A big week for the core podcast workflow. We rebuilt the episode generation page from the ground up, made interview setup smarter with AI-suggested topics, and pushed a series of reliability fixes that make audio production significantly more stable in production.
New Features
Redesigned Episode Generation Page
Creating a new episode is now a guided experience rather than a single form. You choose your source — AI suggestions, a custom prompt, RSS feed, or an interview — and the page adapts to show only what you need. For interview-based episodes, a new segment wizard lets you review every clip from the recording, score them by quality, and hand-pick exactly which moments make the cut before generating.
A live duration bar tracks your selected content as you build, and a real-time progress view shows each step of generation as it happens — no more wondering whether your episode is still running.
AI-Suggested Interview Topics
When creating an interview invite, Mato now suggests conversation topics tailored to your guest. Topics are researched from the guest's public profile and framed from their perspective — so the suggestions feel relevant and specific rather than generic. Clicking a suggestion fills the topic field instantly, and you can refresh for a fresh set of ideas anytime. The suggestion cards are now full-width with the complete topic angle visible, replacing the cramped pill layout.
Improvements
More Reliable Audio Production
Two production failures that could silently kill an episode run are now fixed. Audio processing (the step that stitches all your host recordings together) now retries automatically on transient server errors with exponential backoff — a single hiccup no longer fails an entire run after 15 minutes of work. For long episodes, that final assembly step is now split into two stages with a checkpoint between them, keeping each stage within Vercel's execution limits.
Smarter Segment Include/Exclude for Interviews
When you toggle segments in and out of an interview on the edit page, those decisions now persist and carry through to the Generate Episode wizard. Previously, your selections could be overwritten by the AI's automatic scoring on each page load. Your manual choices now take priority.
Topic Suggestions Now Load Reliably
Interview topic suggestions were frequently timing out or falling back to generic placeholders due to an over-restrictive timeout. The AI model now has the time it needs to generate creative, guest-specific suggestions. Fallback placeholders still appear instantly if something goes wrong, so the invite form is never blocked.
Talent Voice Assignment Fixes
Saving a host's voice settings now works correctly whether you're using Gemini or ElevenLabs voices. Previously, saving a Gemini-only voice configuration could trigger a validation error because the form expected an ElevenLabs voice ID. Voice provider is now tracked explicitly, so switching providers updates correctly in both the talent profile and the podcast host assignment.
Redesigned Interview Camera Card
The camera consent step in the interview join flow has a new look. A persuasive card with animated highlights explains the value of enabling camera — your clips will be ready for Instagram, TikTok, and LinkedIn automatically. The card degrades gracefully on mobile and respects reduced-motion preferences.
Bug Fixes
Fixed a production issue where ElevenLabs voice requests failed silently because the agent ID stored in the environment had a trailing newline character. The system now normalizes these values before sending them to the provider.
Fixed episode delete from the episode card — the delete option in the menu now shows a proper confirmation dialog instead of silently doing nothing.
Fixed a React hydration mismatch in the Mato player's waveform display that caused inconsistent rendering between server and browser.