12:59Luke Scripted · Essay
You Don't Need Fancy Editing to Make Sales from YouTube
A screen-share creator explains why editing tricks don't create sales, and the two things that do.
August 13thThe argument in one line.
Claude Fable 5.1's jump in the systems-building score means it can carry a long multi-step workflow to the end without silently skipping steps, letting a creator one-shot a finished YouTube script from a single unstructured brain dump.
Read if. Skip if.
READ IF YOU ARE…
- You use Claude or another LLM to write long-form YouTube scripts and have hit the problem of it randomly skipping steps in a multi-stage workflow.
- You maintain a Claude skills folder or prompt library for content production and want to know if it's now over-engineered.
- You want to see a real-time test of an LLM going from a raw brain dump to a finished, voice-matched script using sub-agents.
SKIP IF…
- You've never used Claude for scripted, multi-step content workflows and just want general AI news.
- You're looking for a build tutorial that explains the underlying automation step by step, not a live demo and reaction.
The full version, fast.
Claude Fable 5.1 posted the largest jump yet on Anthropic's building-business-workflows benchmark, and this video tests whether that translates to a long job like YouTube script writing surviving start to finish without silently dropping steps, the failure that broke every earlier model in the creator's year of building script systems. Skipping his usual 40-minute automated pipeline, he gives the model a vague brief, a folder of past scripts, and one brain-dump message, then lets it spin up sub-agents to draft, stitch, and quality-check the sections itself. The result needs light editing, not a rewrite. His takeaway: newer models perform better with a goal and one example of 'good' than with an exhaustive rulebook of instructions.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:53
01 · The score everyone is ignoring
Opens on Claude Fable 5.1's business-workflow benchmark score nearly doubling, from 17.1% to 31.4%, beating Opus 5 and GPT-5.6.
00:53 – 01:42
02 · Why long jobs kept breaking
Older models like Opus 5 would randomly skip steps in a multi-step script pipeline, making output inconsistent even when individual steps worked fine.
01:42 – 02:05
03 · Everything I tried before this
Runs through the tool history: ChatGPT, then Jasper and other wrapper tools built on the same underlying models, then building his own system on Claude.
02:05 – 03:03
04 · The 40-minute workflow I normally run
Walks through his standard automated pipeline (outline, parallel section writing, quality pass, linter, stitch) and where it used to fail silently with no error or warning.
03:03 – 04:40
05 · The live test starts
Switches to a screen-share inside an agentic coding IDE and asks Claude to skip the full 40-minute pipeline and just write a script directly, with no prior instructions.
04:40 – 08:40
06 · I give it the full brain dump
Points Claude at a workspace of past 'gold standard' scripts and reference docs, then delivers one spoken and typed brain dump covering the video's angle, stats, and structure.
08:40 – 09:49
07 · Why I kept my instructions vague
Explains he deliberately withheld his detailed rulebook (AI slop ban list, voice doc) because newer models perform worse when overloaded with highly specific rules.
09:49 – 10:33
08 · Sub-agents write the script
Claude proposes and runs a workflow where parallel sub-agents draft each section, a coordinator stitches them together, and a final pass checks flow and AI-sounding phrasing.
10:33 – 12:42
09 · Reading the script it wrote
Opens the finished draft and revision file, reads the AI-written intro aloud, and reacts to the quality: few remaining 'slop' lines, no major mistakes.
12:42 – 13:54
10 · What makes or breaks the output
Reviews the stats (1,243 words, reading level 4.5) and stresses that a thin prompt still produces thin output, then pitches the Client Society community mid-video.
13:54 – 14:55
11 · What this means for business owners
Closes on the competitive implication: AI scripts near top-copywriter quality lower the barrier to entry, and creators who don't adopt the workflow risk falling behind.
Lines worth screenshotting.
- Claude Fable 5.1's business-workflow benchmark score nearly doubled, from 17.1% to 31.4%, beating both Opus 5 and GPT-5.6 on that measure.
- The failure mode in long AI workflows usually isn't bad output quality, it's silently skipped steps that break the pipeline downstream.
- A creator who spent a year building automated script-writing pipelines found the newest model could one-shot a finished script from a single brain dump, no pipeline required.
- Giving a modern LLM a goal plus one example of 'good' now outperforms handing it an exhaustive, highly specific rulebook of instructions.
- Over-specifying instructions to newer models can confuse them and produce worse output than a simpler goal-and-example prompt.
- A roughly 1,243-word YouTube script came back from sub-agents in one pass at a reading level around 4.5, needing light edits, not a rewrite.
- The workflow uses parallel sub-agents: one drafts each section, a coordinator connects the seams, and a final pass audits for AI-sounding phrasing.
- AI script detection increasingly comes down to structural tells, like the 'not only X, but Y' construction, more than obvious cliche vocabulary.
- A YouTube script written at 85 to 90 percent of a top-tier copywriter's quality is now achievable with AI, lowering the barrier to entry for competitors.
- The creator's own AI video-editing system got audited and partially repaired as a side effect of the same brain dump used to write this script.
Long AI workflows fail from skipped steps, not bad ideas
AI WORKFLOW DESIGN
The real gain in Claude Fable 5.1 isn't smarter writing, it's finishing every step of a long job without silently dropping one, and that changes how much scaffolding your prompts actually need.
01The score everyone is ignoring
- A model's specific business-workflow or agentic-task benchmark score predicts long-job reliability better than its general coding or reasoning score does.
- When comparing models for a multi-step job, check the systems or workflow benchmark specifically, not just headline coding or knowledge scores.
02Why long jobs kept breaking
- A multi-step content pipeline (research, outline, draft, revise, quality-check) is only as reliable as its weakest, most-often-skipped step.
- Inconsistent AI output on a repeatable workflow is usually a sign the model is dropping steps randomly, not that the prompts are wrong.
03Everything I tried before this
- Wrapper tools built on top of ChatGPT or Claude inherit the same underlying model's weaknesses, so switching wrappers doesn't fix a reliability problem.
04The 40-minute workflow I normally run
- A quality-control linter for AI writing can be built as literal pass or fail rules: sentence length, banned words, banned patterns like 'not only X, but Y.'
- Building a rules-based AI slop detector turns vague 'sounds like AI' judgment into a checklist a script has to pass before it ships.
05The live test starts
- Deliberately withholding your detailed rulebook and giving only a vague ask is a valid way to stress-test how much a newer model can infer on its own.
06I give it the full brain dump
- Keeping a workspace of your best past outputs lets an AI model learn your style by reading it, not by being told rules about it.
- A single spoken or typed brain dump, combined with reference files already in your workspace, can replace a formal creative brief.
07Why I kept my instructions vague
- Long, highly specific instruction lists can confuse newer models more than they help, because the model has to reconcile too many rules at once.
- Giving a goal plus one strong example of 'good' now gets better results from capable models than an exhaustive rulebook.
08Sub-agents write the script
- Splitting a long-form draft into per-section sub-agent jobs, then running a final full-read pass to smooth the seams, catches both section-level and whole-document errors.
09Reading the script it wrote
- A first AI draft that only needs line edits, not a structural rewrite, is the real signal a workflow is working, not the draft's word count or speed.
- Comparing a revised draft against the AI's original and feeding that diff back into your system is how the system actually improves over time.
10What makes or breaks the output
- The input still decides the ceiling: a thin prompt like 'write a video about X' produces thin output even from a much stronger model.
11What this means for business owners
- As AI-written scripts approach 85 to 90 percent of top-copywriter quality, not adopting the workflow becomes the bigger competitive risk, not using AI badly.
Terms worth knowing.
- AutomationBench
- A benchmark suite that scores AI models on building business workflows, used here as evidence that Claude Fable 5.1 handles long, multi-step tasks more reliably than prior models.
- Sub-agents
- Separate AI agent instances spun up by a main Claude session to handle one piece of a larger task in parallel, then merged back into a single output.
- Brain dump
- A single unstructured, spoken or typed stream of ideas handed to Claude in place of a formal outline or creative brief.
- AI slop
- Content that reads as obviously AI-generated, often flagged by banned words, cliche phrasing, or repeated sentence patterns like 'not only X, but Y.'
- Linter (for scripts)
- A rules-based quality-check step, borrowed from the coding term, repurposed to catch banned words and AI-sounding patterns in a script draft before it ships.
Things they pointed at.
00:00toolAutomationBench
03:04toolGoogle Antigravity IDE
Lines you could clip.
10:38
“So this whole video, every cut and every graphic you're seeing right now was edited by AI. I never touched the timeline.”
11:40
“So the problem was never taste, it was follow through.”
11:11
“Every one of those mistakes look different, but line them up and they're all the same mistake.”
08:55
“It's actually better now to just give it the goal that you want it to achieve, and then an example of something that's good.”
13:56
“Anyone can come and write a YouTube script at 85%, 90% of quality as a top 1% copywriter, which is kind of scary because it's lowered the barrier to entry.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
So CloudFable 5 .1 came out a few days ago, and it's completely changed how you write YouTube scripts. Now, most people are looking at the coding scores, but I think the score that matters to us the most is the one for building business systems, because this one nearly doubled from 17 .1 % up to 31 .4%, and it also beats Opus 5 as well as GPT 5 .6.
And so I basically spent the last year building script writing systems inside of Cloud. I've had the old models run it like Opus 5 and so forth. But often a lot of the steps would get skipped.
The workflows would go really weird sometimes. But now with Fable 5 .1, you can read the whole system and then repair the gaps. And it did everything in one pass.
And so in this video, I'm just going to walk through everything that I changed, as well as do a live test of writing a YouTube script inside of Cloud. With the new fable 5 .1 model and so going back to these numbers right here basically in simple terms What this means is that when running a long job that has a lot of steps This is where fable 5 .1 performs best and with writing YouTube scripts This is especially true because there is a lot of steps first what we need to do is do the research and then you have to do the outline and then the first draft and then you do revisions and we have quality control checkers we have linter steps that run and look for ai slop and so there's many steps that need to be run but often steps will get skipped randomly and so the quality was inconsistent and now all of that is a lot more consistent and also fable 5 .1 is a lot cheaper now i noticed opus 5 although they said a lot of the benchmarks were higher it didn't really perform
as well as fable five and so hopefully fable 5 .1 is even better and so i've tried a lot of things with writing youtube scripts with ai i use chat gpt i use jasper and a few of the wrapper tools but they were basically just clawed or chat gpt under the hood i started using claude recently and and this has been what i found to be the best but sometimes steps will get skipped and the output just wasn't 100 there but with fable 5 .1 i think it's a lot better so We're going to try it out.
But the main thing I want to use Fable 5 .1 for is to actually design the workflow. Now, normally what I would do is I would run a full workflow. And so there'll be a Python script, like parallel writing steps, quality passes, a linter step, which basically what it does is it turns words into a equation.
Like it looks at the number of words in a sentence. It basically asks yes or no. Does it pass these criteria that we set?
So for example, the sentence reading level. Or if a certain word slips in that we don't like, because often AI likes to use really weird words that sound a bit cliche. Also for AI patterns that always comes in, like not only X, but Y.
It loves that one a lot. And so we're going to test it out. But instead of running this super long workflow, because this usually takes 40 minutes, like I would leave it.
It runs for 40 minutes and I get the script back. And often it's good to film. But right now, because I don't want to sit here for 40 minutes while filming this video, I just want to try having it write it.
So I want to see if it can reproduce like the quality of scripts. that i'm looking for so i'm inside of the google anti -gravity ide i'm just going to chat with it and let's see what we can do hey claude i want to write a youtube script and i want to test out the new fable 5 .1 but i don't want to run my typical write script parallel workflow because that's going to take a long time i'm actually filming a video i just want to demo the process and i want to see how well claude can write a script without me giving it any prior instructions and so just look at the scripts in my workspace the fully written scripts we have a lot of them in there so just look at one of them and look at how i structure it look at my writing style and then try to replicate it are you able to spin up a bunch of sub agents to then write the different sections of a script i can give you a brain dump right now and i just want to try it out see how it goes try to keep the the process similar to what we do usually so for the video idea i'm thinking of doing this just title it
Claude Fable 5 .1 just changed video editing forever. Now that's the video title. I'm going to do a brain dump afterwards, but I just want to get started on the process.
So yeah, we just started that. Let's see what it could whip up. So inside of my workspace, I have a lot of my past scripts loaded in here, like the gold standard scripts.
We have all the scripts I've written. So Claude can look at that and use that as the benchmark of what I like. We also have skills and things that I use for different parts of the script writing process.
We have the AI slot bandwidth. There's a lot of files in here that it's going to look through just to get up to speed with my writing process. And so if you want all these things, I actually have a Google Doc.
Like this Google Doc right here just has a bunch of different skills I use, the AI slot ban list, voice doc, YouTube strategy, how to do visuals, a bunch of different things. So if you want this, I'll put a link to this in the description for you to download. So while it's doing all this stuff, I'm actually going to do the brain dump right now just so it has the information.
And then once it's ready, it can now one -shot the script. Hey, Claude, here's my brain dump for the video. And so it's going to be about how ClaudeFable 5 .1 has changed video editing.
it's now made it way easier and so what i want to do is at the start of the video basically just say that this video right now has been edited by ai and so they're visibly seeing the effects happening on screen to get an idea of how good the quality is and so i can dive into my own personal story about how i've tried using cloud previously it would make all these silly mistakes it would cut at the wrong moments cut out bits that I was actually speaking, not remove the right retakes and just do a lot of things like that.
And so dive into my story. And then from there, I want to go through the stats. And so just show on screen, the benchmark stats, the numbers and what Claude is getting better at.
I think the one that's important is the systems building number and how we can now use Claude fable 5 .1 to build editing workflows that are more consistent, I can get us the output we want. And so talking about how I had to go through my current editing workflow, I built this full editing workflow and I had it spot all the weaknesses, the bits where there were issues that we had to fix.
And actually you might have on file the things that it did update because I did this process in this cloud workspace. You can look for that. And then at this point, it'd be a good point to add a CTA for them to join the client society.
So that's my school group. Just say, hey, if you want more of these AI workflows, a deeper dive into them and all the templates and things like that, join client society. from there i want to talk about what this means and so with youtube now you're able to one shot full videos that are fully edited it requires you to give it the information though like the rules of when to do what when to do a pop sound effect when to do a red x sound effect when to do sliding it's like you have to give it the rules now what you could do is give it an example of what good looks like and so you give it a full video you tell it to screenshot every frame in the video and then look at how the elements move in when they move in reverse engineer the editing decisions of when you use a meme versus when you use an animation or a motion graphic versus when you use the talking head.
It needs to be rules, right? And so what it can do is study videos and learn from those rules and then create a rule book that it follows every time it edits. But at a certain point, you can just mass produce videos that are at a high quality.
And there's going to be channels that are just fully AI generated, producing actual good videos. Like it's not going to be AI slop content. There are going to be a lot of AI slop content and you're not going to stand out.
But if you can make videos that are actually good, are high quality, they'll just be a lot of AI, but AI gold. And that's what makes YouTube just completely just super tough for people starting out because these mass produced AI channels are actually making good content.
And if you don't join it, and you don't create AI videos, you're going to fall behind. So something like that, and then end screen call to action to my previous video, which shows how I built my AI editing system and fired my editor. And so this is for the Luke unscripted channel for context.
So I want to write a script that's in my style. in my voice can you run through that please and so as i did that brain dump it actually gave some information here so let me just copy this so i can read this through okay so added this in above is the brain dump combine that with the things you thought of here and what do you think we should do should we just one shot a script or would you go section by section what's your idea and what we can do here would you be able to one shot a script but not take 40 minutes with that full workflow so what it's done is save my brain dump into this little message right there verbatim from the chat.
So you can use that to write the script. And then it's going to look through my workspace. And because this is something I've been doing inside of my workspace, I've literally been building this AI editing system.
And so we can just look at the editing system that I built when I made changes to it, and just write out the actual updates and just use that for the video. And so the cool thing about this is if you're working inside of your cloud workspace, whenever you do things inside of it, and you make a video on it, It's right there.
You don't even have to tell it the information. You can just find it from your past chats. It's going ahead and it's writing the script right now.
Now, one thing you might have noticed is I gave it deliberately very vague instructions. I didn't tell it what to do specifically because with the updates, right? I have these very elaborate skills.
Like if you go to my skills folder, like just this AI slot ban list, as an example, there's... so many lines of it. And this is actually a shortened version.
This used to be like 300 lines. And I noticed that when you give the newer models of Claude super intense instructions with so many specific details, it ends up confusing it even more. It's actually better now to just give it the goal that you want it to achieve, and then an example of something that's good.
And so it looks at the example of what good looks like, and then the goal of what it's trying to do, which is replicate it. And then you have it just go and figure it out. Previously, I would have like a thousand steps and it starts following it and actually messes up and it's more likely to make mistakes because now it's just getting confused.
Because in the past versions of Claude, you would need to give it very specific rules because they weren't as good. But now that they're a lot better, they don't need that many instructions. And in fact, it just overwhelms it with more details that makes it perform worse.
And so right now I'm going through and I'm simplifying a lot of my processes. That way we can get a better output from... Claude when it comes to scripts, editing, whatever you use it to make content for.
Okay, so Claude's come back and it's given me the workflow it recommends. So we're going to go with what it says. So it's got the outline right here.
Let me just look through this and see if I approve. Okay, so looking through this, it looks good to go. Let me just say let's go looks good.
Claude, you can go and start writing the script. So what it's going to do is it's going to spin up a bunch of parallel sub agents to go and write each section. and then the main clod is going to look through all the sections and then connect them together and so there's the seams between the different sections that sometimes they don't really flow together it sounds a bit weird and so a final agent goes through make sure they sync together and they don't sound weird reads the entire script in one go make sure there's no mistakes and then checks for any ai slot before returning the script so i'm going to let that run and come back to it when it's finished okay so the video has been scripted we can see right here there's two versions of it There's the revisions version, which is the one I'm going to look at because this is where I make the changes.
And then after I'm done, I can then compare the before and after, update the system so it learns from that. But let's open this up. So Ctrl -Shift -V.
So this whole video, every cut and every graphic you're seeing right now was edited by AI. I never touched the timeline. And editing was the one job I hated most.
Paying someone to do it didn't go much better. But I built the editing system behind this video, and this week I ran the new cloud through every step of it. So I'll show you what Fable 5 .1 actually changed, what it found wrong with my own system, and how to get one yourself before everyone else does.
So that intro's... Pretty good. There's like no AI slop in there, which is pretty impressive.
Now, if you've ever tried getting AI to edit a video, you're probably thinking, hang on, mine came out like trash. It just doesn't know what a good edit looks like. Yeah, I thought that too.
The last time I handed Claude a video, it cut me off mid -sentence and chopped out bits where I was actually talking. And where I'd redone a line, it kept the wrong take. So I'd end up watching the whole video back myself, hunting for mistakes, which was the exact job I was trying to get rid of.
Now, every one of those mistakes look different, but line them up and they're all the same mistake. Back then, Claude couldn't get through a long job without skipping a step. Because editing is just one small decision after another right to the end.
Stuff like where to cut or which take to keep. And if Claude skips one, the whole video looks broken. So .
So this bit right here, this is a slop. I'll take that out. So it's not going to be perfect.
But as you can see, like you just remove the occasional lines that are slop and just skim through it. but for something where you just let it go create a first version like this is a decent version where you can run with it add in bits that you want and then from there you just chat like maybe there's a section where you want something you added in you could tell it that like hey actually upon reading this i feel like we're missing this other section can you add it in and then claude's gonna go and write that bit in and then you just talk with claude you edit it and then after a while your revisions file will be ready you can then have Claude compare the before and after the original versus your revisions make improvements to the system so that next time it could write in your standard and you just keep doing this over and over again and every single time Claude's gonna get better and better and better until eventually you could just one -shot scripts from your brain dump now the thing that matters is the input because you can't just say go and script the video and expect it to be perfect if you don't give it anything like with this case I gave it a full brain dump
it has access to my workspace, it can see, you know, things I've worked on to make sure things are accurate. In your case, if you're starting from scratch, and you just say write a video on XYZ, it's not going to be that good. Now you could have it scrape the internet, find content and synthesize something to teach, but that's not going to get you a really good result.
A lot of it starts from what you give it. Now I know that a lot of this stuff could feel confusing. There's this weird looking app right here, you might not want to set all of this up yourself.
If you want to just shortcut that and get all the templates all the different ai systems i have to write scripts then you can join the school community like i just recently launched the client society we have a five -day video challenge if you're struggling to get videos posted we have my youtube storytelling course from back in the day the full cloud script writing system guides on packaging scripts call recordings we have weekly call too so lots of useful things and ai systems in here and so we just started out so we're kind of small now but that probably makes it even better because I'll basically be able to chat with you one on one.
But as it starts to grow, it's going to be harder and harder to do that. But if you're interested, come and check it out. I'll put a link to this in the description.
And so what does this new update mean for business owners on YouTube who are trying to get clients and they want to write their YouTube scripts faster? Well, this is a little bit scary, honestly, because anyone can come and write a YouTube script at 85%, 90 % of quality as a top 1 % copywriter, which is kind of scary because it's lowered the barrier to entry.
So if you're someone who's spent your whole life mastering a skill, AI's kind of come in and just level the playing field, which kind of sucks. But it also means that you have to take advantage of the technology. If you don't, there's going to be people that mass produce content and they just win through volume and you're just going to fall behind.
And so we could complain about, oh, AI's taking over and all that. But the truth is, you're just going to have to learn how to do it and adapt. And those that don't, sadly, are just not going to do well on YouTube.
But those are my thoughts on the Fable 5 .1 update. Now, I made a video. last week on why views don't matter is actually horrible advice for business owners.
I cover how you can actually revive your channel if you're starting to fall out of the algorithm. I'll put it up on screen for you to watch next.
The Hook
The bait, then the rug-pull.
Claude Fable 5.1 just posted the biggest jump yet on the benchmark that predicts whether an AI can survive a long, multi-step job, and this creator puts it straight to the test by handing it a raw brain dump and asking for a finished YouTube script in one pass.
Named ideas worth stealing.
02:19list
The old script pipeline
- Outline
- Write
- Quality pass
- Stitch
- Linter
- Wrong script (silent failure point)
His standard 40-minute automated workflow, and the point where older models used to silently drop a step with no error or warning, producing a broken script.
Steal forany long multi-step AI content pipeline where you need to catch silent step-skipping before it ships
01:46list
How I got here
- ChatGPT (2023) - sounded like AI
- Jasper + wrappers (2024) - all the same underneath
- Claude (2025) - built my system, but steps got skipped silently
- Fable 5.1 (Sept 2026) - fixed the gaps
A quick timeline of the tools he's tried for script writing and why each one fell short until this release.
Steal forframing a 'why now' section when reviewing a new AI model release
08:55concept
Give a goal and one example, not a rulebook
With newer, more capable models, an exhaustive list of specific rules confuses them more than it helps. Giving the model a clear goal plus one strong example of 'good' now produces better output.
Steal forany Claude prompt, skill file, or system instructions you've over-engineered with too many specific rules
How they asked for the click.
VERBAL ASK
12:51product
“if you want more of these AI workflows, a deeper dive into them and all the templates and things like that, join Client Society”
Soft-pitches his paid Skool community mid-video, right after demonstrating the free workflow live, framing it as the templates and systems behind what was just shown rather than a hard sales break.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
