I put out a podcast every week, but I don't lose days to production anymore. My own AI pipeline chews through everything except the conversation. Here’s the system, straight up.
Manual Podcasting: Absolutely Brutal
What I did before (never again):
-
Pre-Production (~3 hours):
- Guest research: 1 hour — scouring guest sites, prepping
- Outline questions: 30 min
- Write intro/outro: 30 min
- Tech wrangling: mic tests, Zoom disasters — 1 hour
-
Recording: 1 hour per episode
-
Post-Production (this hurts):
- Edit audio: Chop ums/uhs, level — 2 hours
- Add intro music: 15 min
- More enhancement nonsense: 30 min
- Transcribe: 30 min hoping Whisper didn’t hallucinate
- Show notes: 45 min
- Upload & publish: 15 min fighting platforms
Total: 8 hours per episode. If you do this each week: welcome to spending 32 hours a month glued to your DAW.
My AI Pipelines: 45 Minutes, Done
Here’s what I built:
Pre-Production:
- Guest research: Full auto via Python, runs while I sleep
- Questions: Generated in my style
- Scripts: AI writes them based on my notes or podcast context
My time spent: 0. It’s all hands-off.
Recording:
- I show up, talk for 45 minutes, that's it
Post-Production:
- Edit audio: Descript API handles all the slicing
- Transcription: AssemblyAI, called from Python
- Show notes: GPT does summaries, tags, links
- Publishing: Scripts push everywhere.
My time: Still zero. Agent wraps up edits, notes, and upload in 20 minutes, no human in the loop.
So:
- I spend 45 minutes per episode
- The agent (collection of scripts & prompts & APIs) covers 7+ hours every week
The Actual Components
Guest Research Agent
- Scrapes guest info, pulls social handles, finds their best video examples
- Code:
class PodcastGuestResearcher:
# Pulls LinkedIn, YouTube, and podcast guest spots. Outputs Google Doc.
...
Script Writer Agent
- Composes intro/outro and transitions. I wrote a custom prompt for this in brand_cron.py, and it spits back in my own voice/style.
class PodcastScriptWriter:
# Uses GPT-4.1, Claude Sonnet, picks best samples with local Ollama if required
...
Audio Editing Agent
- Drops in raw audio, calls Descript API for filler word removal, levels, and intro music
- Sometimes I swap in Playwright+CDP with browser-level automation if I need Adobe Podcast AI instead
Transcription & Show Notes Agent
- AssemblyAI generates transcript
- Claude or local Qwen summarizes into bullet show notes, timestamps, links
class ShowNotesGenerator:
# Drop in raw MP3, it emails final markdown
...
Publishing Agent
- Bash script triggers social_poster.py, which fires out promotions to @billy_kennedy_bmx, @nepa_ai, and uploads to Transistor
class PodcastPublisher:
# Instant publish, then calls vision_social_poster for social reels/images
...
Putting It All Together
Every Sunday night:
class PodcastProductionAgent:
# Schedules guest research, script generation, editing, publishing, social push. All chained.
...
Whole pipeline is easily dockerized on my Axon server.
Cost Breakdown
- Descript API: $24/month for edit
- AssemblyAI: ~$0.90/hr audio
- Transistor.fm: $19/month for hosting
- OpenAI/Claude APIs: $10-20/month depending on usage
$65-85 monthly — I used to spend that just buying extra cloud storage for raw WAV files. Net saved time: 28 hours every month.
Real-World Results
Manual:
- 8 hours each ep (painful)
- 32 hours/month (who has this?)
- Quality: hit-or-miss, often rushed
Automated: - 45 mins per ep (just the talk)
- 3 hours/month, zero burnout
- Consistent releases, notes, quality
Want to Try? My Build Plan
Saturday (3 hours):
- Open Descript, upload a test file, see what their API can do
- Spin up a simple guest research script (I’ll just yank my PodcastGuestResearcher class)
- Generate a sample script with GPT, tweak the prompt
Sunday (3 hours):
- Record something, toss it into the post-prod pipeline
- Run through the edit and show notes workflow
- Review: everything publish-ready in one push
Second weekend:
- Automate all publishing/social/scheduling
Third weekend:
- Fully automated episode, start-to-finish without touching your laptop
Wrap-Up
Podcasting can drown you in grunt work. Or you can build and deploy your own agent army and get your life back.
- Guest research, scripts, editing, notes, social: all automated
- Real stack, tested weekly.
- Tools: $65-85/month.
- Time saved: 91%
You can spin this up in two weekends.
Curious? See my actual code and workflows: axon.nepa-ai.com