# Sep 25, 14:13

## Session setup

![David Avalon working at his desk in his room](photos/photo-df5a81fe-399c-450e-92f1-fa0d299b7925.jpg)

![The desk setup: curved monitor, keyboards and drives](photos/photo-a78481fc-58ba-477b-92c9-bd6e76194e27.jpg)

- Working with David Avalon on video techniques, in his room
- Drive plugged in; screen mirrored
- Small Logitech Bluetooth mouse
- Monitor settings fixed
    - Samsung and LG monitors
- Drive is the Samsung, one terabyte
- Project files under the "Anthony" folder

## Reviewing the existing script

### Story file

- Story.md (or HTML) opened and read
- Cold open: frozen cursor over a deploy modal, silent
- Montage: "six days"
- Only partly the Story Sherpa version
- Product: a security product sold to customers

### Approved and production scripts

- Path: audio boards, concepts, script; "approved" MD
- Script review version two; the 23rd is the latest
- Production script sections
    - Style, music, voices, assets, scene one cold open
- Scene 16: end card; pan scan, full frame
- Missing end screen line
    - Roughly "They trusted you with it more than just their password"
- Script can change; judging what works is part of the process

## Source treatments

- Three treatments, not in this folder; Chipp to send them
- Sent around September 9 or earlier
- Made by ChatGPT from the website: "give me three treatments"
- File: security treatments for New Normal Security
- No Story Sherpa applied yet; good starting point

### Aside: new local models

- Chipp installed a new Qwen image model
- Ran tests with LTX 2.5 video model
- Maybe usable for this project

## First pass: Story Brand

### Prompt

- Run the three treatments through Story Brand
- Return one new treatment on Story Brand
- Website URL question: probably already in the source material

### Story Brand vs Story Sherpa

- Story Brand is part of Story Sherpa
- Story Brand: story structure placing the story in a hero format
- Went straight to Story Brand since preliminary work existed
- Story Sherpa would run a much more thorough analysis

### Story Brand output

- Developer is the hero; New Scan the guide
- Agent is the hero's weapon
- Framework rewrites the text; seven elements
    - Hero: a vibe coder who wants to launch
    - Villain: the old way of doing security
    - External: can't tell if the app is safe online
    - Internal: "Am I about to be the guy who leaked data?"
    - Philosophical: AI changed building; security shouldn't need different workflow
- Guide line: "You didn't build it the old way..."
- Authority: deterministic testing on every finding
- Great for branding products

## Full Story Sherpa run

### Prompt

- Run treatments through the complete Story Sherpa workflow
- Insight extraction engine first, then feed existing material
- Goal: 60-second AI-assisted cinematic online commercial
- Added New Normal Security website link, from text messages
    - Website has enough backstory for the product

### Aside: /next list and recording

- Chipp uses /next in all his repos
- Possible podcast or mulligan showing how this works
- Start screen recording: whole monitor, audio on
- Recording stored on Chipp's drive; he will break it down

## Recorded explanation

### The problem Chipp brought

- First thing to settle in a branded film: the story
- Chipp's friend runs a security company with a website
- ChatGPT made three treatments; can't do a great job alone
- David: award-winning film director and producer; also advertising
- Story Sherpa, new name "Box Box Story"
    - Holds 30–35 years of experience
    - Used at David's company
- Ask: analyze the three treatments

### Story Sherpa's parts

- Five pieces in total
- Insight extraction engine
    - Pulls insights from any document
    - Prompt built with research to learn a subject
    - For jumping into a new team's space
    - Research that already thinks in story
- Story Brand
    - Existing outside framework, borrowed
    - Sets up a hero and a guide
    - Product is never the hero; the hero starts dumb
    - Luke Skywalker needed Yoda and Obi-Wan; be the Yoda

## Story Brand explained

### Components

- Seven components; understands brand storytelling well
- Hero, villain, external, internal, philosophical problem, guide, plan, call to action
- Also failure, success and transformation
- "Algebra of storytelling"
    - Algebra is teachable and scalable
    - Art comes through trial and error

### Applied to this product

- Security product for vibe coders
- Hero: vibe coder who wants to launch
- Villain: old security
    - Dashboards, consultants, jargon; built for other coding
- External: can't tell if app is safe to publish
- Internal: fear of leaking users' data
    - Imposter feeling of shipping what he can't see
- Best ads break down into this algebra
- Guide: New Normal Security as Obi-Wan Kenobi
- Philosophical: AI changed how software is built; security workflow hasn't
- Guide shows empathy
    - "You didn't build it the old way..."
- Authority
    - Every finding retested and confirmed with evidence
    - False positives dropped, not dumped on you
- Plan in three steps
    - Hand it to your agent
    - New Scan finds what's real
    - Fix it and ship
- Vague "we're cool" commercials: eye candy only
- Calls to action
    - Direct: "Tell your agent: install New Scan"
    - Transitional placeholder: free first scan, docs list
- Failure
    - Another user's data returned, exposed endpoint, leaked secret
    - Security-incident email to your first 500 users
- Success: clicks deploy, users sign up, closes laptop, sleeps
    - Already builds visuals in David's head
- Transformation: from hoping it's secure to shipping with evidence
    - No transformation, no real story; needs conflict
    - "Travel video" trap: Belize, sun, beaches, no problem solved

## Value for newcomers

- Chipp: never made films or live-actor video ads
    - Has hired David for some of that before
- New AI videographers have no help like this
- ChatGPT doesn't reach this level of detail
- David knows nothing like it; if any, internal to agencies
    - Agency experts in meetings don't scale
- Goal: make the Box Box Story workflow available to the masses
- Works with any LLM
    - Claude, Copilot, Gemini, Grok

## Story Sherpa results: insight extraction

### Input

- Big picture: 60-second AI-assisted cinematic online commercial
- Fed Chipp's three ChatGPT treatments plus newnormalsecurity.com
- Plain natural-language request; no filmmaking experience needed

### Core truth

- Ad sells permission to ship, not a scanner
- Turns a builder's private doubt into a decision
- Lesson from Human Code
    - Lennox HVAC project: buyers rarely buy, every 25–30 years
    - Lennox really in the business of trust
- 30 years of design and industry knowledge fed in

### Problem being solved

- Agentic coding lets builders outrun their understanding
- Security still a separate discipline with separate tools
- That gap is where builders hesitate, or don't
- Real problem also defines what you're selling

### Hidden insights

- Villain isn't a hacker; it's the shrug
    - Agent answers "Are we secure?" with "I recommend a review"
    - Strongest beat across all three treatments
- Trust comes from the part that isn't AI
    - Agent is the hands; New Scan the instrument
    - Agent a pair of hands, not a second oracle
- Brand best expressed by subtraction
- Deploy button is a confessional booth
- Conversion ask is one prompt, not a purchase
- "Same hands" principle
    - Build and check with the agent; check isn't its opinion

### Assumptions

- Vibe coders feel doubt: many are overconfident
    - Ad must briefly create doubt before resolving it
- Viewer's agent runs an MCP tool; viewer runs Docker
    - True for target, but friction the ad shouldn't hide
- Audience needn't know BOLA or SSRF; plain-language stakes
- Website speaks to pen testers and security teams
    - Treatments speak to vibe coders; landing page partly mismatched
- Anthony doesn't want anything overly technical

### Agency comparison

- At Human Code: presentation one to two weeks later
- This took about five minutes

### Story layer

- Character: builder who outran their understanding
    - Talented, fast, alone at 2 a.m.
- Tension: speed vs knowing
- Resolution: same partner brings an instrument that tells truth
- Opening scene: ECU of cursor trembling over "Deploy to production"
    - Laptop fan, fridge hum, 2:40 a.m.
    - Sticky note on the bezel: "Ship it"
- Tone: intimate tech thriller resolving into dry comedy
    - Low-key practical light, shallow depth of field
    - Then warmer dawn light: relief
- Chipp would never have thought of cool-to-warm colors
- Subtext
    - Language between the lines; felt, not seen
    - Great films and commercials full of it

## Aside: mulligan idea

- Chipp: would make a good mulligan with screenshots and avatars
- Process is what's worth money
    - About $7,900 for one Human Code meeting

## Human Code background

- Interactive media powerhouse in Austin, early 90s
- Era of CD-ROMs and the early web
- Made the very first Apple CD-ROM
- Website possibly first with buttons that did actions
- David joined at about eight people, in a house off West Avenue
- Design focus from leadership down, not engineering
- Won many awards

## Story analyzer and rubric

### Story Brand scores

- Scored existing treatments out of 35
    - Treatment one: 25; treatment two: 22
- Red X missed, yellow warning, green nailed
- Treatment one missed call to action; warning on plan
- Treatment two missed most elements
- Hybrid treatment scored 32

### Algebra of storytelling

- Seven components are variables to fill in
- Not fuzzy; gives numbers to compare treatments
- Not just subjective artist intuition
- Great stories share components and beats
- Like UI design: don't reinvent the drop-down
- Learn the algebra, then bring your art

### Aside: analyzing existing ads

- Chipp: can an existing ad, e.g. Coca-Cola, run through?
- Yes; a WhisperX-type tool describes video frames
    - Works on key frames; like image-to-prompt
    - For all-visual ads with no voiceover
- Tool must be local; Story Sherpa asks to download it
- Tell it in the prompt to analyze the video

### Hybrid treatment

- Merged best parts of all three treatments
- Usually a recipe for disaster: too many cooks
- Scored higher across the board

## Four Words, Five Lines

### Four words

- Second framework; from Ted Sorensen, JFK's speechwriter
- Clarity: said the clearest way?
- Brevity: short, cuts to the chase
- Charity: empathy
- Levity: humor, relief from tension, the human side

### Five lines

- Headline: in ten words or fewer
    - Can't do it, you don't know what you're saying
    - Platitudes like "we go the extra mile" don't count
- Chipp's view: bottom line as sub-headline
    - Headline grabs attention in about 1.5 seconds
    - Sub-headline is the evidence for the claim
- Asked Story Sherpa to define bottom line
    - Last thing the audience feels and remembers
    - Five lines: outline, headline, frontline, bottom line
    - Frontline opens; bottom line closes
    - Two jobs: emotional close, credibility anchor
- Chipp's reading was wrong; not the subhead
- Story Sherpa also meant to teach storytelling

### Installing Story Sherpa

- Four prompts, as skills (MD files), in your own repo
- Fifth file: whole system and how parts interact
    - Starting point, like CLAUDE.md
- Four components
    - Story Brand, Four Words Five Lines, story rubric, emotional index

### Scores

- Four Words Five Lines out of 45
    - Treatments: 30, 35, 38; hybrid 41
- Brevity: X on treatment one; charity weak across all
- Levity got a 5 for the password line
    - "Customers trusted you with more than their password"
    - Best part of the script
- Headline nailed only in treatment three
- Totals out of 100
    - Treatments about 70–71; hybrid 91
    - Third treatment didn't pass

## Emotional index and rhetoric

- Video in a rectangle means entertainment, for any company
    - Otherwise make a white paper or podcast
- Show-me-don't-tell-me medium
- Chipp: 50,000 YouTube followers, few comments
    - New product videos: commercial first, then demo
- First part needs a hook before learning mode
- Four rhetorical appeals, asked from Story Sherpa
    - Pathos, 50%: emotion; hover, doubt, relief at click
    - Logos: proof; confirmed issue, false alarm dropped, fix, rescan
    - Ethos: credibility; "no tool can guarantee," nothing inflated
    - Kairos, 10%: timing; last second before launch
- Kairos treated as ignition; flagged first

## David's background

- Designers are natural storytellers; wanted to be a better one
- Human Code: Dell.com website, about 10,000 pages
    - Pre-dynamic pages led to thinking in patterns
- Interface design, information architecture, filmmaking merged
- Chipp recalls David using CD Factory in unexpected ways
- Started on an Atari with 6502 assembly
    - Atari enthusiasts group: ex-Schlumberger engineers and pilots
    - Youngest member; made him a hacker
- Story Sherpa as a tool: started a year ago, with LLMs
- Roots go back further
    - Macromedia days: repeating story elements
    - CD Factory: Human Code's no-code CD-ROM storytelling tool
    - Cheap DSLRs, under $1,000, opened camera work
    - Short film festivals; films shown at Alamo Drafthouse
    - Shooting and writing commercials
- Tipping point: Save the Cat, before IBM
    - Was at a startup bought by Logitech
- Muse storytelling framework when starting at IBM

## The new script

### Opening beats

- Character: Jordan, late twenties, hoodie, "the good kind of tired"
- Agent: generic unbranded terminal chat
- Frontline: ECU cursor trembling over modal
- Kairos: "Deploy to production"; rack focus to Jordan
    - Kairos means timing
- Excite: montage of Jordan typing to the agent
    - Tasks snap to checks; coffee cups multiply
    - Day counter one to nine
    - "Done": social login, payments, admin
- Disturb: hard cut to hover; music drops
    - UI peels back into routes and endpoints
    - Flashes: other user's data, open admin, plain-text API key
    - Heartbeat, slow

### Probing the doubt beat

- Chipp: film vernacular unclear; can ask Story Sherpa
    - Frontline, excite, disturb, the shrug
- Doubt runs 14 to 22 seconds; hinge of the spot
    - Disturb in excite–disturb–assure model
    - Goal: viewer thinks "Have I checked that?"
- Like a Human Code pitch: logic behind choices, no easy button
- Chipp: how would viewer know it's someone else's data?
    - As written, they wouldn't
    - Fix: "Logged in as Jordan" header with avatar
    - URL ticks user 1042 to 1043
    - Stranger appears: Maria Lopez
    - Needs 1.5–2 seconds; cut flashes from three to two
- Script adjusts interactively; optimizes for the score
- Chipp's idea: new user card slides over, red alarm
- Asked for three tweaks
    - Wrong face: header vs. Maria's photo, address, card
    - Email: customer complaint emails stack up
    - Every key fits: apartment hallway, key opens neighbor's door
- First concept chosen and written into script
- Running in Claude inside the Story Sherpa project

### Script first

- Chipp: getting the script right matters most
- On set, "fix it in post" could've been fixed in script
- Reshoots cost a crew day rate
    - $1,600 a day or more

### Rest of the script

- Shrug: "Are we secure?"
    - Agent: "I'd recommend a full security review"
    - Jordan: "Yeah. No kidding."
- Guide: "Install New Scan. Scan the app. Show me what's real."
    - Music re-enters: precise, mechanical, confident
- Plan: New Scan console cascades
    - 42 endpoints discovered
    - Confirmed: another user's data exposed
    - False positive dropped
- Human decides: evidence panel shows Maria Lopez record
    - Agent: "One confirmed, evidence attached. Want me to fix it?"
- Rescan: tests pass
- "We're secure?" Agent: "No tool can guarantee that"
    - "Everything real is fixed."
- Deploy click without hesitation; first blue of dawn
- End card: "Your agent built it. Now let it find what's real."
    - Chipp not sure he likes it

## End line rewrite

### Calling ChatGPT through Codex

- Chipp: Claude weak at headlines; ChatGPT better
    - Claude strong at coding and analysis
- Created a "question codex" skill in Claude Code
- Skill not visible to the other session
- Cloud chat can't run the skill; dropped
- In OMP, /model switches between ChatGPT and Fable

### Three variations

- "No guarantees. Just the truth about what you build."
- "Build fast. Know what's real. Ship it."
- "Ship like you know what's in it."
- Each ends "New Scan. Free."

### "Trust" line

- Chipp proposes: "Your customers trust you with more than their password"
- Wording fixes: "their" password; no question mark
- Story Sherpa assessment
    - Works: pays off the Maria beat; raises stakes to responsibility
    - Respectful of builders; no scare
    - Watch: reopens tension after the dawn release
    - Password slightly off target; spot shows name, address, card
- Option: "...Now you can honor that."
- Kept: "Your customers trust you with more than their password. New Scan. Free."
- Line originally came from ChatGPT
- Chipp to integrate ChatGPT; can't from the cloud

## Saving the script

- Output: Story Sherpa breakdown run plus production script
- Production script: DocX and plain text; no MD
    - Note: add an MD version
- MD version requested and generated
- Saved to new folder "Anthony new" for the whole video
- Opened in MD Docs, set as default
- Pause to grab water; recording stopped
