Introducing Roux: The AI Interviewer Built by Gumbo

Gumbo builds AI products. Usually that means building them for other teams, inside their stack and on their deadlines. Today we get to launch one of our own, and anyone can use it!

It's called Roux (pronounce it "roo"). I'm a South Louisiana boy, and where I'm from, a roux is how every good gumbo starts: flour and fat, stirred low and slow until it's dark as chocolate. Everything else in the pot builds on it. We needed a name for the thing that gathers what everyone knows before we build anything, and at the risk of over-using food metaphors, Roux was too good to pass on.

Why we built Roux

Every project we take on starts with the same question: what does this team actually need us to build? The answers are in dozens (or hundreds) of people's heads, and getting them out used to eat a week of meetings, many Slack threads, and late nights talking to Claude about Granola transcripts.

We tried the old way of gathering and synthesizing context. Forms were easy to send, but the answers came back thin. One-on-one interviews went deep, and at an hour each, we could only get through a handful of people. Group calls mostly gave us the opinions of whoever unmuted first (this is a problem with most meetings, and another reason why we think Roux will be a huge unlock for teams of all shapes and sizes).

So what did we do? We built the interviewer we kept wishing we had. We ran it on our own client work for a while, and it got pretty obvious it was too useful to keep to ourselves. At one client, Roux collected and synthesized 124 responses into a polished deck in 48 hours.

We still use it on ourselves all the time. Last week we were putting together the Gumbo newsletter and wanted to hear from our whole forward deployed team, since they're the ones inside client teams every day. So we sent them a Roux asking what they're seeing with clients when it comes to AI adoption.

Within the hour, we had everyone's answers back, with the context behind them! It gave us useful insights for how we manage client work, for our marketing, and for the newsletter itself. Nobody had to block off an afternoon for it, which the team appreciated more than they'll admit.

What Roux does

Roux is an AI interviewer. You tell it what you want to learn in plain words, and it takes it from there.

• It writes the interview. Roux turns your goal into an actual conversation, with sections that fit what you're after.
• It shares as one link. Nobody needs an account. People can type or talk from their phone or laptop, whichever is closer.
• It asks the follow-ups. Roux listens to what each person says and picks the next question from there, like a good interviewer would.
• It gives you one clear picture. Every response rolls up into themes, real quotes, and next steps, so nobody's reading 40 transcripts on a Friday night.

Roux also plugs into the tools your team already lives in, like Linear, GitHub, Slack, Jira, Notion, PostHog, Amplitude, Mixpanel, and Google Analytics. It pulls that context right into your interviews so it can ask better questions, and it can send findings back out as tickets, Slack messages, or Notion pages. When you're done, you can sync the report straight to ChatGPT or Codex.

What we use it for

Honestly, any meeting that mostly exists to collect what people think could be a Roux. A few of our favorites:

• Kicking off a new engagement. We get a whole client team's input in a day, read the summary right after, and talk it through with them while it's fresh.
• Budget reviews. Send a Roux to every department head beforehand and walk in already knowing where the gaps are.
• Project handoffs. Roux interviews the team that's moving on, so their context stays behind after they go.
• Product info gathering. Send one link to sales, support, and customers, and get back one summary of everything they said.

A growing library of "recipes"

Roux comes with dozens of prebuilt interviews out of the box: from battle tested interviews like Project Kickoff, Brand Discovery, and Team Checkins, to well known frameworks like Garry Tan's YC Office Hours and Matt Pocock's Grill Me.

Why an AI studio ships its own products

Building our own products makes us better at the work we do for clients. We build what we launch, launch what we build, and teach teams to run all of it, and each one makes the other two sharper. Roux is the best example we've got so far.

It's also multiplayer AI the way we always hoped it would work. By multiplayer, we mean your people and your AI working from the same context, where you can trace every answer back to the person who gave it. Roux pulls in context from the tools your team already uses, gets everyone's input into one place, and remembers who said what.

Send your first Roux

Roux is live today! Head to helloroux.com, tell it what you want to learn, and send your team the link this week. You'll probably have answers before you'd have found a time that works for everyone.

Skip the meeting. Send a Roux.