Pipod

Most people want to ask one question. A few want to produce a show.

Behind every episode there are a lot of settings: format, hosts, language, length, depth, sources. Most listeners don’t care about any of them. They have a question and want to hear the answer. A smaller group wants to direct the whole thing, with their own PDF and their favourite hosts.

And under both, there’s the wait. An AI needs time to write and voice an episode, and waiting is the opposite of pressing play.

Home used to be a catalog. Now it opens on a question.

My first designs were built for browsing. Home opened on mood chips (Curious, Inspired, Study, Work) above rows of episodes, podcasts and hosts. Quick Episode was already there, but behind the + button, one option out of three.

That made Pipod look like any other podcast app, and the one thing it does that others don’t was two taps away. So the question moved to the top of Home: “What do you want to hear about?”, typed or said. The mood chips went, and the + menu lost an option once Studio could create podcasts too.

BeforeHome

ShippedHome

BeforeCreate menu

ShippedCreate menu

If Pipod misreads a request, you should see it before it costs a credit.

Quick Episode reads your request and comes back with a plan: format, vibe, language and hosts, each one a pill you can change. Early on, a short request, “Ai au marché marocain” (AI in the Moroccan market), came back as an episode about souk crafts. The search had dropped “Ai”, and the script followed the travel pages it found.

I fixed the search and the prompts, and I changed the interface so the next misread wouldn’t go unnoticed. The plan now starts with “Understood as…”, Pipod’s reading of your request in plain words. If it’s wrong, you see it right there, above a Generate button that shows the cost. Edit the request and Pipod reads it again.

Ask

Type it, or tap the mic and say it.

Reading

The plan’s outline shows while Pipod reads.

The plan

“Understood as” first, then pills you can change.

An episode no longer needs a podcast first.

Studio is for people who want control. In the earlier design it began with “Pick a podcast, then create a new episode”, so a one-off episode meant creating a whole show first. Now an episode can stand on its own. Adding it to a podcast is optional, and picking one fills in that podcast’s hosts and style, which you can still change for this episode.

The long list of conversation vibes (Myth Busters, Meme Chats, Friendly Debate and more) became two levels: pick a vibe, then one of the few styles that fit it. Every setting sits on one page, and length shows its cost before you choose.

BeforePodcast first

ShippedTopic and length

Depth and vibe

Style and format

Hosts and ambience

How credits work

A Short (3 to 6 minutes) costs 1 credit. An In-Depth episode (15 to 20 minutes) costs 3. Voices are about 95% of what an episode costs to make, so that’s what credits follow. If an episode fails, or you cancel it before it plays, the credits come back.

I stopped designing the progress screen and removed the wait instead.

In the first version, an episode took minutes to make. I designed a progress screen with three stages (sources, script, voices). It was clear, but it was still a wait.

Now episodes stream. The script is written line by line, each line is voiced as soon as it’s ready, and the player starts on the first one, about six seconds after you ask. The episode is marked Live while the rest arrives, and you can scrub once it’s finished. The transcript follows along, and the mini-player keeps it going while you browse.

Player

Change the ambience or follow the transcript.

Transcript

Every line, who said it and when.

Mini-player

The episode keeps playing while you look around.

Onboarding changed more than any other flow.

It went through four versions in five weeks, the last one after launch. Each time I was weighing the same thing: how much to ask before someone hears their first episode.

1. Oct 1

   Pick a few topics, then a free first episode, playing within a minute.

2. Oct 4

   Three steps: name and language, topics, first episode.

3. Oct 14

   Welcome, language and topics move before sign-in, on new illustrated pages. The free first episode goes: new accounts start with their 3 monthly credits.

4. Nov 8

   A Start here card on Home suggests episodes from your topics, so the first one is still a tap away.

The whole app works in English, French, Spanish, German and Arabic. In Arabic the layout mirrors, down to the direction of the arrow button, and the hosts read numbers out as words.

Welcome

Language

Topics

In Arabic

Daily briefings that know when to stay quiet.

News Radar makes short audio briefings on a topic, right now or every day at a time you pick. Two rules keep it from turning into noise. If there’s no news on your topic, there’s no briefing and no credit spent. If three briefings in a row go unheard, the series pauses itself until you turn it back on.

The topic field is the same bar as on Home, so you can type it or say it. An earlier version had its own text field with an Enhance button. I took it out.

News Radar

A daily series

New briefing

I rebuilt Editor’s picks seven times and ended up close to where I started.

Editor’s picks are podcasts I plan and produce myself, grouped by subject, and they’re the first thing in the Library. I went through a deck of covers, a deck that folded like a Rolodex, and a looping stack that peeked above and below. They were fun to build.

But a stack shows one podcast at a time, and the Library is where people come to scan. Version 1.0.1 went back to subject chips and rows of covers, with the newest episodes underneath. The release notes call it “a refreshed Editor’s Picks”, which is a generous way to say I went back.

Editor’s picks, shipped

Profile

The create buttons that lived here became a streak and stats.

The seven versions

1. 1Featured cards and subject chips
2. 2No featured row, square covers
3. 3A deck of covers
4. 4A deck that folds like a Rolodex
5. 5A looping stack, peeking above and below
6. 6The stack without chips
7. 7Back to subject chips, now with rows of covers

The App Store page and the website were design work too.

I designed the App Store screenshots and pipodapp.com, which also hosts the pages people see when someone shares an episode with them. Pipod went live on November 6, 2026.

The App Store screenshots

I built Pipod myself, with AI coding tools, and that changed the design work. When waiting was the problem, I could change how episodes are made instead of polishing a loading screen. In September 2026 I moved the backend from Supabase and Python to Cloudflare so episodes could stream, and shipped about six weeks later.

App

Expo / React Native / TypeScript / React Query / Zustand

API

Cloudflare Workers / Hono / D1 / R2 / Durable Objects

AI

LLM via OpenRouter / Fish Audio voices / FLUX covers / Tavily search

Accounts

Sign in with Apple / RevenueCat subscriptions / push

Pipod has only been out for a few days, so I don’t have numbers worth showing yet. The ones I’m watching: how many new listeners play a first episode, how often people change the plan before tapping Generate, and whether the streak brings anyone back the next day.