	Full power. Reads files, runs commands, does the work. Your main surface.
Desktop or phone app	Quick questions, reading, light edits away from your desk	Same Memory, less friction, no terminal needed
Voice and meeting capture	Raw input: calls, thoughts, brainstorms	Lowest-friction capture. You talk; `/meeting-cortex` files it into Memory
A dashboard (web)	The bird’s-eye view: decisions, pending changes, your backlog	Read-only. Shows your files as boards, not a chat. The review surface

A word on the dashboard: it’s not another place to chat with the Cortex, it’s a place to see it. It reads your files straight from where they live and shows the decision board, the pending changes, and the backlog. When the job is “show me the state of the business,” a board beats a conversation.

The “where to do what” rubric

The rubric is the real deliverable of this chapter. Learn it and surface friction disappears:

Building, editing, running a skill?      → Claude Code (terminal or editor)
Quick read or question, away from desk?  → desktop or phone app
Capturing a call, a meeting, an idea?    → voice, recorded → /meeting-cortex
Reviewing decisions or pending changes?  → your dashboard (web)
Approving a change?                      → dashboard to see it, Claude Code to apply it

The through-line: capture where it’s easiest, build where it’s most powerful, review where it’s clearest. Don’t capture in the terminal, don’t review in a chat, don’t build on your phone. Each surface has a job. The rubric keeps you in the right one.

Pitfalls (the “if this, then that” guide)

Situation	What to do
You force everything through the terminal	Stop. Capture wants voice, review wants the dashboard. The terminal is for building. Forcing one surface is how the habit dies.
You try to do real building on your phone	Don’t. The phone is for reading and quick questions. Deep work needs the full tool. Park the build until you’re back at your desk.
You type meeting notes by hand	That’s friction you don’t need. Record it and let `/meeting-cortex` file it. Capture should be effortless or it won’t happen.
You review by scrolling raw files	Use the dashboard. Seeing your files as boards is the whole point of a read surface. Raw files are for editing, not surveying.
You stand up the dashboard before your files are tidy	Backwards. The dashboard shows well-formed files. Do the Memory chapters first; a read surface needs something structured to read.
You add a fifth and sixth surface for novelty	Each one is upkeep. Four cover the modes of work. Add another only if a real kind of work isn’t served.

Maya’s version

Maya lands in each surface without thinking about it now. She builds and edits in Claude Code at her desk. When she’s out, she uses the phone app to ask a quick question or read something back. Calls and voice memos get recorded and routed in by /meeting-cortex. Once a week she opens her dashboard to see open decisions and pending changes as boards, instead of scrolling files.

The switch that stuck: she used to type up notes after every supplier call. Now she just talks, the recording files itself, and the note writes itself. Her highest-volume input got the easiest door.

Two things worth keeping:

The dashboard reads the same files the terminal edits. It’s one source of truth shown two ways: edit it in Claude Code, survey it on the dashboard. Not two apps, one system with a shell and a dashboard.
Capture is deliberately the easiest surface. The most valuable input, what gets said on calls, comes in through the surface that takes the least effort: talking.

You’ve got it when

Each kind of work (build, read, capture, review) has a surface you reach for without thinking
Claude Code is set up where you do deep work
The app is on your phone or laptop for when you’re away
Capture happens by voice or recording, routed in by /meeting-cortex, not by hand-typed notes
A read-only dashboard shows your files as boards for the bird’s-eye view
You’ve caught yourself about to force the wrong surface and switched

If build, read, capture, and review each have a surface that fits, the sense is live. The last one means the rubric is now a habit.

The skills that help here

This chapter installs surfaces, not commands, with one exception: /meeting-cortex routes recorded meetings into Memory. It’s in the starter kit, with the full list in ../03-habits/01-skills-and-commands.md.

→ Next: ../03-habits/01-skills-and-commands.md. Leave Senses for Habits: the skills, commands, and workflows that run on top of the surfaces and connectors you just installed.