# Our First Big HDR Episode, Edited by an AI Agent

We delegated the full edit of a 1.5-hour podcast to the Luna model through Codex—from bloopers and chapters to smart notes and true HDR. The episode explores vibe coding, synthetic CustDev, AI stacks, and context architecture.

Author: Maxim Tolmachev
Published: 2026-09-06
Language: en
Canonical: https://maximtolmachev.com/blog/our-first-big-hdr-episode-edited-by-an-ai-agent/
Original source: https://t.me/mtolm/104

![Our First Big HDR Episode, Edited by an AI Agent](/blog-images/tg-104.jpg)

“This 1.5-hour podcast was edited entirely by an AI agent”: our first big episode in HDR

We released the second episode of the “Big Potential” podcast with Misha Gorbunov 🎙

We recorded a long conversation about vibe coding, the future of product managers, and automation. But the most interesting part began during production.

I decided not to hand 1.5 hours of raw footage to a human editor (my wife 😅) and instead delegated the entire edit to the Luna model (on High effort) through Codex.

She edited the episode overnight for several days in a row, fine-tuning the timing down to the frame. And most importantly, the subscription limits were barely used—which is exactly what I talked about in my previous post on model economics.

Here’s what the agent did on its own, without any manual editing in Premiere 👇

🤖 What the AI editor delivered end to end:

• Bloopers in the intro: it found funny outtakes in the footage and turned them into a dynamic opening hook.

• Intro and chapters: it created the speaker introductions and visual transitions between topics.

• Dynamic progress bars: it calculated the timings and rendered a timeline bar for each thematic section.

• True HDR: it preserved the camera’s 10-bit dynamic range and rendered all generated graphics in HDR.

• Context-aware smart notes: it looked up the mentioned services (Lovable, Figma, UX Pilot) and methods (CustDev, Opportunity Scoring), then added pop-up cards with pricing and descriptions.

👥 Who’s on camera:

• Maxim Tolmachev: 10+ years in product management, CPO, building AI agents and automations.

• Mikhail Gorbunov: Operations Manager at Xsolla, former developer, analyst, and product manager. He builds his own AI apps and automations and looks at technology through the lens of engineering practice: databases, architecture, and security.

Two product managers sat down for an honest conversation about where AI is already saving weeks of work—and where human taste, systems thinking, and responsibility for the outcome are still essential.

🎙 What the episode covers (1 hour 15 minutes of substance):

1. Are junior specialists no longer needed? Why vibe coding is forcing everyone to become systems architects with refined taste.

2. Synthetic CustDev (the Palmolive case): how AI personas matched real respondents by 90%, and why this changes research.

3. 1,400 landing pages: how to cut the test launch pipeline from three weeks to three hours.

4. Our personal AI stack: local MLX Whisper on Mac (transcription in 5 minutes instead of 45), voice input with Handy + GigaAM, and Claude Code.

5. Context architecture: why we dropped heavyweight MCP servers and switched to Skills, saving 90% on tokens.

YouTube (best viewed on an HDR-capable screen):
youtu.be/Bixtvy2z4j0
