Workflow: Podcast Episode Production (AI & I)

Layer: L2 (Workflow) Date: 2026-04-01

Purpose

This workflow produces episodes of “AI & I,” Every’s podcast hosted by Dan Shipper. The podcast features conversations with builders who are actively using AI in their work — not commentators, not theorists, but practitioners with demonstrated experience. The show produces 41 episodes/year and has accumulated 83,957 listening hours, making it a significant audience touchpoint and credibility builder for Every.

Why it matters: The podcast is Every’s highest-trust medium. Articles make arguments; the podcast lets listeners hear builders think out loud. A great episode surfaces 3+ actionable insights that listeners can’t get from reading the guest’s blog or Twitter. A bad episode — a softball interview with a guest who talks about AI but hasn’t built anything — directly undermines Every’s builder credibility.

Trigger

A new episode cycle begins when:

  1. Rachel Braun (Podcast Producer) identifies a scheduling slot in the production calendar (targeting ~41 episodes/year, roughly weekly with breaks)
  2. Dan Shipper identifies a guest he wants to interview based on something they’ve built or published
  3. An editorial opportunity arises (e.g., a major AI development where Every has a unique perspective from a builder)

Steps

Step 1: Topic and Guest Selection

  • Responsible party: Dan Shipper (host) + Rachel Braun (producer)
  • Input: Production calendar with upcoming slots, current AI landscape, Every’s editorial priorities
  • Process:
    1. Guest identification: Dan maintains a running list of potential guests — people he’s encountered through writing, conferences, reader recommendations, or the AI builder community. Rachel also flags potential guests based on listener requests and trending topics.
    2. Builder credibility screen (non-negotiable): Every guest must pass this filter:
      • Has the guest actually built something with AI? Not advised, not invested in, not written about — built.
      • Can the guest speak to specific, concrete details about what they built, what worked, and what failed?
      • Would Every’s audience learn something actionable from this person’s experience?
      • If the guest is primarily a “thought leader” who doesn’t build, they are not a fit — regardless of their audience size or name recognition.
    3. Topic framing: Dan and Rachel discuss the angle for the episode. Not just “Interview [Guest] about AI” but a specific frame: “How [Guest] rebuilt their company’s entire workflow using AI agents” or “What [Guest] learned about AI reliability from shipping to 10,000 users.” The frame determines which questions to prepare.
    4. Diversity and coverage check: Rachel reviews recent episodes to ensure variety in guest backgrounds, industries, company stages, and AI application areas. The show shouldn’t inadvertently become “the same type of guest every week.”
  • Output: Confirmed guest with topic frame and target recording date
  • Criteria for completion: Dan approves the guest and topic frame. Guest passes the builder credibility screen. Recording date is on the production calendar.

Step 2: Guest Outreach and Scheduling

  • Responsible party: Rachel Braun
  • Input: Approved guest and topic frame
  • Process:
    1. Rachel reaches out to the guest (or their team) via email or DM. The outreach message:
      • Explains what AI & I is and Every’s audience
      • States the specific angle/topic frame (not generic “we’d love to have you on”)
      • Provides logistics: estimated recording length (typically 45-75 minutes), format (video call via StreamYard or equivalent), and estimated publication timeline
    2. Rachel schedules the recording, confirming:
      • Date and time that works for Dan and the guest
      • Technical setup requirements (good microphone, quiet space, stable internet)
      • Any pre-recording materials needed (e.g., demos the guest wants to screen-share)
    3. Rachel sends a calendar invite with all technical details and a brief prep document.
  • Output: Confirmed recording date on calendar, guest has received prep materials
  • Criteria for completion: Recording is scheduled, guest has confirmed, calendar invite sent with all logistics
  • Time budget: 1-2 weeks (scheduling can take time with busy guests)

Step 3: Recording Preparation

  • Responsible party: Dan Shipper + Rachel Braun
  • Input: Confirmed guest, topic frame, guest’s published work and background
  • Process:
    1. Research: Dan reviews the guest’s recent work — articles, talks, products, social media posts. The goal is to arrive at the conversation knowing enough to ask specific questions, not generic ones. “Tell me about your AI strategy” is generic. “You mentioned in your March blog post that you rebuilt your underwriting pipeline with Claude — what broke the first time?” is specific.
    2. Question preparation: Dan drafts 8-12 questions organized in a loose arc:
      • Opening: What did you build, and why?
      • Middle: What surprised you? What failed? What did you learn that wasn’t obvious?
      • Depth: Technical specifics that the audience (AI-savvy builders) will find valuable
      • Closing: What would you do differently? What’s next?
    3. Rachel’s production prep:
      • Verify recording platform (StreamYard or equivalent) is configured
      • Test audio/video settings
      • Prepare intro/outro scripts
      • Note any specific segments to highlight for social clips
    4. Dan and Rachel do a brief (5-10 minute) pre-recording alignment: What’s the one insight we most want to surface in this episode?
  • Output: Dan’s prepared question list, Rachel’s production checklist completed
  • Criteria for completion: Dan has reviewed the guest’s work and prepared specific questions. Rachel has confirmed all technical setup. Both have aligned on the key insight target.
  • Time budget: 1-2 hours of prep per episode

Step 4: Recording

  • Responsible party: Dan Shipper (host), Rachel Braun (producer/tech)
  • Input: Prepared questions, functioning recording setup, guest on the call
  • Process:
    1. Rachel handles technical setup — confirms audio levels, recording is active, backup recording is running.
    2. Dan and the guest have a brief (2-3 minute) warm-up conversation before recording starts (helps the guest relax and sets the conversational tone).
    3. Dan conducts the interview. Key principles:
      • Follow curiosity, not the script. The prepared questions are a safety net, not a railroad. If the guest says something surprising, follow that thread.
      • Ask “how” not “what.” “What’s your AI strategy?” gets platitudes. “How exactly does your team decide which tasks to give to AI vs. keep human?” gets insight.
      • Push past the surface. If the guest gives a generic answer, ask for the specific example. “Can you give me a concrete case?” is always a valid follow-up.
      • Make it a conversation, not an interview. Dan shares his own experience where relevant. “We found the same thing at Every — when we…” This is not about Dan; it’s about making the guest comfortable going deeper.
      • Don’t softball. If the guest makes a claim that seems too good to be true, ask about it respectfully. “That’s impressive — what didn’t work along the way?”
    4. Rachel monitors audio quality throughout and flags any issues (background noise, echo, lost connection) to Dan via chat.
    5. Target recording length: 45-75 minutes raw (will edit down).
  • Output: Raw recording file (audio + video)
  • Criteria for completion: Full conversation recorded with acceptable audio quality. Both Dan and Rachel feel the conversation surfaced at least 3 specific, actionable insights.
  • Post-recording: Dan and Rachel do a 5-minute debrief: What were the best moments? Anything to flag for editing? Any follow-up needed with the guest?

Step 5: Post-Production

  • Responsible party: Rachel Braun (using Descript and AI-assisted editing tools)
  • Input: Raw recording file, recording debrief notes
  • Process:
    1. Transcript generation: Import the recording into Descript, which generates a time-stamped transcript automatically.
    2. Editing (Descript + manual):
      • Remove dead air, long pauses, verbal fillers, and technical glitches
      • Cut tangents that don’t serve the listener — if a 5-minute segment doesn’t contain an insight or advance the conversation, cut it or tighten it
      • Preserve the natural conversational flow — editing should be invisible, not robotic
      • Flag the strongest segments (highest-insight moments) for social clip extraction in Step 7
      • Target edited episode length: 30-60 minutes depending on content density
    3. Audio cleanup: Normalize audio levels between Dan and guest. Remove background noise. Ensure consistent audio quality throughout.
    4. Add intro/outro: Insert the standard AI & I intro and outro, including any episode-specific sponsor reads or announcements.
    5. Rachel’s quality check: Listen through the full edited episode to verify:
      • Audio quality is clean and consistent
      • Edits are seamless (no jarring cuts)
      • The episode flows logically from topic to topic
      • At least 3 clear, actionable insights are present in the final edit
  • Output: Edited episode file ready for distribution
  • Criteria for completion: Rachel approves the edit. Dan listens to the final cut and confirms it represents the conversation accurately. No listener would be confused or find the editing distracting.
  • Time budget: 3-5 hours of production work per episode

Step 6: Show Notes, Transcript, and Metadata

  • Responsible party: Rachel Braun (AI-assisted)
  • Input: Edited episode, Descript transcript
  • Process:
    1. Show notes: Write a concise summary that:
      • States who the guest is and what they’ve built (builder credibility upfront)
      • Lists 3-5 key takeaways with timestamps so listeners can jump to what interests them
      • Includes links to the guest’s work, tools mentioned, and related Every articles
      • Written in Every’s voice — conversational, specific, not corporate
    2. Transcript cleanup: Review the AI-generated transcript for accuracy. Fix names, technical terms, and any AI transcription errors. The transcript should be readable as a standalone document.
    3. Metadata: Prepare episode title, description, tags, and category information for podcast platforms.
    4. Episode title format: Specific and thesis-driven, not generic. “How [Guest] Rebuilt [Thing] with AI” not “Talking AI with [Guest].” The title should make someone want to listen.
  • Output: Show notes, cleaned transcript, episode metadata
  • Criteria for completion: Show notes include timestamps and key takeaways. Transcript is accurate. Title is specific and compelling.
  • Time budget: 1-2 hours

Step 7: Social Clip Generation

  • Responsible party: Rachel Braun + Anthony Scarpulla (social distribution)
  • Input: Edited episode, flagged highlight moments from Step 5
  • Process:
    1. Clip selection: From the moments flagged during editing, select 2-4 clips (30-90 seconds each) that capture the highest-insight moments. Each clip should stand alone — a listener who sees only the clip should understand the insight without needing the full episode.
    2. Clip production: Edit clips for social media format:
      • Add captions/subtitles (most social video is watched on mute)
      • Add Every branding (minimal — logo, episode number)
      • Optimize format for each platform (square for Instagram, landscape for YouTube, etc.)
    3. Social post drafts: For each clip, Anthony’s team generates accompanying post text that:
    4. Dan approves the final clips and social posts before publication.
  • Output: 2-4 social clips with accompanying posts, ready for distribution
  • Criteria for completion: Clips capture genuine insights (not just entertaining moments). Posts pass the social gate. Dan has approved.
  • Time budget: 2-3 hours

Step 8: Distribution

  • Responsible party: Rachel Braun
  • Input: Final episode file, show notes, transcript, social clips
  • Process:
    1. Podcast platforms: Upload the episode to all distribution platforms (Apple Podcasts, Spotify, YouTube, and others). Include show notes, transcript, and metadata.
    2. Every’s website: Publish the episode page on Every’s site with embedded player, show notes, full transcript, and related content links.
    3. Newsletter mention: Coordinate with the editorial team to include a mention of the new episode in the next newsletter, with a specific hook (not just “new episode out” but “In this week’s AI & I, [Guest] explains why [specific insight]”).
    4. Social rollout: Anthony publishes social clips on a staggered schedule (not all at once) to maximize reach across platforms.
    5. Guest notification: Send the guest a link to the published episode with thank-you note and social assets they can share.
  • Output: Episode live on all platforms, social promotion active, guest notified
  • Criteria for completion: Episode is accessible on all target platforms. Social clips are scheduled or published. Guest has received their link and assets.
  • Time budget: 1-2 hours

Execution Model

  • Sequential dependencies: Step 1 (Guest Selection) → Step 2 (Outreach) → Step 3 (Prep) → Step 4 (Recording) → Step 5 (Post-Production) → Step 6 (Show Notes) → Step 7 (Social Clips) → Step 8 (Distribution)
  • Parallel stages: Steps 6, 7, and 8 can run in parallel — show notes, social clips, and distribution prep can happen simultaneously after post-production. Step 1 for the next episode can begin while Steps 5-8 of the current episode are in progress.
  • Convergence point: Step 8 (Distribution) — where the edited episode, show notes, social clips, and platform metadata all come together for launch.
  • Blocking gates: Builder credibility screen in Step 1 (non-negotiable). Social Media Publication Gate before Step 7 clips are posted.
  • Dan dependency: Steps 1, 3, and 4 require Dan’s direct involvement (guest selection, prep, hosting). This is a structural bottleneck — the podcast is Dan’s personal brand surface.

Feedback Loops

  • Listener Data → Guest Selection: Rachel tracks listener engagement per episode (downloads, completion rate, listening hours). High-engagement episodes inform future guest selection and topic framing.
  • Social Clip Performance → Clip Selection: Which clips get the most engagement? Anthony and Rachel refine clip selection criteria based on performance data.
  • Guest Feedback → Process Improvement: Post-recording feedback from guests (was the process smooth? were they prepared? was the conversation what they expected?) improves Steps 2-3.
  • Episode Insights → Articles: Conversations sometimes surface insights that become Chain of Thought articles or Source Code pieces, creating cross-content reinforcement.

Quality Gate

There is no dedicated podcast publication gate in the current gate architecture. Quality is enforced through:

  1. Guest selection (Step 1): Builder credibility screen is the primary quality gate — it’s a hard filter, not a soft preference. Guests who don’t build are not booked, period.
  2. Editorial review (Steps 4-5): Dan’s host judgment during recording and Rachel’s editing judgment during post-production serve as taste gates. The debrief after recording explicitly checks: “Did we get 3+ actionable insights?”
  3. Social distribution: Social clips pass through the social-media-publication gate before posting.

Recommendation: If podcast volume grows beyond 41 episodes/year or if new hosts are added beyond Dan, a formal podcast-publication gate should be designed following the pattern established for articles and consulting deliverables.

Stranger Test

Could someone with zero context execute this workflow from this spec alone?

For production (Steps 2-3, 5-8): Yes. Rachel’s production role is well-specified and follows standard podcast production practices. A competent podcast producer could follow these steps.

For editorial (Steps 1 and 4): Partially. The guest selection criteria are explicit (builder credibility screen), but:

  1. Dan’s interview style: Dan’s ability to follow curiosity, push past surface answers, and connect the guest’s experience to Every’s is a taste skill built over 41+ episodes. A new host would need to listen to 10+ episodes to internalize the style, and even then, the specific rapport Dan builds comes from his own builder experience. The spec tells you what to do (“follow curiosity, not the script”) but not how to do it at Dan’s level.

  2. Guest network: Dan’s guest pipeline comes from his personal network, writing, and reputation. A stranger wouldn’t have this network. Rachel can help with outreach, but guest identification requires knowing the AI builder community.

  3. Tool access: Rachel needs access to Descript (editing), StreamYard or equivalent (recording), podcast distribution platforms (Apple, Spotify, YouTube), and Every’s website CMS. Standard tools, but accounts and credentials are needed.

  4. Social distribution pipeline: Step 7 depends on Anthony’s social media system. A stranger would need access to Anthony’s tools and approval workflow.

  5. Production calendar and editorial coordination: Rachel coordinates with Eleanor and the editorial team on newsletter mentions (Step 8). A stranger needs to know who to contact and when.

Verdict: A podcast producer could execute the production pipeline from this spec. The editorial/hosting function is Dan-dependent and not fully transferable through documentation — it requires builder credibility, interview skill, and a network. If Dan were unavailable, the spec provides enough structure for a substitute host to deliver an adequate episode, but not an exceptional one.

Iteration Protocol

This workflow improves through the following mechanisms:

  1. Listener metrics tracking: Rachel tracks downloads, completion rates, and listening hours per episode. High-performing episodes are analyzed: Was it the guest? The topic? The specific angle? Patterns feed back into guest selection (Step 1) and topic framing.

  2. Social clip performance: Track which clips get the most engagement. High-performing clips reveal what the audience values most — specific technical details? Counterintuitive insights? Personal failure stories? Use these patterns to guide what Dan probes for in future interviews.

  3. Guest feedback loop: After publication, ask guests how the experience was. What would make them recommend the show to other builders? Their feedback reveals production friction and quality perceptions from the guest side.

  4. Episode retrospective: Rachel and Dan briefly review each published episode: What moment was the best? Where did the conversation drag? What question should Dan have asked but didn’t? These notes accumulate into Dan’s interview playbook.

  5. Builder credibility screen refinement: As the AI landscape evolves, “builder credibility” changes. Someone building with GPT-3 in 2023 was impressive; the same work in 2026 is table stakes. Rachel and Dan periodically recalibrate what constitutes “has actually built something meaningful” for guest selection.

  6. Production efficiency tracking: Rachel tracks hours spent per episode across each production step. If post-production consistently takes 5+ hours, investigate whether Descript’s AI editing has improved enough to reduce manual work. If scheduling consistently takes 2+ weeks, consider building a scheduling agent or using Calendly-style automation.

  7. Cross-pollination with articles: When a podcast conversation surfaces an insight that deserves deeper exploration, flag it for the editorial team as a potential article topic. When an article generates strong reader response, flag the topic for a potential podcast episode. The two workflows should feed each other.