Independent / Native Mac app
CueLoop
CueLoop is a native Mac app that helps you get more out of customer interviews. It records both sides of a call, transcribes it, coaches you live, and saves what you learned.
I built it with AI coding tools, and my product experience went into deciding what it should do.
Where the idea came from
I was sitting in on a colleague’s customer interview over Zoom, typing coaching prompts into a shared Google Doc. I could suggest a follow-up, flag a leading question or point out something worth exploring while the participant was still there to answer.
It helped, but it took two people. I wanted to see whether I could build something that offered that kind of help without needing someone else on every call.
That was around 2023 or 2024, well before coding agents, when AI mostly meant autocomplete and chatbots. My first tries came in early 2025: a Chrome extension that popped up interview tips, then a Mac menu bar app that recorded both sides of an interview. I kept coming back to the idea.
I think the time builders spend with their users is some of the most valuable time they have. You can still ask leading questions, miss something worth following up on, and come away feeling like you learned more than you actually did.
Building it and trying it
The first work was getting microphone and system audio into the app, then transcription and live coaching. From there, it grew into a compact native interface, onboarding, saved sessions, project context and a way to distribute the app.
Coming from a business background, I hadn’t traditionally been the person building the software. With AI coding tools, I could explain what I wanted, work through the decisions with an agent and then go try it in the app. If it wasn’t doing what I expected, we could figure out why and keep going.
Applying to Y Combinator in February 2026 and launching on Product Hunt that March showed me how much deadlines matter, even artificial ones. Without an end date, a project can go on forever, especially when you’re working with coding agents that never quit.
Deciding what to take out
Live Sharing started as a proof of concept. Another AI agent could read the conversation while CueLoop was running. But how often was I actually going to switch to ChatGPT in the middle of an interview to ask it something? I decided to remove it and keep the agent’s role around preparing for the call and reviewing it afterward.
What can one interview really tell us?
A product review pointed out that CueLoop could treat something from a single interview as validated. That needed fixing.
Hearing something in one interview doesn’t tell you how common it is or whether someone will pay for a solution. We changed the saved findings to describe what the participant reported, keep the supporting quotes and make the unanswered questions clear.
Rebuilding the coaching
By September, the live coaching was the part that wasn’t working, and it’s the part that makes the product different. One version fired something out every minute, which was too noisy. In a mock interview, a later version showed a single cue. I called it a complete failure. How could anyone use this?
I liked the interface, so I decided to replace the engine and keep everything around it. My first rebuild got over-engineered. After a couple of weeks, it still hadn’t shown a cue in the real app. So I started again with something smaller that only tried to show the right follow-up in time to ask it.
I also needed a way to test without running a mock interview every time, so we built synthetic ones. In one replay, cues went from 2 of 17 pauses to 7 of 17, and six of those seven appeared about a second after the participant stopped. Those are test numbers, not real interviews.
The interview is the interface
You’re in the middle of a live conversation, so you shouldn’t have to click around. I want the screen to change on its own based on what’s being said, so all you have to do is run the interview.
I saw what that could feel like by accident. During a test, a cue changed to the word “Cleared” once it was answered. It wasn’t supposed to happen, but it felt good. Now when you ask a cue, it gets a green check without you touching anything.
I’m taking that further with a prototype called the Interview Map, which draws the threads of a conversation as it happens, so you can see where it’s gone without clicking anything. It hasn’t shipped. The biggest risk is that it becomes a distraction.
Where it stands
The new coach went out in the app on October 13, and I’m still fine-tuning and testing it. A few trusted friends are trying it, but no one is using it in real interviews yet, so I don’t know whether the cues help.
With a young family and a full-time job, CueLoop has become a hobby I chip away at. But I keep coming back to it, and every time the technology underneath improves, the idea looks more promising.
If I started it today, I’d take it more seriously from the start, bring someone in to build it with me, and release it sooner. I tinkered with too many ideas, and I keep coming back to the core one: showing live cues during an interview.