Sep 25, 14:13

A working session with David Avalon, in his room, on video techniques: using his Story Sherpa workflow to turn three rough treatments into a 60-second commercial for New Normal Security’s NewScan product.

David Avalon at his desk in his room, leaning in to work at the keyboard

The setup: David's ultrawide monitor, two keyboards and the drives on his desk

Setup

  • Chipp plugged his drive into David’s machine, and David mirrored it to his screen. David fixed his monitor settings first (“Mirror 4”, “Optimize 4”): one screen is a Samsung, the other an LG, and they look a little different.
  • Chipp asked about David’s mouse. It’s Bluetooth, a little Logitech one, and David likes it.
  • The drive isn’t called “Video Library”; it’s the Samsung, one terabyte. The project lives under the Anthony folder.

The Existing Material

The old story file

In the Anthony folder we opened the story file (Story.md, or the HTML version; either one). David asked whether it had already been through Story Sherpa, or whether it was an older version. It has elements of a story:

  • A cold open: the frozen cursor rests on a modal, dead silent, no one talking yet, an on-screen “deploy”.
  • A montage: six days.

Chipp: “Yes and no. That’s the story that I have. We can change it.” Part of the process is looking at it and saying, that kind of works, it kind of doesn’t.

The product is a security product that Anthony sells to customers. The background of the story is a guy in an office. There’s also a “finish” section noting that clips land at 720 and get upscaled.

The production script

Chipp had modified the script since then, so we went into the audio boards, concepts and script folders and looked at approved.md, then the script reviews (version two, dated; the 23rd is the latest), then the production script. It has sections for style, music, voices and assets, then scene one (the cold open) all the way down to scene 16, the end card. The end card’s intent: make the last thing read the same path as the problem. It’s pan-and-scan, full frame.

The end card is supposed to have a screen that says something like “They trusted you with more than just their password.” It doesn’t, but that doesn’t really matter, because whatever we end up with, we end up with.

The three treatments

The starting point was three treatments Chipp made earlier. They weren’t in this folder; he had sent them to David, probably around September 9th or earlier, and we found them in a folder called “Security treatments” for New Normal Security. That’s where we start.

Chipp made them by giving ChatGPT the website and asking, “Give me three treatments.” So no Story Sherpa had been applied at all, which makes it a good place to start.

Side note: new models

Chipp had just installed the new Qwen image model and run tests with the LTX 2.5 video model, and it’s working. It might be a good one to use for this project.

First Pass: Story Brand Only

David opened Story Sherpa in his projects. Because Chipp already had the core of what we needed, David cut to the chase and went straight to Story Brand, which lives inside Story Sherpa. His prompt:

Can you take a look at this script? There’s three treatments in here, but we want to run the treatments through Story Brand and give me a new treatment. Take a look at the three and just give me one new treatment, but make sure that it is on Story Brand.

Chipp asked whether to give it the website URL too. It could, but the treatments had already been generated from that URL, so it wasn’t an issue. With three treatments already in hand, Story Sherpa just takes them and gets them into Story Sherpa shape. Starting from scratch would mean running the whole Story Sherpa process. It produces one treatment, picking the best parts of each.

What Story Brand is

Story Brand is part of Story Sherpa. It’s the story structure that puts a story into a hero format: hero, guide, plan. It’s a framework, and it always puts the story that way. It doesn’t just reformat the text; it rewrites it.

The main Story Brand fix across the originals: the developer is the hero, NewScan is the guide, and the agent is the hero’s weapon.

Story Brand is very much about brand. It’s great for branding products, which is really the way we sell products today.

What Story Sherpa adds is a much more thorough analysis. In the end it may come out the same, but for the sake of it, David ran the full Story Sherpa to see what’s different.

Recording the Session

We decided to record the screen as well as the talk.

  • Chipp keeps a /next to-do list in all his repos (“You call them projects, I call them repos”). This session could go on it as either a podcast or a Mulligan.
  • David had room to record (“I got 20”), and it could also go on Chipp’s little drive. Chipp can convert it and break it down later.
  • We recorded the whole monitor, since we’d be popping back and forth, with audio on.

Why Chipp Came to David

Chipp asked David to explain what we’d done so far.

David: in creating a cinematic short film, commercial film or branded film, the first thing you want to settle is the story. Chipp already has a really good story; it just isn’t on the Story Sherpa or Story Brand framework.

Chipp: “I have a friend with a website. It’s a security company. I asked ChatGPT to create three different treatments, and ChatGPT did its best. But I know ChatGPT can’t do a great job of creating a video, and I know there are a lot of steps.”

David is an award-winning director and producer of films, and he works in the advertising space as well. His tool, Story Sherpa, now has a new name: Box Box Story. It’s a collection of 30 to 35 years of experience, and at his company everybody uses parts of it. Chipp wanted him to take the three treatments and do an analysis.

How Story Sherpa Works

Story Sherpa consists of five pieces.

1. The insight extraction engine

It can take any document and pull insights from it, so it understands the space it’s in. The prompt was built from research and asks a set of questions to understand the subject.

David jumps onto different projects for different teams in completely different spaces. The first thing he always had to do was ingest all the information so he could jump into a new space. Insight extraction does that, but from a story-development framework: it does a natural kind of research and extraction, but it already thinks in story.

2. Story Brand

A framework that exists out in the world; David borrowed it. It’s based on setting up a hero and a guide. Your product or service never wants to be the hero, because the hero is kind of dumb in the beginning. Look at Luke Skywalker: he didn’t know anything. He needed Yoda and Obi-Wan to bring the wisdom. You want to be the Yoda; your product and service want to be the Yoda.

Story Brand has seven components and understands brand storytelling really well. There’s a hero, a villain, an external conflict and an internal conflict. The external is the threats out there in the world. The internal problem is the one the hero has to solve, just as Luke Skywalker has an internal problem to solve. It looks at a story that way because good storytelling is really character storytelling.

The full set: a hero, a villain, an external problem, an internal problem, a philosophical problem, a guide, a plan and a call to action. It also highlights failure, success and transformation. All these components are in good brand storytelling.

3. Four Words, Five Lines

Covered below.

4. The story rubric

Covered below.

5. The emotional index

Covered below.

The algebra of storytelling

David calls this the algebra of storytelling. “I can teach the algebra, and that is scalable. I can’t teach the art.” The art comes through trial and error, and people get better at it, but the algebra is there. Story Brand’s algebra is hero, villain, external problem, internal problem, philosophical problem, guide and so on.

What Story Brand Found for NewScan

The product is a security product for vibe coders.

  • Hero: a vibe coder who wants one thing: to launch.
  • Villain: the old way of doing security. A separate world of dashboards, consultants and jargon, built for a way of coding he doesn’t use.
  • External problem: he can’t tell whether the app is safe to put on the internet.
  • Internal problem: “Am I about to be the guy who leaked his users’ data?” The imposter feeling of shipping something he can’t fully see.
  • Philosophical problem: if AI changed how software gets built, security shouldn’t still require a different workflow.
  • Guide: New Normal Security is the Obi-Wan Kenobi of the story.
    • Empathy: “You didn’t build it the old way. You shouldn’t have to secure it the old way.”
    • Authority: deterministic testing, with every finding retested and confirmed with evidence. False positives get dropped, not dumped on you.
  • Plan:
    1. Hand it to your agent.
    2. NewScan finds what’s real.
    3. Fix it and ship.
  • Direct call to action: tell your agent, “Install NewScan.”
  • Transitional call to action (placeholder): free first scan, docs, list or similar.
  • Failure: another user’s data returned, an exposed endpoint, a leaked secret, the “We had a security incident” email to your first 500 users.
  • Success: he clicks deploy without hesitating, users sign up, and he closes the laptop and sleeps.
  • Transformation: from hoping it’s secure to shipping with evidence.

David’s commentary

  • You can see it already setting up a story. If you broke down some of the best ads out there, you’d find this underneath. That’s where the insights come from: why does this resonate? Here’s the underlying algebra of good storytelling.
  • It lays out three perspectives on the problem (external, internal, philosophical) and then asks how we solve it. We’ve got a guide, and New Normal comes in with empathy and then authority. A great framing for how the story can be told.
  • A good guide gives you wisdom and a one-two-three plan. A good commercial doesn’t give you a vague “We’re cool, you can be cool too.” You watch that and say, it was eye candy, but that’s it. It won’t win any awards. Good work gives you a plan and a direct call to action, and tells you what you can physically do.
  • The success beat already builds visuals in David’s head: someone closing a laptop, going to sleep, resting. There’s a whole image there.
  • You don’t have a real story without transformation. That’s why you need conflict, a villain and an external problem, and why so many boxes are spent on the problem. When people show off products, they often make what David calls a travel video: “Come to Belize, because we have sun and beaches and little umbrellas on our drinks.” Is that cool? Yeah, but you didn’t tell me anything. You didn’t give me a problem you’re fixing. That’s why you spend a lot of time on the problem, because then transformation happens.
  • “From hoping it’s secure to shipping with evidence” is a real transformation, not a vague “Try our product.”

Why This Tool Matters

Chipp: “I’m a guy who hasn’t made 50 films. I’ve never done a video advertisement with live actors. I’ve hired you to help me with some of that in the past.” People just getting their toes wet in videography, especially AI videography, have none of this help. You can ask ChatGPT, and Chipp has, but it doesn’t get to this level of detail. This is the first time he’s seen anything like it.

David doesn’t know of anything like it. If it exists, it’s internal only, at an agency, where they bring someone like David into a meeting to lay it out for you. That isn’t scalable.

The goal for Box Box Story is to make this workflow available to the masses: “Get your Box Box Story workflow.” It works with your existing AI: Claude, Copilot, Gemini, Grok, any LLM you want.

The Full Story Sherpa Run

The prompt

David fed it two things: Chipp’s three treatments (made in ChatGPT) and the website, newnormalsecurity.com. The prompt:

Let’s go ahead and run these treatments through the complete Story Sherpa process workflow, because we need the insight extraction engine first. That will now feed into what we already have. Using the insight extraction along with the material that’s already there, we can start to determine what’s the best story to create, because in the end we’re wanting to create a 60-second online commercial. Obviously this commercial is going to be an AI-assisted cinematic commercial.

I’ve also put the link there for New Normal Security. Take a look at the website. There’s plenty of information there to give any backstory that you need for the product that we are highlighting in this commercial.

The target: a 60-second, AI-assisted, cinematic online commercial.

It’s a very natural-language request. You don’t need years of filmmaking experience to say “Here’s what I want, give me a great commercial,” and let the Story Sherpa brain do what it does best.

Insight extraction: the core truth

The first thing it does, and what David loves: “Here’s the core truth. We’ve got a real thesis. The ad isn’t selling a scanner. It’s selling permission to ship. The product is the thing that turns a builder’s private doubt into a decision.”

David learned this at Human Code. In a meeting about a big project for Lennox heating and air conditioning, they asked what Lennox was really in the business of. They sold HVAC systems, but how many HVACs do you want to buy? Hopefully one in 25 or 30 years. They weren’t saying “Upgrade your HVAC every year”; that would be a crappy HVAC. They decided Lennox was in the business of trust. Same thing here: get to the real truth of what the product is really in the business of. That knowledge from 30 years of working in design and industry is fed into this.

Why the treatment exists

Agentic coding has let builders outrun their own understanding of what they build. Security still feels like a separate discipline with separate tools, and that gap is where they hesitate, or worse, don’t. You don’t really have transformation until you know the real problem, and finding it also helps decide what you’re really selling.

Hidden insights

  • The villain isn’t a hacker; it’s the shrug. The strongest beat across all three treatments is the coding agent answering “Are we secure?” with “I recommend a comprehensive security review.” That non-answer is the real antagonist: the moment the trusted build partner goes vague.
  • Trust comes from the part that isn’t AI. The positioning says it plainly: deterministic tooling that agents operate. The agent is the hands; NewScan is the instrument. The ad should make the instrument feel solid and the agent feel like a pair of hands, not a second oracle.
  • The brand is best expressed by subtraction.
  • The deploy button is a confessional booth. (David: “I like that one.”)
  • The conversion ask is one prompt, not a purchase.

These insights are story-driven. It isn’t just information; it’s information applied.

Mental model

The “same hands” principle: the tool that creates should be able to verify, through a separate, trustworthy instrument. Build with the agent, check with the agent, but the check isn’t the agent’s opinion.

(Chipp broke off laughing here. He couldn’t see the text David was reading, then found it right in front of him.)

Assumptions

  • That vibe coders feel doubt at deploy. Many don’t; they’re overconfident. The ad must briefly create the doubt before resolving it. It takes an adversarial approach, and it helps you understand the potential customer.
  • That the viewer’s agent can run an MCP tool and the viewer can run Docker. True for the target, but it’s friction, and the ad shouldn’t pretend it isn’t there.
  • That the audience doesn’t need to know what BOLA or SSRF means. Plain-language stakes (“another user’s data”) must carry it, unsaid.
  • That the landing experience will meet a vibe coder where the ad leaves them. The website speaks mostly to pen testers and security teams; the treatments speak to vibe coders. Right now it partially doesn’t.

Chipp: the website isn’t focused on vibe coders. When he talked to Anthony, Chipp said he didn’t want something overly technical. If you target technical people, they want the specifics, and anyone else looking at it goes “What the hell?” A vibe coder kind of gets it.

It also covers implications and context; we didn’t go through all of it.

What this would have been worth

David: if he’d had this tool at Human Code, he’d have been the big dog. A client would come in, and they’d come back one or two weeks later with the big presentation. This took five minutes. Then you put graphics on top, but all the insight is there to say, “We’re ready to put your 60-second spot together.” It’s the equivalent of that meeting. It would have been worth not $79 but something like $7,900 for that one meeting alone.

From insight to story

Even in the insight stage it’s already thinking in story, and now it translates that. “Thanks for all the insight, but when am I going to get my 60-second commercial?”

  • Character: a builder who has outrun their own understanding. Talented, fast, alone at 2 a.m.
  • Tension: speed versus knowing. “I made this. Do I actually know what it does?”
  • Resolution: the same partner that helped build it now brings in an instrument that tells the truth.
  • Cinematic translation, opening scene: ECU of a cursor trembling over “Deploy to production.” Near silence: laptop fan, a fridge hum. 2:40 a.m. A sticky note on the bezel: “Ship it.”
  • Key visual metaphor: it translates all of that into something very visual.
  • Tone and genre: an intimate tech thriller that resolves into dry comedy. Low-key, practically lit, shallow depth of field, then warmer dawn light: relief.

Chipp would never have thought of moving the colors from cool to warm through the piece, from a dark night to the dawn light. David: that’s subtext, and it’s huge. Great films and great commercials have subtext all over them. It’s the language underneath, between the lines. Most people don’t see it, but they know it because they feel it.

“Speaking of water balloons,” as Chipp’s wife says when changing subjects: Chipp thinks this could be a really good Mulligan, with screenshots and a couple of avatars. David agreed.

Human Code

Human Code Interactive Media was an interactive media powerhouse in Austin in the early '90s, when CD-ROMs and then the web were breaking out: new technology, the same way we’re in an AI revolution now. Human Code was at the forefront of that high-tech evolution and revolution.

  • They did the very first Apple CD-ROM, and Apple was the first to do CD-ROMs.
  • Chipp thinks their website was the first on the planet with buttons that actually did actions.
  • It had some of the most creative minds working. David got there when it was about eight people in a house off West Avenue, and then it grew.
  • Everyone from the leadership down had a real design focus, not an engineering or technology focus, and that helped all the products. They won tons of awards, and that was the proof.

Scoring the Treatments

The insight is done from a story-brain standpoint, with an eye (or an ear) for story, not as pure research. Next comes the story analyzer.

Story Brand rubric (out of 35)

It ran the existing treatments through character, problem, guide, plan, call to action and failure:

Treatment Score (out of 35) Notes
Treatment one 25 Totally missed the call to action; a warning on the plan
Treatment two 22 Missed character, problem, guide, plan, call to action and failure
Hybrid 32 Much higher across the board

The rubric shows a red X for a complete miss, a yellow warning sign when you’re not doing well, and green when you figured it out. The best original, at 25 out of 35, is about two-thirds of the way.

What the algebra means

Just as Story Brand has seven components, those are variables to fill in, and how you fill them in drives the score. Chipp: so it isn’t warm and fuzzy; it’s deterministic. It rates things and gives you numbers, so you can compare them, rather than “this seems to work pretty well and this one doesn’t.”

David: it isn’t just subjective artist intuition. There’s a reason great stories are great stories: they have components and beats, and there’s a lot of evidence for that. So why not start with best practices? It’s like UI development. If you want a good drop-down, don’t start from scratch; learn from years of UI work about what makes a great drop-down and what functions it needs. You could invent a new drop-down, but you don’t want to ignore all the knowledge already there. The same goes for story development. This is the science of storytelling: don’t ignore it. Good storytelling comes from understanding the algebra. Then go off and bring your art to it.

The hybrid

The hybrid took all three treatments, tried to find the best parts of each, and merged them. Normally that’s a recipe for disaster: too many cooks in the kitchen. Chipp was curious to see how it turned out.

Side Question: Analyzing an Existing Ad

Chipp: can I run an existing ad, say a Coca-Cola ad, through Story Sherpa? David: absolutely.

There are tools from OpenAI; David thinks WhisperX has one. If a commercial is all visual with no voiceover, how do you judge it? The tool is the equivalent of transcribing audio, but for video: it grabs every so many frames and shows you what’s in each one, like image-to-prompt. It reads every frame, or at least every key frame, to work out what’s on screen. If there’s a voiceover, Whisper transcribes it and it understands the context.

  • WhisperX isn’t built into Story Sherpa; Story Sherpa will ask to download it for you. You need it local.
  • You have to tell it in the prompt: “I need you to analyze this video. Go off and analyze this video.”

Four Words, Five Lines

This is the second framework, from Ted Sorensen, JFK’s speechwriter. He was known for cutting to the chase; JFK was renowned for great speeches, and the speechwriter deserves a lot of the credit.

The four words

Story Sherpa understands these and applies them when it writes a script:

  • Clarity: is this said in the clearest way?
  • Brevity: is it short, is it sweet, does it cut to the chase?
  • Charity: is it empathetic?
  • Levity: humor and relief. Where there’s a lot of tension, you get a point of relief. It’s also the human side, bringing your own words to the table.

The five lines

  • Headline: you should be able to write a headline for what this is about in ten words or less. If you can’t, you probably don’t know what you’re trying to say. Chipp asked whether platitudes like “We go the extra mile” count, since anyone can say that. They don’t; it’ll catch that, because the headline has to resonate with the story. It takes the whole story into account and sums it up tightly, like newspaper headlines back when newspapers had that power.
  • Bottom line, front line and side line follow.

What “bottom line” means

Chipp’s understanding from web design: a headline grabs your attention (you have about a second and a half to keep someone from clicking off the page), and the subheadline is the evidence for the headline, like evidence for a claim in a courtroom. He asked whether that’s the bottom line.

David asked the LLM to define it as if for a client. Its answer:

The bottom line is the last thing we want the audience to feel and remember when the spot ends. It’s the note we leave them on, right before we ask them to do anything. Most people hear “bottom line” and think of a summary of the business takeaway. In Story Sherpa it’s more specific. It comes from the Five Lines framework: outline, headline, frontline, bottom line. The front line is how you open, and the bottom line is how you close. In our framework the bottom line has two separate jobs, and we never combine them into one line:

  • The emotional close: the moment the audience feels the hero’s win. It’s usually a picture or action rather than words.
  • The credibility anchor: the short, factual reason to believe the guide, the brand, delivered that win.

So Chipp’s description was wrong: it isn’t the subhead, it’s something different.

Story Sherpa is also meant to educate. Someone could install it only to learn to tell better stories and sit with it all day: “Teach me how to tell better stories. What is Story Brand? What is this? What is that?”

Four Words, Five Lines rubric (out of 45)

Treatment Score (out of 45)
Treatment one 30
Treatment two 35
Treatment three 38
Hybrid 41
  • The third principle got a warning; the first treatment got an X on brevity. Brevity and charity didn’t really hit across the three treatments.
  • Levity scored a 5 on the treatment with the best line: the funny line about customers trusting you with more than their password. It’s the best part of the script because it gets your attention, and the rubric picked up on that. David loves that about the science of storytelling: it knew that line delivered.
  • The headlines didn’t really nail it, except on the third treatment.

On the overall score out of 100, the third treatment didn’t pass at all. The first two kind of did; the treatments weren’t bad, and around 70 or 71 feels about right. The new treatment went up to 91.

The Emotional Index and Rhetoric

The minute you decide to tell a story in a rectangle (video), you’ve entered the world of entertainment whether you want to or not. David has told teams this for years. It doesn’t matter what company you are, even a medical company doing something serious: you have to play the entertainment game. Otherwise make it a white paper or a podcast. Put it in a rectangle, and welcome to the world of entertainment. There are rules for that, with an emotional side: you want people to feel it. Video is a “show me, don’t tell me” medium first. Otherwise, why make it a video if it’s just text or a talking head?

Chipp has 50,000 YouTube followers and very few comments. David: a demo is different; “I’m going to show you how to use this software” isn’t a commercial. Chipp would like to do more commercials. When he launches a product, the first part of the video is a commercial and the end is a demo. David: that first part needs a good hook. Once viewers know they’re entering learning mode, their brain shifts, but the hook has to play the entertainment game.

David asked Story Sherpa to list the rhetorical principles with simple definitions. There are four rhetorical appeals, in order of weight:

  • Pathos (50%), emotion: makes the audience feel something so they care. In the spot: the hover over deploy, the doubt, and the relief when he finally clicks.
  • Logos, logic: gives the audience a clear reason, proof or evidence. In the spot: the scan screen showing a confirmed issue, a false alarm dropped, the fix and the rescan.
  • Ethos, credibility: earns trust by showing the messenger is honest and knows what it’s doing. In the spot: “No tool can guarantee that,” every finding retested, nothing inflated.
  • Kairos, timing: shows why this matters right now and why this is the moment to act. In the spot: the last second before launch, set against the broader shift to agentic coding. Kairos is weighted at only 10%, but it’s treated as the ignition, the moment that makes the story start, so it’s always looked for and flagged first.

All of this is the algebra, the science of storytelling, and that’s what David loves about it.

How David Built Story Sherpa

Where it came from

David came to it as a designer, and designers are natural storytellers; he just wanted to be a better one. At Human Code, where so many great things happened for him because there was so much creativity, he worked on the Dell.com website: 10,000 pages. Pages weren’t yet fully built on the fly; they built pages from templates. That got him thinking in patterns, and information architecture and design started to gel. Then he got into filmmaking, which nailed it. Interface design, information architecture, filmmaking and storytelling smashed together, and that’s how Story Sherpa came about. Chipp added: how to make things simple, using macros and tools you build, and process thinking.

Chipp remembered David as one of the first people using CD Factory. Chipp looked at what David made and thought, “Wait a minute, CD Factory doesn’t do that.” David: “Oh yeah, it does. Let me show you how we turned this thing upside down.”

The Atari starter kit

David’s origin story: he started on an Atari with 6502 assembly language (it was BASIC or assembly). Owning an Atari meant being scrappy; you had to hack everything. At the Atari enthusiasts’ group there were a bunch of ex-Schlumberger engineers and pilots. He was the youngest kid there, and they took him in (“Hey, got a youngster in here”) and dumped all this cool stuff on him. There was no software yet; you had to code it yourself and figure it out. That’s how the early process started.

How long it took

Chipp asked how long it took to build and how many years of knowledge are in it. Story Sherpa as a tool started about a year ago, because LLMs became everywhere and that helped codify it and build the connections even stronger. But the story work goes way back:

  • In the Macromedia days he was already working on story. Doing interactive work, some elements kept repeating.
  • CD Factory worked the same way: what are the components of a CD Factory piece? This component, that component.
  • CD Factory, for people who don’t know, was Human Code’s own program for creating CDs, back when there were few authoring apps and you couldn’t do it unless you could write code. It was one of the first no-code CD-ROM builders that could actually tell a story. Others could make a slideshow; that wasn’t what CD Factory was.
  • CD-ROMs led to interactive story projects; then the web came along with information architecture.
  • Cameras got cheaper. The DSLR gave him filmmaker quality from a camera costing, give or take, under a thousand dollars, which opened the floodgates for camera work and telling stories.
  • Film festivals: a short film festival offered to show his films at the Alamo Drafthouse. He’d have paid the entry fee just to see his film there, and seeing it on the big screen was so awesome. That taught him to tell short stories in front of an audience.
  • Then doing, shooting and writing commercials.

The tipping point: David was at a startup that Logitech bought, before IBM, and was already making a lot of films. He read Save the Cat, which says there are really only seven stories in the world. That really kicked it off. When he started at IBM, the Muse storytelling framework came out, and that, along with Save the Cat, is how it evolved.

Installing Story Sherpa

It’s four prompts that go into your LLM; you put them in your own repo. They’re four different skills, four MD documents:

  1. Story Brand
  2. Four Words, Five Lines
  3. The story rubric
  4. The emotional index

There’s really a fifth: the one you start with, which describes the whole system and how the parts work together, like a CLAUDE.md. Understanding how they work together is important.

The New Script (Version A)

  • Character name, for production continuity (never spoken): Jordan. Late twenties, hoodie, the good kind of tired.
  • The coding agent: a generic, unbranded terminal chat-pane agent.

Front line

ECU: cursor trembling over a modal, with sound. Kairos: “Deploy to production.” Rack focus to Jordan’s face in the screen glow. Sticky note on the bezel.

“Kairos” is the word for timing.

Excite

Whip-rewind montage. Jordan types to the agent; each task snaps to a check. Coffee cups multiply. On the whiteboard a day counter ticks from day one to day nine. Driving synth. Jordan whispers: “Done.” Social login, done. Hook up payments, done. Admin, done. A very visual script.

Disturb: the doubt

Hard cut back to the hover. The music drops out. Jordan’s eyes drift across tabs: auth, payments, database. Two-second insert: the polished UI peels back into a tangle of routes and endpoints. Three quick, quiet UI flashes:

  1. One user’s profile loading someone else’s data.
  2. A forgotten admin route answering “200 OK.”
  3. An API key sitting in plain text.

He glances at the sticky note. Room tone. Heartbeat, slow: boom, boom, boom.

Chipp’s reading: he’s realizing there’s a problem with his program, somebody’s API key is out in the open, and he’s worried.

Asking Story Sherpa about “the doubt”

Chipp: there’s a lot of film vocabulary here he didn’t understand (front line, excite, disturb, the shrug). They mean something to David, but Chipp can always ask Story Sherpa what they mean. David asked it to elaborate on the Disturb section, the doubt:

The doubt runs from 0:14 to 0:22. It’s eight seconds, and it’s the hinge of the whole spot: everything before it is momentum and everything after it is resolution. This is where the audience has to stop and feel the question for themselves. In the Excite–Disturb–Assure model, this is the Disturb. It has one job: turn a private, unspoken worry into something the audience can see. We aren’t trying to scare anyone. We’re trying to make every builder watching think, “Wait, have I checked that?” That recognition is what makes the product feel necessary rather than pitched.

Beat by beat: hard cut back to the hover. The montage music dies in one frame. That sudden silence is the loudest moment in the spot. We’re back to the deploy button, but now the audience knows how much work is riding on it.

David: this is like being back in a room at Human Code two weeks later, pitching the client and explaining why we did what we did. They came to Human Code paying the big bucks. Why? Because we had the science and the reasoning behind it. It’s not David or Chipp in the other room sipping iced tea and twiddling their thumbs. They thought about it deeply, and that thinking is there. We didn’t hit an easy button; we put all this knowledge into the system.

Fixing the “someone else’s data” flash

Chipp asked: how would the viewer know the profile is loading someone else’s data? Story Sherpa’s answer:

As written, they probably wouldn’t. A profile page flashing by in under a second looks just like another profile page. That flash needs a visible contradiction the eye can catch instantly. The clearest fix is to show a mismatch between who’s logged in and whose data appears:

  • The header says who it is: the top corner reads “Logged in as Jordan,” with his avatar.
  • The URL changes by one digit: user 1042 becomes user 1043. You can animate the number ticking over.
  • A stranger appears: a different face, a different name (Maria Lopez), her email.

The eye gets pulled to the conflict. A non-technical viewer reads it as “he’s seeing someone else’s private info,” and a developer reads it as an access-control bug, but both get it. The trade-off is time: that read needs about 1.5 to 2 seconds, not 8 to 10 frames. I’d cut the flashes from three to two.

Chipp: what’s cool is that as you try to understand what’s going on, it adjusts the script on the fly. This is a proposal: two beats instead of three, so the fact that another user’s data is showing really lands.

Chipp suggested making it more explicit: a new user picture slowly sliding over the existing one, with a little red alarm flashing above it saying “Who is this?” David: it looks at the spot as a whole and always optimizes for the score. It will tell you if you just lowered your score or raised it (“That was a great move”), so it keeps that in check. Chipp mainly wanted to be sure that beat hits, which he wasn’t sure it did yet.

David asked for three possible tweaks before putting anything in. (We were talking to Claude, but Claude running inside the Story Sherpa project, so it already has everything.)

  1. The wrong face (the audience sees the bug happen). Hard cut to the hover; the music dies. His eyes drift to a test tab. The header reads “Logged in as Jordan” with his avatar. Two-second beat: the URL ticks over, and his profile snaps to a stranger, Maria Lopez: her photo, home address, and a card ending in 4417. A slow push-in lands on “Logged in as Jordan” sitting right above Maria’s face. Quick second flash: an API key sitting in plain text. He glances at the sticky note. Why it hits: it’s accurate, and developers will recognize the exact bug. Non-technical viewers need the header-versus-face contrast to be unmistakable, so it depends on sharp UI design. Maria’s record comes back later in the evidence panel.
  2. The email (plain human language: the audience sees the consequence in a user’s own words). Hard cut to the hover; the music dies. His eyes drift to his inbox tab, and he imagines it a week from now: a single new email slides in from a customer. The subject line reads, “Why can I see someone else’s account?” The preview says, “I can see another customer’s name and address.” Two more emails stack in behind it.
  3. Every key fits. Hard cut to the hover, then a match cut from his screen to a long apartment hallway lit by the same screen glow: his app imagined as a building. A resident unlocks her own door, but her key also turns in the neighbor’s lock. The door swings open onto someone else’s life: family photos, and so on. It goes a little deeper.

Chipp liked the first one best, and David agreed: the first concept wins. David asked it to fold that into the script and give the whole script back.

Why the script matters most

David: it gave a full understanding, gave the rubric as the reason for each change, and then let us break the script down, and the whole time we can interact with it and tweak at will. That’s why you want to know the story first.

Chipp: the most important part of all the work on a video, including staging, storyboards and everything else, is getting the script right, so spend time here.

David: that’s true in real filmmaking and commercial making too. On set he’d think, “Crap, it’s ‘fix it in post.’ This could have been fixed in the freaking script.” When shooting someone else’s script, you’re on set asking, “Can we shoot it this way?” because you saw a hole in it: we don’t see the email, or we need to see that on the line, because how else will the audience get it? The problem is when you find the hole after you’ve shot it. “Can we go back and shoot it?” Sure, if you want to spend another $1,600 day rate, or $2,000 to $3,000 a day depending on the crew, just to shoot that little part, and get the sound in.

The Revised Script, Continued

It came back as plain text, and we lost the nice formatted version from the Sherpa run. But the new flash is in: in the Disturb section Jordan’s profile snaps to Maria, with “Logged in as Jordan” sitting directly above Maria’s face.

The shrug

Jordan types. Agent replies. Close on Jordan’s face.

  • Jordan (typing): “Are we secure?”
  • Agent: “I’d recommend a full security review before deploying.”
  • Jordan: “Yeah. No kidding.”

Chipp: that reinforces what just happened.

The guide

Jordan types the line that changes the film. Enter.

  • Jordan (typing): “Install NewScan. Scan the app. Show me what’s real.”

The music re-enters: precise, mechanical, confident.

The plan

The agent drives. The NewScan console cascades:

  • 42 endpoints discovered.
  • Testing auth on user ID.
  • Candidate → retest → confirmed: another user’s data exposed.
  • Candidate → retest → false positive → dropped. The line dissolves off screen.

Sound: clean mechanical ticks, with a soft release tone on the drop.

The human decides

Evidence panel: the request, the response, and Maria Lopez’s record, the same one from the doubt beat. Jordan leans in, recognizes it, reads, nods.

  • Agent: “One confirmed. Evidence attached. Want me to fix it?”
  • Jordan: “Fix it.”

The rescan

Diff. Replays. Test passes. Confirmed.

  • Jordan: “So now we’re secure?”
  • Agent: “No tool can guarantee that. But everything real is fixed.”
  • Jordan: “There.”

Back to deploy

Back to “Deploy to production.” This time he doesn’t hesitate. Click. The first blue of dawn through the window. He leans back and exhales. The sticky note is in frame.

End card (as generated)

Your agent built it. Now let it find what’s real. NewScan. Free.

Chipp wasn’t sure he liked that, so we asked for three variations on that last line, the bottom line.

Detour: Getting ChatGPT to Write the End Line

Chipp wanted ChatGPT for this, because Claude is terrible at headlines and end lines, and ChatGPT is way better; Chipp and Eric have probably proven that. Claude kills on the coding side and a lot of other things, but not on writing like this. David has run Story Sherpa through Codex too; he uses both and jumps back and forth out of habit. The analysis is probably fine on Claude, but the writing is probably better in ChatGPT, and it’s faster in some cases.

What we tried:

  • David has Codex, logged in. Codex can reach ChatGPT, but you can’t give it the context; Claude can.
  • In a new Claude Code session (Chipp uses the CLI), David asked Claude to create a new skill called question codex: connect to Codex, send it a prompt, have ChatGPT answer the question, and return the answer. The good news is that once you have a skill, you just type the skill name and say, “Please ask…”
  • Claude created the skill, but the other session couldn’t see it. We told it the skill was available to all his projects and pasted the location. It offered three ways to make it work (attach the SKILL.md to the chat, copy it somewhere it could reach, and so on) and finally said the skill is a set of instructions it can’t run, because that chat runs in the cloud.
  • We dropped it. David will figure out how to make it work later (“Open the folder in Finder for the skill so I can copy it into another project”). Chipp didn’t want to spend much time on it.

The plan: when we get to this part, stop the recording and start a new one. Before we finish, generate the final script as a separate MD.

The three variations

The first try rewrote whole versions (Version B, Version C), but we only wanted the last beat, so we asked again for three ways to replace just that line, keeping everything else from the original:

  1. “No guarantees. Just the truth about what you built. NewScan. Free.”
  2. “Build fast. Know what’s real. Ship it. NewScan. Free.”
  3. “Ship like you know what’s in it. NewScan. Free.”

Chipp’s line

Chipp suggested: “Your customers trust you with more than their password. NewScan. Free.” It took a few passes to get the wording right (“their password,” not “your password”; drop the extra “with”; no question mark, because it isn’t a question).

Story Sherpa’s assessment:

It’s a strong idea, and it gives the spot something the other options don’t: it turns the story outward. Up to this point the stakes are all Jordan’s nerves. This line reminds us that the real stakes are Maria and every customer like her.

What works:

  • It pays off the doubt beat. We showed Maria’s address on screen, and this line names why that moment felt bad.
  • It raises the stakes from “my launch” to “my responsibility.” That’s the deeper, more adult reason to care.
  • It’s respectful. It treats the builder as someone who already cares about customers rather than scaring them.

What to watch:

  • It reopens tension after the release. After the click and the dawn exhale, the emotional win, a line about what customers trust you with pulls the mood back toward worry, so the spot ends on the problem rather than the resolution.
  • “Password” is slightly off target. The spot never shows a password; it shows her name, address and card.

Small wording fix: drop the second “with.” It should read, “Your customers trust you with more than their password.”

It offered two endings:

  • “Your customers trust you with more than their password. NewScan. Free.”
  • “Your customers trust you with more than their password. Now you can honor that.”

Chipp and David both liked keeping “NewScan. Free.”

Chipp noted that ChatGPT came up with that trust line originally, and it’s better than any of the ones Claude came up with. So David will need to integrate ChatGPT, just not from the cloud. Chipp’s tip: in OMP you can just type /model, switch to ChatGPT, create the line, then /model back to Fable. David may get to that this weekend; he’d told Larry he needs to, since he jumps back and forth all the time.

Output Files

  • There are three outputs: the Story Sherpa breakdown run, the production script as a DOCX, and a plain-text version. The DOCX is nice because it’s visual and in color.
  • There’s no MD version yet. Note to self: add an MD version of the script to the skill. David asked for one (“Give me an MD version of the script”), and it generated it.
  • We downloaded it into a new folder on Chipp’s drive. Not “Anthony Story Sherpa,” because the whole video project will go in there; we called it Anthony new.
  • We opened it in MD Docs and made MD Docs the default app for MD files.

Then we paused for a glass of water and stopped the recording.