Designer, mid‑compile

Portfolio • Case Study 01

MARLOWE

Role
Owner: design, architecture and build
Timeline
2025 — present
Status
Version 2.0.4 on my Mac; live acceptance pending

What is it?

A voice-first macOS companion for thinking out loud, researching ideas and turning clear requests into saved documents, account actions and delegated design or coding work.

01

Problem framing

I think faster out loud than I type, and typing flattens a requirement. By the time an idea is a paragraph in a prompt box, the reasons behind it are gone, and a coding agent gets a thinner brief than the one in my head. I wanted to close that gap: talk a problem through and come out with a task an agent can act on, rationale intact.

The harder half was trust. An assistant that can speak and also act on your machine is a liability unless the line between talking and acting is explicit. So the brief had two parts: keep conversation fluid, and make every action specific, permissioned and accountable.

MARLOWE stands for Multi‑Modal, Adaptive, Reasoning, Learning, Operations, Workflow, Engine. It began as SideKick; the name came once it was clear what it was for.

The problem: typing a requirement flattens the reasons behind it, and a spoken assistant that can act needs a clear line between talking and acting. 1 · Typing flattens the thinking Idea in your headreasons, trade-offs, context Typed into a promptthe why gets flattened Agent gets a thin briefbuilds the wrong thing reasons lost reasons lost 2 · Talking and acting blur together You think out loudhalf a sentence, a maybe Model hears a commandno line between talk and act It runs on your machinenothing to approve or undo The goal: keep conversation fluid, and make every action specific, permissioned and accountable.
Two failures, one goal. Dashed: what gets lost between thinking and the brief. The solution answers both rows.
02

Research through use

There was no panel of users. There was me, using it every day, and a saved record of every conversation. MARLOWE began as an inherited prototype whose notes described designs that no longer existed, so the first rule was to trust what ran over what was written. After that, every transcript, with its timing and tool outcomes, was evidence, and each finding changed the design:

  • Pauses. Measuring time to first audio per turn showed where the waits landed, and filler lines like “give me a moment” read as stalling. Routine work now happens quietly, with a short, truthful update only when the wait needs one.
  • Noise. A speech classifier meant to reject background sound failed on short replies, so a simpler loudness gate stayed, with its limits written down.
  • Proof. A transcript shows what was said, not what happened, so outcomes are recorded on their own, and a task that was submitted, returned and verified are three different states.
  • Cost. Tracking spend per conversation showed what a daily habit of talking to it would actually cost.
  • Writes. In a live session I asked for a new research document, and MARLOWE reached for an older one instead. The write was refused and nothing changed, but the lesson was structural: creating a document and revising one are now separate tools, and a revision has to name the version it read.
03

Solution as a system

The findings came down to one structure: the voice talks, the app acts, and nothing claims more than it knows. A live speech model handles listening, speaking and interruption. When a request needs a tool or deeper thinking, it goes to a separate reasoning model, and the voice explains the result that comes back.

Every proposed action is checked against the current request, its target, the permissions granted and the state of the task. A clear request can authorize scoped work without a ritual confirmation, but sending mail needs an exact preview and a separate yes, calendar changes need review, and reading a folder never implies writing to it.

Through that gate, MARLOWE researches with cited sources, drafts and revises documents, works in mail and calendar, calls connected MCP tools, and hands Codex, Claude Code or Pen one brief carrying the original request and its exact workspace: the gap from the start, closed. Everything lands in a vault you can read, conversations in one folder and tasks in the other. Stopping work does not undo it, and the interface says so.

MARLOWE workflow: conversation, delegated reasoning, app-checked actions and recorded outcomes. GPT-Live handles ordinary conversation and delegates requests when needed. The app validates current intent, target, permissions and task state before tools execute. Tool receipts inform replies and persistent task records. A local supervisor monitors background work. You speak, interrupt, request review when required GPT-Live voice ordinary conversation delegates when needed Reasoning backend interprets requests proposes tool actions speech spoken reply request + context results + receipts App-owned execution checks Current request · exact target · permissions · task state · duplicate and uncertain outcomes Required previews and approvals are enforced here. A preview alone is not consent. Recall saved conversations and source context Research web sources and saved reports Files vault documents granted workspaces Mail & Calendar read and prepare review before writes Coding agents Codex / Claude Code scoped task dispatch Design direct Pen route Figma via Codex Recorded outcomes and follow-up Saved paths, sources and task state inform the next reply and continuation. Submitted ≠ completed. Returned ≠ verified. Uncertain actions need reconciliation before retry. Marlowe Vault transcripts / tasks Conversations, outputs and readable task records Hidden ledgers retain operation and recovery state. Local supervisor Background research and task monitoring can outlive the window while awake. Window close ends voice; folder grants stay. Capabilities depend on configuration and access; records are evidence, not permission to act.
Conversation delegates work as needed. The app validates actions, tools return recorded outcomes, and the voice reports those results. The capability groups summarize multiple tools; available routes depend on settings and permissions.
04

Interface

Inside the app the brand steps back. The conversation screen is the Nebula, the microphone and a settings control, nothing else. The Nebula is 26,000 points under a reflective skin that breathes; speech moves its ridges and bloom, not its size, and it has no eyelids or expressions. Light mode is its own rendering, not an inversion.

Settings is where trust becomes visible. Each connection states what it can do and how far it is trusted: Claude Code is enabled but unchecked until dispatch, Gmail sends with review, Calendar changes with review, and a transport that cannot be verified says so. As the design boards put it, an animation is not evidence that a task succeeded.

Waiting is shown honestly too. While a reply is pending, a thin sweep traces the Nebula’s surface, and with reduced motion it holds still. Research sources gather in a Web Sources tray that can be cleared without ending the conversation or deleting what was saved.

MARLOWE speaking in dark mode: a glowing violet particle sphere centered on deep indigo, a caption below it partly highlighted as it is spoken, and the microphone button at the bottom.
MARLOWE ready in light mode: a soft lavender particle sphere on a near-white background with the microphone button at the bottom.
The Nebula speaking in dark mode, its caption highlighting as the line plays, and ready in light mode. Both are renderer captures with synthetic input.
Marlowe Settings, General tab in dark mode: a choice of two voices, dark or light appearance, and microphone and speaker selects set to follow the system default.
Marlowe Settings, Connections tab: Claude Code and Codex enabled, Gmail and Google Calendar connected with review noted for sending and changes, and OpenAI Dot marked transport unavailable.
Marlowe Settings, Tools and MCP tab: Figma via Codex and pen.dev enabled as built-in integrations, and an empty custom MCP servers list with an Add server button.
Settings: General, Connections, and Tools & MCP. Each connection says what it may do and how far it is trusted, down to a transport it cannot verify.
05

Design system

The design system exists for the same reason the guardrails do: so the interface never claims more than the software knows. The brand palette stays out of the interface, so the personality lives in the mark and the Nebula and every control reads plainly. Eight semantic roles, from canvas to danger, carry meaning in both themes, and color never stands in for a label, so a status is always written in words. Every button has six specified states, busy included, because a button that can be pressed twice is a message that sends twice.

It is a contract, not a style guide. Eight numbered boards in Marlowe.pen describe it, the code owns it, and a test renders 108 views across both themes and two widths to check the two still agree. When the design changes, the source changes first and the boards are regenerated from it, so the picture never drifts from the product.

Marlowe interface system board in dark mode: primary, secondary and destructive buttons in default, hover, focus, pressed, disabled and busy states, text, select and error fields, switches, and a spacing and behavior summary.
Board 02 of the design contract: three button types in six states each, field patterns, switches and the spacing rhythm, all on the same semantic tokens.
06

Brand

The mark is a flowing M: one continuous ribbon, cool on the way in and warm on the way out, running from sky blue and violet through pink to amber and warm cream on deep indigo. A voice product gives you little to look at, so the icon carries the warmth the interface keeps quiet.

An assistant that listens to you and acts on your machine has to earn trust before it does anything, so I wanted the mark to feel warm and calm rather than clever or robotic. The ribbon never breaks, the way a good conversation doesn’t, and its color turns from cool to warm the way a rough thought becomes something you can hand off. And it had to be unmistakable: an M at 16 pixels, on light or dark, because the tools you trust are the ones you recognize at a glance.

The final MARLOWE app icon board: a flowing ribbon M in a blue-to-amber gradient on deep indigo, shown at 1024 pixels, scaled down to 16, on dark, light, indigo and warm backgrounds, with its five-color palette and gradient stops.
The final icon: the flowing M at 1024 pixels and scaled to 16, on four grounds, with the palette and gradient it is built from.
07

Takeaway

Saying a requirement out loud forces you to finish the thought. The most useful thing MARLOWE does is make me answer the question before the agent has to.

The guardrails cost more than the features. Scoped permissions, required reviews, honest result states and a stop control that admits it cannot undo prior edits all took work to design. The goal is to keep a clear request moving while preserving the checks that make its effects accountable. Version 2.0.4 passes 460 automated tests and all 23 regression stages, but those tests use stand-ins for the models, so they prove routing, not conversation. The last live session, on 2.0.3, finished its research and kept its sources, and it surfaced four defects: a folder grant that went stale, a new document routed to an old one, stale job records blocking new work, and failures explained too vaguely. Version 2.0.4 repairs all four. Talking to it is still the acceptance gate, and for 2.0.4 it is still open.

Nothing touches the calendar until I say so. MARLOWE drafts the event from what I said, shows exactly what will be saved, and waits for a spoken yes before it acts.