CrewOS.

CrewOS is a genuinely agentic AI podcast production pipeline built on a perceive-decide-act-learn loop — not another point-and-click editing tool. Upload a podcast audio file once and the agent transcribes, decides what content to generate, executes across platforms in under 60 seconds, and learns the creator's preferences over time so each episode makes the agent smarter.

Five hard differentiators against Descript, Riverside.fm, and Castmagic (none of which are fully agentic): (1) Genuinely agentic — perceive-decide-act-learn loop, not workflow automation; (2) Memory layer — preferences (tone, platform flags, content length) persist and improve across episodes via Supabase; (3) Multi-platform distribution in one workflow — show notes, chapters, summary, LinkedIn posts, email newsletter, YouTube descriptions all generated together; (4) Approval-first — agent proposes, creator approves before publish; (5) Zero prompt engineering required — creator just uploads.

Architecture: Lovable UI for the creator interface; Make.com webhook receives the file; OpenAI Whisper transcribes; Claude reads the transcript + Supabase preferences and decides the task list; a router executes the relevant content branches (show notes, summary, chapters, LinkedIn, YouTube, newsletter); outputs return to Lovable for review. The learning loop closes via Supabase — skip LinkedIn once, the Agent Memory card updates the platform flag, and the next episode's plan reflects the preference without re-asking.

Targeting 500M+ podcast listeners globally with a wedge into the post-production labor cost (5–10 hours per episode for independent creators publishing 2–4 episodes/month). B2C for independent creators; B2B for podcast networks, churches, media agencies managing multiple shows. Pre-launch pilot with 3–5 podcasters in the creator network, with scale trigger at 50+ users (upgrade Make.com plan, add user_id-based preference isolation, full per-user Supabase row-level security).