Podcast Summary: "Nobody wanted to do this work": How Emmy Award–winning filmmakers use AI to automate the tedious parts of documentaries

Podcast: How I AI
Host: Claire Vo
Guest: Tim McLear, Producer & Technologist at Ken Burns’ Florentine Films
Date: November 17, 2025
Length: ~48 minutes

Overview

The episode explores how Tim McLear—an Emmy Award–winning producer at Florentine Films—uses AI not for creative content generation, but as a practical tool to automate the "technical mess" of post-production in documentary filmmaking. The conversation is a deep dive into building custom AI-powered tools that automate data entry, image and metadata management, field research workflows, and archival document transcription, freeing creative teams from the drudgery of tedious, manual work.

• Main Theme:
  Using AI to automate, organize, and optimize the vast, repetitive tasks behind the scenes in documentary production, enabling teams to focus on research and storytelling.

• Purpose:
  To demystify how non-engineers and creative professionals can practically apply AI in highly specific, impactful ways—beyond flashy generative applications—by building or leveraging their own workflows.

Key Discussion Points & Insights

1. The Problem: Post-Production is a Data Mess

Timestamps: 00:00–04:45

• Documentary filmmaking involves handling hundreds of hours of footage and tens of thousands of images, which must be meticulously logged, tagged, and made searchable for future use.
• Manual data entry ("logging") is time-consuming and widely disliked.
• The shooting ratio for documentaries can be extreme. Example: For an 8-hour Muhammad Ali series, over 20,000 stills and 100+ hours of footage were gathered.
  • Quote:
    "Post production is like a technical mess of media management. ...The data management piece...is the mess that I have used AI to tackle."
    (Tim, 00:08)

2. From ChatGPT to Custom Automation: The Evolution of the Workflow

Timestamps: 05:05–15:15

• Initial Inspiration: The arrival of ChatGPT’s image upload was an "aha!" moment, opening doors for automated database entry.

  • Quote:
    "It was this insane day for us...an aha moment where it was like, ‘Oh my god, this thing can see.’"
    (Tim, 05:21)

• Early Days:

  • Started with simple scripts to generate descriptions of images via OpenAI’s vision models.
  • Quickly realized that generic AI-generated descriptions weren’t sufficient; context (e.g., exact location, date) was missing.
  • Solution: Designed scripts to first extract embedded metadata and append it to AI prompts—elevating both accuracy and richness of database entries.
    • Quote:
      "When you give [AI] the tools and information to write a better description, it’s gonna get there."
      (Tim, 10:46)

• Choosing Models:

  • Used OpenAI for vision/image analysis (first to market with vision APIs).
  • Used Anthropic Claude models for coding via automated prompting in tools like Cursor.
  • Quote:
    "All the vision preview API calls were already there, and the switching costs were too much."
    (Tim, 12:24)

3. Scale and Sophistication: Building a Custom, Extensible AI Logging System

Timestamps: 12:41–21:22

• Current State: A REST API called “Autolog” handles multi-step metadata extraction for any file—stills, video, music.

• For images:

  1. Gather file specs
  2. Move/copy files
  3. Parse for metadata
  4. Scrape web for additional info
  5. Generate robust, AI-assisted descriptions

• For video, uses frame sampling (at 5-second intervals for efficiency), extracts key stills, audio transcripts (with Whisper), and aggregates all into detailed, prompt-based reasoning for AI models to produce descriptive events/summaries.

  • Quote:
    "For the frame captions themselves, I’ll use a cheap model… But then all the data goes to a reasoning model to get what’s happening in the video."
    (Tim, 15:25)

• Embeddings power semantic search, making archives vastly more discoverable:

  • Uses CLIP for image embeddings, OpenAI models for text, then fuses them.
  • Editors can now "find similar" images or assets by vibe/similarity instead of crude keyword matching.
    • Quote:
      "Now the ability to discover semantically is, I think, the most robust part of the system."
      (Tim, 19:38)

4. Automating Field Work: The ‘Flip Flop’ App for Archival Research

Timestamps: 24:44–31:35

• Pain Point: In archives, teams take massive numbers of iPhone snaps (front and back of images) that used to be an organizational nightmare.
  • Prior method: chaotic camera rolls, hard-to-match image pairs.
• Solution:
  • Tim "vibe coded" (AI pair-programmed) an iOS app called Flip Flop.
  • Lets users create collections (folders), snap fronts and backs, associate pairs, and embed OCRed backs as EXIF metadata in real time.
    • Quote:
      "Now anytime anybody uses one of these images…that image is embedded with that metadata."
      (Claire, 29:19)
  • Drastically streamlines organization and retrieval—even for people outside the app ecosystem.
    • Quote:
      "I had two colleagues out in the field a couple weeks ago…they came back with 1400 images…Flip Flop is certainly making the process easier."
      (Tim, 29:33)

5. AI for Archival Document OCR: 'OCR Party' App

Timestamps: 32:03–36:44

• Problem: Transcribing only portions of historical documents (old newspapers, handwritten letters) is tough; generic OCR is often not accurate.
• OCR Party: A custom Mac menubar app lets users crop the document, choose between Apple’s vision OCR or an AI API, and extracts just the needed text—including challenging sections.
  • Also highlights AI's strength in reading cursive, noisy scans, translation, and partial documents.
  • Genealogy use case: Claire’s mother uses similar workflows to transcribe historical names from old documents, which AI excels at.
    • Quote:
      "AI is really good at OCR of old documents. It’s really good at handwriting. It’s pretty good at translation, too."
      (Tim, 32:20)

6. Broader Lessons & Takeaways

Timestamps: 31:35–36:44

• Custom AI-Enabled Tools:

  • With today’s AI, creative and technical professionals can "vibe code" specific, high-impact personal tools—which would never make sense for a commercial product.
  • Apply AI + light software engineering to domains/files you know well.
    • Quote:
      "No one was going to make me this app. And so the ability to make an extremely specific app…has been an unbelievable moment."
      (Tim, 35:39)

• AI and File Types: Consider what information can be embedded or extracted from various file types—a new use case for AI emerges when you “load up” files with context via code.

Lightning Round: Philosophy and the Future

Timestamps: 37:46–44:49

• Learning & Upskilling:

  • Tim draws analogies with learning creative tools (Photoshop, Premiere), seeing AI coding as creative design more than hard engineering.
    • Quote:
      "Coding feels so much more creative than technical...these tools feel really like creation engines."
      (Claire, 39:06)

• AI Skepticism & Ethics in Film:

  • The industry is wary of generative video’s potential for job displacement and authenticity risks.
  • Tim distinguishes:
    • Nonfiction/archival: Should not generate fake footage or mislead.
    • Practical AI tools: Empower researchers and creatives by cutting toil, not replacing core skills.
    • Quote:
      "We should not be generating archival footage. ...[but] there’s a place in the process for it which allows you a place to learn without thinking it needs to end up in the final product."
      (Tim, 40:46 / 43:18)

• Prompting Advice:

  • If AI is not responding well, start a new thread or ask for a “resume work” prompt to clarify.
  • Be polite to the AI; prompts often go better.

Notable Quotes & Moments

• "Automate away toil. That's what we want to do."
  (Claire, 00:36 & 37:46, recurring theme)

• "Nobody wanted to do this work" (regarding manual data entry)
  (Summed up in title and throughout)

• "I think the best argument I have for all the work I've done...is that the same people who used to write this data were the ones responsible for doing the research. So you've now freed them up to just look more."
  (Tim, 22:01)

• "I have a button down here where...if I like an image I can click 'find similar' and it's just going to go and find every image that kind of has that vibe."
  (Tim, 22:51)

• "No one was going to make me this app. And so the ability to make like an extremely specific app that makes a workflow...easier, it's been an unbelievable moment."
  (Tim, 35:39)

Section-by-Section Timestamps

• Intro and Overview: 00:00–03:49
• Motivation & AI’s Role: 03:49–05:21
• Early Automation & Demos: 05:21–12:41
• Building Out the System: 12:41–19:10
• Embeddings & Search: 19:10–22:51
• Semantic Search & Discovery: 22:51–24:44
• Flip Flop App & Field Work: 24:44–31:35
• File Embeddings & Takeaways: 31:35–32:03
• OCR Party & Handwriting: 32:03–36:44
• Product Philosophy/Learning: 37:46–40:15
• Industry Concerns & Advice: 40:15–44:49
• Prompting Techniques & Wrap: 44:49–46:24
• Contact & Show Info: 46:24–end

Practical Workflow & Tips from the Episode

• AI + Manual Guardrails: Always fuse AI’s generative power with hard metadata or source context to boost trust and accuracy, especially for archival/historical work.
• API-first Automation: Even "non-engineers" can build powerful workflow automations via API, AI script-writing, and micro-apps.
• Use ‘Vibe Coding’: Design UIs and features by “speaking in screens” and partnering with AI coding assistants.
• Think Custom: If you have a repetitive task nobody wants to build a product for—build your own AI-powered micro tool!
• Embed Metadata Early: Automate embedding of all critical info into files at the moment of capture, not downstream.

Conclusion

Tim McLear provides a masterclass in “AI for real work”—demonstrating how filmmakers and creatives can use today’s tools not as replacements for human creativity, but as crucial force multipliers for the tedious, time-sapping, and universally dreaded parts of research and post-production. The episode is both practical and inspiring, emphasizing that anyone can leverage AI coding—even in highly specialized fields—to free up more time for meaningful, creative human work.

For more:

• Tim’s site: timmacular.com (features his own AI chatbot "GP Tim")
• How I AI Podcast and episode archive: howiaipod.com

Ideal Listener:
Anyone overwhelmed by digital assets, curious about practical AI, or itching to automate away tedious work—especially in creative or research-heavy industries.