Why Modern Creator?

Dream Labs AI · YouTube

How I Make Claude Opus Think Like Fable 5 (Best Method Tested)

A creator runs the same workout-app build through Opus 4.8, Fable 5, and three advisor-mode hybrids to find the cheapest way to get Fable-level output without paying full price.

Posted

July 10th

1 months ago

Duration

09:41

Format

Review

educational

Views

264

14 likes

Part of the collectionThe Fable 5 PlaybookAll 45 Fable 5 breakdowns, synthesized into one page.

Read the playbook

Big Idea

The argument in one line.

Fable 5 produces the best-designed AI-built app, but running it as the sole builder is slow and can cost roughly $50 per project; using it only as a design advisor over a cheaper, subscription-included executor model gets close to that quality for near zero extra cost.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You use Claude Code or a similar agentic coding tool and want more design polish without burning API credits on the priciest model for every build.
You've heard of using a premium model as an 'advisor' or 'orchestrator' over a cheaper executor model and want to see whether it actually works.
You're deciding between Claude and Codex-based tooling and want a head-to-head look at output quality, not just benchmarks.

SKIP IF…

You already have unlimited budget for premium-model API credits and don't care about cost.
You want a step-by-step coding tutorial rather than a comparison of AI-model workflows.

TL;DR

The full version, fast.

The host built the same 90-day workout-tracker app five ways to find the cheapest route to Fable 5-level design quality. Fable 5 alone produced the best UI but took 43 minutes and could cost up to $50 in API credits; Opus 4.8 alone was faster and cheaper but visually generic. Using Fable 5 as an on-call 'advisor' over an Opus 4.8 executor (Head Chef mode) was fastest of all and landed roughly halfway between the two in quality. A prompt-only method that just told Opus to 'think like Fable' produced broken, non-functional output. The winning combination swapped the executor from Opus to OpenAI's Codex under the same Fable-advisor pattern, producing the closest match to Fable's design sensibility while running on an existing subscription instead of paid credits.

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 →

Chapters

Where the time goes.

00:00 – 00:46

01 · Intro

Fable 5 is the best but most expensive model; the host sets up a test of cheaper ways to approximate its output.

00:46 – 02:28

02 · Test setup

A detailed single-file build spec for a 90-day 5x5 workout tracker ('NINETY') is sent to three parallel Claude Code sessions: Opus 4.8, Fable 5, and Opus 4.8 with Fable 5 as advisor.

02:28 – 02:52

03 · Build times and cost breakdown

Opus 4.8: 23 min. Fable 5 alone: 43:46 (up to $50 in API credits). Opus + Fable advisor: 16:29, the fastest of the three.

02:52 – 03:48

04 · Opus 4.8 results

Clean, functional 'classic Opus' UI; workout flow, celebration animation, wall mode. Scored 6/10.

03:48 – 04:40

05 · Fable 5 results

Visibly higher-class UI -- tighter spacing, nicer stopwatch and settings, top-aligned nav. The quality bar the rest of the video is chasing.

04:40 – 05:45

06 · Head Chef mode results

Opus 4.8 executing under a live Fable 5 advisor call. Better than Opus alone, still short of full Fable -- lands 'halfway between' the two.

05:45 – 06:54

07 · Nate Herc's 'think like Fable' method

A prompt/skill file telling Opus to imitate Fable's design process, with no live advisor call. Produces broken, non-functional UI -- called out as a failed approach.

06:54 – 08:06

08 · The Codex workaround

A Peter Steinberger reply to the same advisor-pattern tweet suggests swapping the executor to OpenAI Codex. The host wires Claude Code to Codex under a Fable advisor.

08:06 – 09:41

09 · Final results and winner

The Codex-executed build is judged closest to Fable's design sensibility of any cost-saving option -- heavier typography, smoother chart, cleanest layout -- while running on an existing subscription. Video ends with a community-offer CTA.

Atomic Insights

Lines worth screenshotting.

Running a premium AI model as the sole builder of an app took 43 minutes and could cost up to $50 in API credits for a single small web app.
The same build task run through a cheaper model alone finished in 23 minutes but produced a visually generic, 'classic AI' interface.
Using the expensive model only as an on-call design advisor over a cheaper executor model was the fastest of five tested setups, finishing in under 17 minutes.
An advisor-mode build landed 'halfway' between the cheap and expensive model's design quality -- better than the cheap model alone, but short of the expensive model's full output.
A prompt file that instructs a cheap model to imitate an expensive model's design process, without an actual live advisor call, produced broken UI with non-functional buttons.
Swapping the executor model in an advisor pattern from same-vendor to a different vendor's coding agent got closer to the premium model's design quality than keeping both models from the same vendor.
None of the four cost-saving workarounds fully matched the premium model's output quality -- the advisor pattern narrows the design gap, it doesn't close it.
The advisor/executor pattern being tested originated from a developer-tools account's tweet, and the winning variant came from a reply to that same tweet suggesting a different executor.
A subscription-included coding agent that 'feels free' compared to paid API credits was, in this test, the closest cost-effective substitute for a premium model's design output.

Takeaway

The cheapest path to premium AI output quality

MODEL STRATEGY

Using an expensive, high-quality AI model only as a design advisor over a faster, cheaper executor model captures much of its polish without most of its cost.

Running a premium model as the sole builder produced the best design but took nearly twice as long as a cheaper model and could cost roughly $50 in API credits for one small app.
Delegating execution to a cheaper, subscription-included model while keeping the expensive model as an on-call advisor cut total build time below either model running alone.
An advisor-mode build lands partway between the cheap and expensive model's quality -- meaningfully better than the cheap model alone, but not a full substitute for the expensive one.
A prompt that simply tells a cheap model to imitate an expensive model's style, without an actual advisor relationship, produced broken, non-functional output.
Pairing the advisor model with a different vendor's coding agent as the executor got closer to the premium model's design sensibility than keeping both models from the same vendor.
None of the cost-saving workarounds tested fully matched the premium model's output -- the advisor pattern narrows the design gap, it doesn't close it.

Glossary

Terms worth knowing.

Advisor / Head Chef mode: An agent pattern where an expensive, high-quality AI model is called in only for design decisions and hard problems, while a cheaper model does the actual code-writing -- like a head chef who plans and tastes but doesn't do the chopping.
Executor / workhorse model: In the advisor pattern, the model that actually writes and edits code, taking direction from the advisor model rather than making high-level design calls itself.
Orchestrator pattern: A workflow where one AI model coordinates and delegates tasks to one or more other models or sub-agents, rather than doing all the work itself.

Resources

Things they pointed at.

00:53productStrongLift

05:46channelNate Herc -- "Fable Mode: How I Make Opus Think Like Fable 5"

07:19channelPeter Steinberger (@steipete)

09:26productDream Labs Community ↗

Quotables

Lines you could clip.

00:20

“It's more of the head chef of the kitchen that delegates, looks at the menu, tweaks some things, tastes things with his pinky, but he's not doing the chopping, the stirring, and the cleaning up.”

Clean, memorable analogy for the advisor/executor pattern→ TikTok hook↗ Tweet quote

02:29

“Fable actually took the longest at forty three minutes and forty six seconds, which if you're using API credits for this, it could be upwards of $50 for this one website.”

Concrete cost stat, easy pull-quote→ newsletter pull-quote↗ Tweet quote

06:33

“It absolutely had an aneurysm and did not work.”

Punchy failure line→ IG reel cold open↗ Tweet quote

09:15

“This nice bar at the top that goes gradient colors, this to me by far wins and feels the most like Fable.”

Clear verdict statement→ TikTok hook↗ Tweet quote

The Script

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.

metaphor

Fable five is an absolute miracle, but it's extremely expensive. And so in this video, I wanna test three different ways to get fable mode outcomes without paying a crazy hefty price tag. And the first theory starts from Claude Dev's official Twitter account.

They said, use Fable five as an orchestrator. So instead of it doing all the heavy grunt work, it's more of the head chef of the kitchen that delegates, looks at the menu, tweaks some things, tastes things with his pinky, but he's not doing the chopping, the stirring, and the cleaning up. Instead, you have Fable five delegate all of that work to free if you're on the subscription plan with Claude, things like Sonnet five or Opus 4.8.

But I've seen a lot of people talk about this, but very few people actually test it. I wanna put this to the test in this video with you. And so I just built a home gym in my spare bedroom, and I want a workout tracking app for all the compound lifts.

There's one called StrongLift that already exists, but I want my Claude to build this from scratch in fable mode. And so first, I spun up this prompt. Now this prompt is going to be the same across all three instances of this test.

This is a highly detailed explanation of exactly what I want them to build for me for my workout app in my new gym. And so I've spun up three instances of Claude code in the terminal. Instance one is Opus 4.8.

And so I'm gonna send that prompt off to him so he can build that app for me. The second one I want as a baseline test. What does Fable five's output actually look like even if it is expensive?

Send him off that exact same prompt. The third one though is really special. First of all, I have to change the model back to Opus 4.8, the free model in the subscription plan.

So you can see up here, Opus 4.8. And then I'm gonna send him this exact prompt, the head chef mode where we have him spin up a fine Fable five sub agent who's gonna do all the design, all the organization, and use his superior brain to then instruct Opus to go ahead and build it.

So we're gonna copy this prompt out and paste it into the third instance of Claude code. And then once he acknowledges that, he says, ready. Head chef, Fable five mode is on call for the hard calls, and then Opus will do the work.

What are we building? We're gonna send him the exact same prompt to build me my software for my new gym. Okay.

So all three of those have now finished, and they all took very different times. The Opus 4.8 model took twenty three minutes to build the app.

Fable actually took the longest at forty three minutes and forty six seconds, which if you're using API credits for this, it could be upwards of $50 for this one website. And Opus 4.8, but using Fable five as head chef, actually took the quickest at sixteen minutes and twenty nine seconds.

But what do these apps actually look like? So this one here is Opus 4.8, which looks classic Opus 4.8.

Very AI inspired. Um, it's not bad. It's got little boxes here where we put our starting weights and our starting body weight, and you click start my ninety day challenge, which is five by five on the compound lifts, which is what I'm currently working out in case you're interested.

Then you begin your workout here and you tick it off as you go. It's got a little celebration and a timer between each set, which is kind of cool. And if you don't get the amount of reps, you can click it down.

So it's pretty damn good. And before AI, it would be absolutely mind blowing if I could get a developer to build something like this for me. I've also got a progress mode over here, and I've asked them to build a ninety day demo data so we can actually see what it looks like when someone's put data in.

And this is what it looks like, which is not bad. It's a little bit small. The stars are a little bit small.

There is a wall mode, which if I'm having an iPad hanging on my wall, it tracks my lifts over time. It's decent. I give it a six out of 10.

So let's compare Opus 4.8 to what Fable created. So Fable always blows me away. This is a very clean looking app with plus and minus little buttons in here to adjust the initial weights.

Body weight also has it. But it's just like if we go back and we go back to that home page, like there's such a world of difference in class between those these two apps. Now it also lays them out across the top, which I find a lot easier.

And the spacing's much nicer. The stopwatch is much nicer. Everything about this, even the settings are way more compact.

I'm gonna load the demo data in so you can see what it actually looks like, how it logs it down there. The chart is somewhat similar, and it is a very clean, very amazing app, even the rounding's down here.

And it just feels quicker and more Fable like. This is what everyone wants to keep, but we don't wanna pay hundreds of dollars for every time we wanna spin up a website. So what does Headchef Opus and Fable mode look like?

Well, this is what it looks like here. Initially, looks much better than Opus, but in my opinion, still is not looking or feeling as good as Fable.

So even inside here, if I go back to Opus 4.8 and start the ninety days and then begin workout, you can see these circles are weirdly spaced across where Fable has instructed Opus 4.8, and it is much cleaner, much nicer. It doesn't actually have the number in there, which I would like, but it does if you miss your workout and does all the timing and stuff.

Even the settings are a lot cleaner and nicer with the plus and minus buttons. The colors are nice. If we have a look at our progress over here, let's actually load in that demo data, load sample data.

It's looking pretty good. It's nowhere near what Fable created. But, honestly, I think it lands halfway between Opus and Fable, which I'm kinda disappointed by because we want Fable outputs.

Especially if we're paying Fable credits to be instructing its workers, I would like some exceptional output. And so I was a little bit lost until I came across this video by Nate Herc.

Nate Herc, an absolute monster in the AI space, has this video called Fable Mode, How I Make Opus Think Like Fable five. And Nate actually went and had Fable recipe out its own method and then give that to Opus four point eight so it could think like Fable.

And you don't need any Fable API credits. You don't need as the head chef. You just need to instruct Opus four point eight to think like the miraculous Fable.

So I followed Nate Herc's advice even though we didn't give an example in his video, downloaded his skills files from his school community, and spun up an instance of this exact same app using 4.8 thinking like Fable five. And I don't know what happened, to be honest, but it absolutely had an aneurysm and did not work.

Like, it literally the settings button doesn't work. The progress button doesn't work. It doesn't start you on a place where you can put your weight, and it doesn't even look like Fable five designed at all.

It's looking a lot like our Opus model designed it. And so once again, I was a little bit confused and pretty stuck because I want these Fable five outputs without paying the crazy prices.

At this point, I was gonna scrap the video. I'm like, I don't know if it's possible to actually do this. We're gonna have to wait till the new ChatGPT model comes out and hope it's as good because they're including that as the subscription, which it is kind of weird because I think Fable will come back a lot quicker than we expect just because ChatGPT and Cord are gonna be at war with each other to have the best model.

And if we have to pay for one and the other one's included in the subscription, we're all gonna migrate over to ChatGPT. Then our friend, Peter Steinberger, saw the same tweet that we started at with Chord Devs, where it's using the adviser of Fable five, the head chef as Fable five, and the executor is Sonnet five or Opus 4.8 as we're using.

Peter Steinberger actually says, if you run this workflow, ask Fable to make Codex the workhorse.

Now Codex is ChatGPT's equivalent of Claude code. And so I got my Claude code to link up to my Codex so that now Fable five, yes, is still the head chef, but Codex's latest model and remember, there's a new model dropping really soon, so these results are gonna improve more and more from here on in, was actually the Workhorse.

And Codex's latest model, of course, is included in their subscription. I'd say free, but we are paying a subscription for it. But it does feel free compared to paying for API credits.

And when I saw this result, I was impressed. It is a different style. Let me actually get back to the home page of this, reset this program.

It is a different style. It is more pragmatic. I feel like Codex is always more pragmatic, but it's smooth and works in a way that is most like the Fable outcome.

So you put your body weights in here. It's less color coded. There's no plus minuses.

But once you get in, it's laid out just like Fable laid it out. We've got the New York subway style font up here, uh, and it is a heavy style font. But if we populate it with real data, and you can see this is nice and compact.

It all fits on one screen. Load that demo data in here. Uh, you can see less colors, as we've been saying, but way more functional, way more clean.

The chart looks amazing. It feels amazing. It's way more smooth.

You've got your little menu down here. There is a little bit of an error here where they give us the tick and the fire, but those things are very easy to come in and fix these minor things. It's the big design choices that Fable is so good at that I'm looking for here.

And even this little timer popping up here in the black is nice and clean. This nice bar at the top that goes gradient colors, this to me by far wins and feels the most like Fable. And if you wanna get personal help with your AI to set it up in the best way possible so it outperforms all of your competitors, come and join us inside our Dreamlabs community where we are literally there twelve hours a day to help onboard you into the best settings, supercharging your AI, and taking you by the hand and walking you step by step through this AI revolution so that you can conquer it and come out ahead.

Me and the team are literally waiting in there for you now. Click the link in the description below and come and join us.

The Hook

The bait, then the rug-pull.

Fable 5 is, by the host's own account, the best AI model for building well-designed apps -- and one of the most expensive to run outright. So he builds the exact same workout-tracker app five different ways, from full-price Fable to a broken prompt-only imitation, to find the cheapest setup that gets closest to Fable's output without the Fable-sized bill.

Frameworks

Named ideas worth stealing.

00:20model

Advisor / Executor (Head Chef) pattern

Advisor model (Fable 5) makes design and hard-problem calls
Executor model (Opus 4.8 or Codex) writes the actual code
Executor consults advisor only for non-trivial decisions, not routine edits

A workflow where an expensive, high-quality model is called in only for design decisions and hard problems while a cheaper model does the routine build work.

Steal forAny AI-assisted build where a premium model is too slow/expensive to run for every line of code but its design judgment is worth keeping in the loop.

CTA Breakdown

How they asked for the click.

VERBAL ASK

09:26product

“Click the link in the description below and come and join us.”

Soft-launched at the very end after the technical content wraps, framed as ongoing personal help setting up AI workflows rather than a hard sales pitch.

MENTIONED ON CAMERA

09:26productDream Labs Community ↗

FROM THE DESCRIPTION

PRIMARY CTAWhere the creator wants you to go next.

🤖 Join the Dream Labs Community ↗

Storyboard

Visual structure at a glance.

cold open

hookcold open00:00

build spec

promisebuild spec00:46

Fable results

valueFable results03:48

Codex winner

valueCodex winner08:06

community CTA

ctacommunity CTA09:26

Frame Gallery

Visual moments.

cold open

Frame at 00:08 from How I Make Claude Opus Think Like Fable 5 (Best Method Tested)

Frame at 00:19 from How I Make Claude Opus Think Like Fable 5 (Best Method Tested)

Frame at 00:25 from How I Make Claude Opus Think Like Fable 5 (Best Method Tested)

Frame at 00:32 from How I Make Claude Opus Think Like Fable 5 (Best Method Tested)

Frame at 00:43 from How I Make Claude Opus Think Like Fable 5 (Best Method Tested)

Watch next

More from this channel + related breakdowns.

09:00

Dream Labs AI · Tutorial

Andrej Karpathy Predicted AI Agents Need to "Dream" — Anthropic Just Shipped It

A nine-month-old Karpathy prediction about AI memory becomes Anthropic's new overnight consolidation feature — plus a free routine that replicates it for anyone not on an enterprise plan.

August 3rd

12:27

Dream Labs AI · Tutorial

Stop Prompting Claude — Use Andrej Karpathy's Loops Instead

A breakdown of Claude Code's native /loop and /goal commands, shown live on a race-simulator agent and a newsletter-writing agent that grades its own drafts until they pass.

July 17th

10:42

Dream Labs AI · Reaction

Boris Cherny's Five Archetypes: The Business Structure Built to Run Thousands of AI Agents

A tweet-reaction breakdown of Anthropic's viral five-archetype framework — and the Slack-embedded Claude agent quietly filling the sixth role nobody named yet.

July 8th

12:53

Dream Labs AI · Tutorial

Claude Code's Creator Uses This to 10x his AI Workflow (Self Learning MD)

A YouTube host reverse-engineers Boris Cherny's one-line CLAUDE.md snippet, then runs a live three-way Claude Code test to prove a self-learning lessons file beats a static one.

July 5th

16:16

Dream Labs AI · Essay

Claude Code's Creator Reveals Every Level of AI (and How to Level Up)

A viral Boris Cherny tweet mapped five stages of Claude Code adoption — this breakdown turns the jargon into a pizzeria anyone can picture.

August 6th

Video of the Day15:19

Dream Labs AI · Demo

This NEW Claude feature is the best thing I've used in years

A narrated screen recording, no code and no prompt, turned into a repeatable skill — demonstrated live on YouTube competitor research and a dinner order.

August 11th