17:39Ryan Doser · Tutorial
Claude Fable 5 is Here! The Best AI Model Ever?
A 17-minute practitioner verdict on whether Fable 5 justifies double the price with live side-by-side demos against Opus 4.8.
June 10thThe argument in one line.
Building your operations around a single AI vendor hands that vendor control over your pricing, your uptime, and eventually your ability to do the work — and the deeper your dependency, the higher the exit cost when any of those change.
Read if. Skip if.
READ IF YOU ARE…
- You are running an agency, coaching business, or solo practice where Claude Code handles a meaningful share of your daily output.
- You have automated email sequences, client deliverables, or content workflows in Claude and have not thought about what happens if it goes down.
- You are trying to understand whether OpenRouter or local models are worth the setup complexity for your use case.
- You felt the quality drop between Claude 3.5 and later versions and want a framework for thinking about model-switching.
SKIP IF…
- You use Claude occasionally as a writing assistant but it is not load-bearing in your business — the risk framing will not apply.
- You are already running a multi-model routing setup with OpenRouter — the resilience stack section covers ground you already know.
The full version, fast.
Heavy reliance on Claude Code exposes a business to four compounding risks: rate limits that hit at the worst moment, quiet model degradation that erodes output quality, price increases you cannot negotiate, and policy changes that can break your workflow overnight. The antidote is a three-layer stack — OpenRouter to route tasks across providers by cost and capability, a local model (Ollama or LM Studio) as an offline fallback for 30% of workflow, and a hard manual baseline for the three things no tool should ever own: audience understanding, offer positioning, and relationship building. The underlying principle: the real danger is not a Claude outage lasting a day, it is the slow atrophy of your ability to do the work without it.
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 – 01:08
01 · Hook and setup
Establishes credibility and frames the video as a real-talk departure from typical AI hype. States the three-part structure upfront.
01:08 – 02:21
02 · Failure Mode 1 — Rate limits
Rate limits hit mid-project even on max plans. Opus 4.6/4.7 usage adds up fast; the limits are worse than they were two months ago.
02:21 – 03:07
03 · Failure Mode 2 — Model degradation
Model quality quietly declines between versions. The supplier analogy: formula changes without warning after you have built your workflow around the old recipe.
03:07 – 03:58
04 · Failure Mode 3 — Price increases
Lock-in removes pricing leverage. If your operations depend on Claude, you pay whatever Anthropic charges.
03:58 – 04:22
05 · Failure Mode 4 — Policy changes
Usage policies can change overnight. A workflow built around one permitted use case can become non-compliant with a single update.
04:22 – 06:16
06 · Resilience Layer 1 — OpenRouter
Route tasks by type: complex work to Claude, repetitive tasks to cheaper models. Set a spend cap to prevent runaway bills during automatic provider-switching.
06:16 – 07:28
07 · Resilience Layer 2 — Going local
Ollama or LM Studio for on-device inference. Quality is lower than Claude 4.7 but enough to handle 30% of workflow during an outage.
07:28 – 09:10
08 · Resilience Layer 3 — Manual baseline
Three categories never fully automated: audience understanding, offer positioning, and client relationship building. Skill atrophy is real and faster than expected.
09:10 – 10:47
09 · Mindset shift and close
The real risk is not a one-day outage but losing the capacity to think and work without AI. Auto-steer driving analogy. CTA to community and free resilience checklist.
Lines worth screenshotting.
- The deeper your Claude dependency, the less leverage you have when Anthropic raises prices — you pay because the alternative is rebuilding everything.
- Model quality degrades without a formal announcement — the same prompts that worked in Claude 3.5 silently produce worse output in later versions.
- Rate limits do not respect your client deadlines; a max-plan subscription does not make you immune.
- OpenRouter lets you set a monthly spend cap so a model-switch during an outage cannot generate a surprise bill.
- Running even 30% of your workflow locally means a Claude outage does not fully block you — partial coverage is the goal, not parity.
- Skill atrophy is faster than people expect: one practitioner noticed their words-per-minute dropped after weeks of relying on speech-to-text.
- Audience understanding, offer positioning, and client relationships are the three categories where AI output is structurally weaker than the practitioner who owns the relationships.
- The supplier analogy applies directly: a supplier who changes the product formula without warning after you have built your workflow around the old spec is the same structural problem as a model update that breaks your prompts.
- Policy changes are the least discussed risk but the most binary: a single usage-policy update can make your core workflow non-compliant overnight.
- The people who win long-term are the ones who can switch between tools because they understand each one, not the ones who have automated their judgment away.
What to do before Claude owns your business.
WHAT TO LEARN
The dependency that makes AI feel indispensable is the same dependency that removes your negotiating power — on pricing, quality, and terms.
- Rate limits, model degradation, price increases, and policy changes are four distinct failure modes that compound: each one is manageable alone, but all four can hit a heavily dependent operation simultaneously.
- OpenRouter lets you route tasks by cost and complexity — use a premium model for strategic work and a cheap or free model for repetitive formatting tasks, with a hard monthly spend cap to prevent billing surprises.
- Running a local model via Ollama or LM Studio does not need to match Claude quality to be useful; even 30% workflow coverage during an outage means you are not fully blocked.
- Skill atrophy from over-automation is real and fast — practitioners who stopped typing noticed degraded typing speed within weeks; the same erosion happens to writing, analysis, and client judgment.
- Three categories of work resist AI augmentation not because AI is bad at them but because the value lives in who is doing them: knowing your own audience, positioning your own offer, and maintaining your own client relationships.
- The sharpest long-term competitive position is the ability to do the work with or without AI — not the ability to automate the most.
Terms worth knowing.
- OpenRouter
- A routing layer that sits in front of multiple AI model APIs and directs each request to a specified provider. A single API key gives access to Claude, GPT, Gemini, and open-source models, with routing rules set in system instructions.
- Ollama
- An open-source tool that downloads and runs large language models locally on a personal computer. No internet connection, API key, or usage fee required once the model is downloaded.
- LM Studio
- A desktop application with a graphical interface for running open-source language models locally. Similar function to Ollama but with a visual UI rather than a command-line interface.
- Manual baseline
- The set of skills a practitioner deliberately keeps un-automated so they retain the ability to do the work independently if AI access is interrupted or degraded.
- Rate limit
- A hard cap on how many API requests or tokens a user can consume in a given window. On Claude, hitting a rate limit blocks all further requests until the window resets or the user upgrades their plan.
Things they pointed at.
Lines you could clip.
00:36
“It's actually the fact that Claude is incredible which is causing it to be a massive problem.”
03:16
“Anthropic has us all by our necks, like, be honest.”
09:13
“The most dangerous thing isn't losing access to Claude for a day. It's losing the ability to think for yourself.”
08:10
“If you outsource every single thing — your thinking, your writing — and Claude goes down, you will actually feel you got dumber.”
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
00:00If you spent the last few months building your business around Cloud Code or with Cloud Code, I need you to stop and watch this video before you learn something the hard way like I did. Because to be truthful, there's something going on right now that not a lot of people will post a video on because it's just not going to get the amount of views and attention that people want.
00:15But my goal in this video is educating you and keeping it real so you could actually use AI to help your business and not harm it. And listen, I'm not gonna tell you Claude code is bad. It is incredible.
00:24I've made so many videos on it. It's helped me 10 x my following, my views. It's helped me get leads from my business and amazing things.
00:31But as I always say in my videos, the moment something changes and you're subscribed, I will keep you updated. And it's actually the fact that Claw is incredible which is causing it to be a massive problem. If you build your operations, your marketing, or anything of value to your business on top of Cloud Code or with it.
00:46Alright? Because what used to take two hours when we were doing it the manual way, the caveman way, now it takes nothing. It almost takes literally no time because Cloud Code has automated it, built a dashboard around it, or helped it improve in some way.
00:57But now you have Claude Code, this technology, which is not infinite and not always right, running a massive part of your business. And in this video, I'm gonna show you the four failure modes I've experienced and I've seen other people in the space experience as well. I'm gonna show you how to build AI resilience so you actually get around this and actually use this as a net positive thing.
01:14And then three manual anchors that I personally never recommend automate. Alright. So if you're using Cloud Code, these are the failures I've actually seen, and hopefully this doesn't happen to you.
01:22If you watch through, you could avoid these mistakes. So the first thing is just classic rate limits. Hitting your limit at the worst time when you need to get something done.
01:29You've all been there building stuff, and then boom. You get rate limited. You cannot use the service anymore unless you pay for it.
01:34And those four point six and four point seven Opus models, they really add up in terms of usage. Right? So some people can't do that.
01:40I think it's worth it, but sometimes you can't do that, and now you're stuck. You need to do this client work. You're in between it, and you just cannot finish.
01:46And we all know that recently, the limits have been awful. The glory days of building, I think, were, like, two, three months ago when the limits were amazing. 4.6 was out.
01:52It was great. Now the limits are terrible, and it feels like it's almost a glitchy to a certain extent. Alright.
01:57So that is one of the worst things. This will happen to you only a matter of time, and I'm talking about from someone that has a max plan. So if you're really building this, really relying on it, this will happen to you.
02:05And then you have to start in the middle of something. You might know exactly what you're doing, all your skills, your plugins, you can't access them, and it does become a mess. So that's the first thing.
02:12Not the biggest deal because you can still do the work yourself. It's just something to be aware of, especially if you give claud to your employees. Number two is that the model gets dumber, and we have seen this.
02:203.7 is not like it was with 3.6. It's getting dumber.
02:24Like, in my opinion, you could see everyone's moving to codex now. Like, people are switching back and forth. Everyone's going to the smartest model because these models do get dumber.
02:30Right? Now I don't wanna completely nerd out here with the capability of these models progressing, but I think there is a ceiling to what these can do. Alright?
02:37And we're sort of seeing it. Claude, four points that it's just not where it used to be. So when the models get dumber, not only does it get more frustrating, but the quality of output you're getting is terrible, and it's actually setting you back.
02:46Instead of just you doing the work, you're going to Claude, it's doing bad work, and you have to rewrite that work and then do the work. So you might as well just do the work anyways. Right?
02:53So that's the second point of failure is the model is just getting dumber and it's happened multiple times now. First happened to ChatGPT and OpenAI, now it's happening to Anthropic and Claude. Also, did put together this resilience check for you guys.
03:03It's a few questions that I think you and your team should answer about AI. And are you too overleveraged on Claude?
03:09I'll leave that for completely free down below if you wanna check that out and fill it out. It might help you get some clarity on the situation and how to use AI to actually help you instead of hurting. Now failure point number three is the price goes up, and this is unfortunate.
03:20Alright? , like, be honest. Right?
03:23Because if we're all using this and it's working and it's great, they could overnight double the price. Then what? You you have to pay because everything is there.
03:31You're over relied on that software, so you have to pay it. Right now, again, I think it'd still be worth it to an extent, but they have us, again, by the throat. They they could do whatever they want when it comes to the pricing.
03:40And we've seen it before. The models are only getting worse. The limits are only getting worse, and the prices are staying the same if not getting bumped up.
03:48Or, technically, it's getting more expensive if your limits are going down. You know what I'm saying? So that's another thing that you should be aware of.
03:52That's failure point number three. Then finally is policy changes. This is more of like a a risk to your industry.
03:57Anthropic usage can change. Like, they could rewrite stuff. Like, again, we are completely operating under whatever they say goes.
04:03So if they change something, they remove a model, they do anything, we're relying on that thing, or we're thinking that it's a secure part of our business. They could change it. We don't own that.
04:12Alright? Those are the four failure points. We have increased insensives, the model just getting dumber, getting rate limited, and then policy changes and just, again, realizing that Anthropic owns this.
04:21This is not ours. And, again, none of this is bad, but if it's one tool and you're over relied on that tool, it's never a good thing. So how do we overcome that?
04:27How do we build a resilient stack? How do we get around this? Now, the solution is not to stop using Cloud Code.
04:32Cloud is is amazing, and there's a lot of stuff you can do, especially if you've build your infrastructure properly. I talk about it on my channel all the time. Folder structures and local on your machine and all these different things, then you could plug and play, you're you're honestly good to go.
04:43As for people that haven't thought that through yet, this is your resilience stack. These are a few ways that we're getting around some of these pitfalls when it comes to overreliance on cloud code and stuff like that. So the first layer is what we like to call OpenRouter.
04:54This is a service. Essentially, OpenRouter does, it's a pretty neat tool. It routes your specific task to the model that you want.
05:01So you put in a prompt. It goes to Open Router. What Open Router will do is it'll pick the model for the task.
05:09So I might pick ChatGPT. I might pick Claude. I might pick Deepsea.
05:14Right? And the great thing about this is you can control that within the system instructions. So you could say, hey.
05:19Using the OpenRouter API key for the less intensive questions that I'm asking you, reroute to a free model or a very cheap model, at least. If they have intense questions, go to 4.7, go to ChatGPT, blah blah blah, all those different things. Right?
05:31And with the open source coming out, especially out out of China, like Kimi and all these different things, there's a lot of good options out there that won't be as good as Cloud or Gemini, but they will be good enough to at least get some stuff back. Alright? So OpenAuto is great.
05:43You create a free API key. I'm not getting paid by them to say this. Can literally take the API key, uh, give it to Cloud like you would do with anything else, and it'll help you sort of protect yourself.
05:51Right? Repetitive tax can be routed to cheaper models. Uh, when Cloud hits limits, you can then route to Gemini as well, but you do have to take into account the cost.
05:59And one hack for you is in your open router, say, limit spend, like, whatever, a $102,100 bucks a month. So by accident, if you switch, you don't, like, run up a crazy bill without you noticing. Alright.
06:07So that is layer number one. It's open router, and it's routing to correct models. This is an example of what you could do for strategy, creative writing, complex analysis, system design, coding.
06:15You could use Cloud. And then for formatting emails, pulling data, repetitive stuff, use a cheaper model. Next is going local.
06:20Alright. So I'm not technical. Had to learn all this stuff, and it's really cool how local works if you set it up right.
06:24You run AI models directly on your laptop with no Internet, no rate limits, no API cost. You use OLAMA for something like this or LMSudio, but you have to take into account these models will not be as star smart as 4.7, etcetera. So what you can do is have everything local on your computer since you're already working in folders on your computer, and then just use Olama to run something that is quote unquote dumb model, but it gets the repetitive stuff done for you.
06:45If for some reason, Cloud goes offline, which does a lot nowadays, or something else happens, at least all your data is there and you could still interact even if you have no Internet, etcetera. Even if you could offload 30% of your workflow in an online crisis, clauders out, your Internet's out, whatever, you could at least maneuver.
06:59You can manage. Alright? Going local, open source, all these different things, um, is a very easy way to do this.
07:05Again, the quality is not gonna be there. It's not even gonna be close to what you're used to with 4.7, but it's an alternative. I'm not saying these are perfect.
07:12I'm saying these are options. Alright? Gemma four came out from Google.
07:15Great model. Um, using those different things. I'll make a whole another video on that in another day, uh, for my business owners that wants you to understand how to use this stuff.
07:21But going local is always an interesting option, especially because a lot of you guys already have a great setup with Claude. Uh, it's just downloading Olaama and then routing a few things together. And, it's not as good as Claude.
07:30I will not say it is, but it works fine for some basic tasks if you need stuff to be done. Layer number three is the manual baseline. Now AI could do a lot of tasks for you, but it shouldn't do all tasks for you because you will lose your skill faster than you think.
07:42If you outsource every single thing I'm talking about, your thinking, your writing, your ability to format emails, your ability to do anything, and Claude goes down, you will actually feel you got dumber. Now an example of this for you is I use speech to text, um, applications, Glado, Whisper Flow, all those types of things.
08:00Right? One time, I just wasn't using that, and I had to type, and I my words per minute went down significantly. I almost forgot.
08:06Like, I was like, oh, man. I haven't typed in a while. Overreliance on anything is a bad thing, especially when it's knowledge work.
08:11Like, your knowledge work for your agency, your coaching, that is all you bring to the table. You're packaging it up and selling it. If you don't know what you're doing, you're cooked.
08:17Right? AI will not always be there. Know your stuff first instead of hard baselines where it's like, yeah.
08:21AI could do this. I should do this just to build a skill and get really good at something, then use AI to supplement that going down. Draw that line in the sand wherever that may be for you.
08:29There's a lot of things I still do manually because I need to build a skill that makes me money. Right? I'll always have it.
08:34So for me, that's understanding my audience and who I'm selling to. Claude will never get that as good as I do. My offer positioning as well.
08:40Claude will understand frameworks, but I won't really understand the stuff that's a little bit deeper. I don't have a word for it, but it's a little bit Like, why I'm charging what I charge, who I'm actually selling to, why I'm selling to those people, the problems I'm solving and positioning myself to solve for. I know that better than anyone, so I do that thinking for myself.
08:54And then, obviously, relationship building. Alright? All my Slack channels, my emails, I'm writing that stuff.
08:59I'm not having claud in my Slack. I think it's one of one one of the worst things you could do, especially in your client Slacks. Keep it you.
09:05Just build relationships as you would build a relationship, um, and keep it that way. That's a manual anchor 100%.
09:10And the mindset shift I really wanted you to just have in this video is the following. The most dangerous thing isn't losing access to Claude. It's losing your ability to think and reason for yourself.
09:18Because, again, if Claude goes down and Claude has all your knowledge and does everything for you, you just cooked yourself because you don't know what you're doing anymore. So that's something that you do have to take into account, um, when using Claude. That is the real risk.
09:28It's overreliance to the point or when that thing that you're over relying on gets taken away, you just don't know what you're doing. Good example of this, if you drive a car that has, like, automatic steering for you, you do that for years and years and years. One day, if that goes down and you have to drive, you might be a little nervous.
09:42Same thing sort of applies here. And I truly believe the people that win are the people that could switch between the tools, the people that really know their stuff no matter what AI is up or not, having the judgment when and when not to use AI, and then again keeping their skills sharp so they could always serve the market and always be valuable within their industry and what they're selling.
09:57So this is sort your takeaway here, the Claude code, pitfalls, rate limits, model getting dumber, pricing changes, and policy updates. Use open router, local, and manual baseline, so drawing lines in the sand of what AI does not do for you. Those are how to overcome this.
10:09And then what I want you to just think about is do not let yourself lose your ability to do your skill in your profession if you do not have AI. You need to always have that capability so you could always do things yourself, alright, just in case AI goes away or Claude Co. Shuts down for a week or something like that.
10:26Right? So that is the video I have for you. If you want more resources about how to use AI properly for your AI business, your marketing business, whatever it is, the whole community about AI to get leads and market your business, I will leave that down below.
10:38If you enjoyed this video, you will like this video right here. You use Claude responsibly, and I wish you all the best luck. I will talk to you soon.
10:46Take care.
The Hook
The bait, then the rug-pull.
The warning most Claude Code channels will not publish because it does not generate the views the hype videos do: four ways deep AI dependency can quietly break your business, and what a resilient stack actually looks like.
Named ideas worth stealing.
01:08list
Four Failure Modes of Claude Dependency
- Rate limits
- Model degradation
- Price increases
- Policy changes
Four ways building core business operations around a single AI vendor creates compounding risk.
Steal forRisk audit slide when pitching a multi-model strategy to a client or team
04:22list
Three-Layer Resilience Stack
- OpenRouter (model routing)
- Going local (Ollama/LM Studio)
- Manual baseline (skills never automated)
A practical three-layer hedge against vendor dependency: route intelligently, run a local fallback, protect irreplaceable human judgment.
Steal forAny AI operations setup conversation — positions you as thinking about durability, not just capability
07:28list
Manual Anchor Categories
- Audience understanding
- Offer positioning
- Relationship building
Three categories of knowledge work that should remain human-owned because AI cannot replicate the relational context that makes them valuable.
Steal forFraming conversations with clients about where AI augments vs. where it should not replace
02:21concept
The Supplier Analogy
An AI model provider is like a supplier who can change their product formula without notice after you have already built your workflow around the old spec.
Steal forExplaining model degradation risk to non-technical business owners
How they asked for the click.
VERBAL ASK
10:20link
“If you want more resources about how to use AI properly for your business, I will leave that down below.”
Soft double CTA at the end: free resilience checklist on Gumroad plus paid Skool community. Delivered verbally after the mindset close.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
