Call notes in.
Shooting docs out.
- Client
- Commercial production studio
- Role
- Designer and builder
- Stack
- Claude, Google Slides API, Python
- Status
- Used on every project
- 01
Call notes
Input - 02
Pitch deck copy
Claude skill - 03
Google Slides
Slides API - 04
Script approved
Director - 05
Pre-production deck
Claude + Python - 06
Call sheet
Spreadsheet
The machine moves the work.
People check it.
Each step hands a finished document to the next. Between them, people check the work: script approval, product details and brand guidelines, among others.
AI needs humans to perform well in the marketplace at the moment, so I find strategic check points for it. I know what it does well, and where it needs the human touch.
Pre-production took six weeks.
Every commercial moves through the same process. Call notes become a pitch deck. Won pitches become a script. An approved script becomes a pre-production deck. That deck becomes a call sheet for the shoot.
That is the norm on a typical project, before a single frame is shot.
Six skills. One pipeline.
Pitch deck copy
Turns discovery call notes and a brand’s website into finished pitch sections.
Pitch to Slides
Writes that copy straight into the studio’s Google Slides template through the API. No copying and pasting.
Pre-production deck builder
Takes an approved script and a reference deck, and builds a complete new pre-production deck. I later rebuilt it as a single-run Python script.
Call sheet generator
Converts the finished deck into a shooting schedule in the studio’s standard call sheet format.
Wizard agent
A browser agent that completes the studio’s five-step pre-production wizard: objectives, characters, scripts, storyboards and casting.
Style enforcer
One writing standard on every output. Short sentences, a banned-word list, and script conventions like frame numbers from 101 and INT./EXT. headers.
The AI drafts. People decide.
Automation never gets the final word. A person approves every script before the deck is built. People QC every product detail. People check every output against the brand guidelines.
When I automated everything, engagement on the final videos dropped. So I pivoted, and now humans stay in the loop at every key moment of creation. AI needs constant innovation, because clients’ needs change.
I flipped the order. The workflow worked.
The studio’s internal production platform generated a pre-production deck first, then the script. There was no room for real creative to form, and one change to the script meant starting over from scratch.
Because I understood how the models work, I proposed generating the script first. It fixed the whole workflow. The team went from constant regenerations to simple edits on the writing.
Packaged for a team.
I moved the system into Claude Enterprise: skills packaged, context documents exported, standards built in.
The team adopted the system first. Then the company used those findings to launch its own pre-production AI platform.
What changed.
Pre-production dropped from six weeks to under four, more than a third faster. That is two weeks or more back on every project.
It is not a pilot. Every project now runs through the pipeline, and the entire team uses it.
Also built: a brand strategy generator on Google Gemini. An intake form in, a full content strategy draft out in 15 to 30 minutes. It ran as a lead magnet and a working test of Gemini.
What I chose not to automate.
Fine-tuning the script. Reference images for pre-production. Cinematography and art direction. Anywhere a person could make the work more unique, I kept a person.
Automation took the repetitive work. The craft stayed human.