Memoth

The idea came when I was pregnant. Between medical checks, ultrasounds, and a calendar that kept growing, I was forgetting things from one call to the next: what the doctor said, what I needed to ask, when the next appointment was. The information lived in fragments across my notes app, my calendar, and my memory. None of it talked to each other.

Once maternity leave started, I built the whole thing during it. Between feeds and naps, with Claude as the constant engineer.

AI voice notes existed, but the experience always felt like two products awkwardly joined together: a recorder and a transcript tool. I wanted something that felt like one thing: an app that captures every conversation that matters, across every place a conversation happens, and turns it into something I can actually use afterwards.

The shape of the product: automatic detection of meetings on Mac (Zoom, Teams, Google Meet, Webex, FaceTime), one-tap recording on iPhone, iPad, and Apple Watch, a WhatsApp bot for voice memos, and transcript-to-summary-to-action-items in a single flow. Privacy non-negotiable. Everything stays in iCloud, nothing on third-party servers.

The bet behind the bet: build it solo, ship it on the App Store, and see if a designer can move from concept to live product in months, not quarters, with AI as the engineering partner.

Just me. No PM, no engineer, no design partner, and unusually no Figma. I owned every decision: product strategy, app design, the memoth.com marketing page, and development. Claude was the engineer; the design lived in language, not pixels. I'd describe a screen (what it should do, what it should feel like, what to hold back) and pair with Claude Code to ship working Swift directly.

This is what "designer who ships" looks like now. The roles I used to hand off to other people (engineer, PM, sometimes even researcher) collapsed into one person plus the right AI tools. Memoth is my proof that the workflow holds up under the pressure of a real shipped product.

The loop was tight, and unusual for a designer: I never opened Figma. I'd describe a screen to Claude in plain English: what it should do, what it should feel like, what to hold back. Claude scaffolds the Swift, I run it on device, screenshot what didn't land, feed it back. The fastest design iteration cycle I've ever had. Partly because I never had to wait on anyone else, partly because Claude doesn't have ego about killing decisions.

The thing I had to learn: how to design with words first. Specs that work for human engineers (with their ambiguity, their context, the assumptions you can lean on) didn't always work for Claude. I had to describe interaction with more precision than I ever would in a Figma file: the easing, the moments of friction, the state at every step. The closer my description of the design, the closer the build matched it.

It changed how I think about design fidelity. The right resolution to design something at depends on how cheap it is to change once it's real, and with Claude, "real" is suddenly very cheap. The canvas became language, not pixels.

The thing I didn't expect: working with Claude didn't replace my taste, it sharpened it. When engineering is suddenly cheap and infinite, every design decision becomes a choice with real consequences, fast. There's no "we don't have time to build it" shield to hide behind.

The bigger shift in how I work: I no longer think of designer and developer as separate roles. They're two surfaces of the same skill: taste applied to product, taste applied to code. The thing AI gives back isn't "I can ship faster." It's "I never have to compromise on what to ship."