CrewOS.

The problem

Independent podcasters spend 5–10 hours per episode on post-production beyond recording — editing audio, writing show notes, drafting social posts, building newsletters, uploading to Spotify/RSS, posting LinkedIn, sending email. Creator burnout is the dominant attrition mode for podcasts that have audience traction but unsustainable founder workload.

The deeper failures of existing tools: time cost of writing platform-specific content from scratch every episode (most severe), forgetting to post on platforms or inconsistency in quality, and no memory of what worked — every episode re-decides tone, format, and structure as if it's the first. Descript, Riverside.fm, and Castmagic are excellent at parts of the workflow but none are fully agentic: the creator still drives every decision.

The solution

CrewOS turns post-production into an agentic workflow with memory. Upload the audio; the agent transcribes; reads Supabase preferences (tone, platform flags, content length); decides what content to generate based on the episode and preferences; executes all the content branches in parallel; returns outputs for approval.

The compounding moat is the memory layer: every episode, the creator's tone preferences, platform decisions, and content-length choices reinforce the Agent's model. By episode 10, the agent generates content that matches the creator's voice and platform strategy without prompts.

How it works

Pipeline: Creator uploads audio (m4a/mp3/wav, max 25MB) to Lovable UI → Make.com webhook receives the file → OpenAI Whisper transcribes → Claude reads the transcript + Supabase preferences → decides the task list based on platform flags → router executes the relevant content branches (show notes, summary, chapter markers, LinkedIn posts, YouTube descriptions, newsletter) → outputs return to Lovable for approval.

Required inputs: audio_file (binary), user_id (text, defaults to "default" for solo MVP). Optional preferences (stored in Supabase, read before each run): tone (conversational / educational / inspirational), platform flags (linkedin_enabled, youtube_enabled, newsletter_enabled), content length (short / medium / long).

Learning loop: skip LinkedIn on one episode → Supabase updates the linkedin_enabled flag → next run reflects the change. Edge cases handled: 2-min audio clip (agent identifies insufficient content), >25MB file (transcription timeout handled gracefully), transcript mentions "video" (youtube_enabled flag correctly set).

Who it's for

Primary (B2C): independent podcasters publishing 2–4 episodes/month who currently handle post-production manually and are losing the battle against burnout. They have audience traction but unsustainable founder time investment.

Secondary (B2B): content managers at churches, educational institutions, and media companies managing multiple shows — the volume creates the leverage where agentic automation compounds across episodes and shows simultaneously.

Pre-launch pilot: 3–5 podcasters in the creator's network for qualitative feedback on agent decisions and the learning loop feel before broader release.

Why it matters

500M+ global podcast listeners, AI cost reduction enabling per-episode automation economics, and no dominant AI-native agentic tool in the post-production category — the timing is structural. Descript, Riverside.fm, and Castmagic have validated demand but left the agentic wedge open.

CrewOS's structural bet is on agentic-with-memory as the defensible position. Workflow automation is easy to copy; a system that compounds creator-specific knowledge across episodes is much harder, because the moat is in the data + preferences + outcomes loop rather than in the prompts. Hallucination risk is low by design (agent doesn't generate content outside the transcript scope), and Anthropic + OpenAI policies are respected throughout — letting the product start narrow and earn trust before expanding.

At a glance

Project

CrewOS

Built by

Unekwu Ojoshaibu

One-liner

Genuinely agentic podcast production — perceive-decide-act-learn pipeline that auto-generates show notes, socials, newsletter, and YouTube content while learning creator preferences.