-
Stop Using Claude Like This: The AI Mistake That Could Cost You Thousands
Inspired by Digital Storm Weekly #171 by Dr. Joerg Storm
For the past two years, the AI conversation has largely revolved around one question:
“Which model is the smartest?”
Today, that’s the wrong question.
The better question is:
“Which model should I use for this specific task?”
This shift has become more important than ever with Anthropic announcing that Claude Fable 5, its most advanced reasoning model, is moving from being included in subscription plans to a pay-per-token pricing model. While many users see this as a price increase, the real lesson isn’t about cost—it’s about developing smarter AI workflows.
The future of AI productivity won’t belong to people who always use the most powerful model. It will belong to those who know when to use it.
The Biggest AI Mistake Most Users Make
Imagine hiring the senior partner of a consulting firm to spend the day adjusting fonts in PowerPoint.
It sounds absurd.
Yet that’s exactly how many professionals use AI today.
Every task—whether it’s summarizing an article, writing an email, brainstorming ideas, or solving a complex business problem—is sent to the most powerful model available.
While this may produce good results, it is also one of the fastest ways to increase AI costs without gaining proportional value.
Not every task requires frontier-level reasoning.
Some tasks require intelligence.
Others simply require speed and consistency.
Knowing the difference is becoming one of the most valuable AI skills in the workplace.
Think of AI Models Like a Team
One of the best analogies from the newsletter compares AI models to a consulting firm.
Imagine your organization has four employees:
-
Haiku – the fast intern who handles repetitive, straightforward work.
-
Sonnet – the experienced professional who manages everyday tasks efficiently.
-
Opus – the senior specialist who tackles challenging projects.
-
Fable – the strategic partner who is brought in only when judgment, planning, and deep reasoning truly matter.
Would you ask the strategic partner to rename files or reformat documents?
Probably not.
Instead, you’d involve them in defining the strategy, making high-stakes decisions, or solving ambiguous problems. Once the direction is clear, the rest of the team executes the work.
AI should be used exactly the same way.
Why AI Costs Increase Faster Than You Think
Many people assume they only pay for the latest message they send.
That’s not how modern language models work.
Every time you send a prompt, the model receives:
-
Your newest message
-
Every previous message in the conversation
-
Previous AI responses
-
Uploaded files
-
Instructions
-
Context from earlier turns
In other words, every new prompt asks the model to reread the entire conversation before generating a response.
This means the longer a conversation continues, the more tokens are processed—and the more expensive each interaction becomes.
A discussion that starts out inexpensive can quietly become costly after dozens of exchanges.
This is why starting a fresh conversation for a new project isn’t just good organization—it’s also a smart cost-management strategy.
Stop Using Frontier Models as Your Daily Driver
Frontier AI models such as Claude Fable 5 are designed for problems that require sustained reasoning, planning, and decision-making.
They excel at:
-
Strategic planning
-
Multi-step analysis
-
Complex coding projects
-
Business decisions
-
Research synthesis
-
Long-horizon AI agents
-
Challenging architectural design
However, they are unnecessary for many everyday activities.
Routine tasks like:
-
Formatting documents
-
Writing standard emails
-
Summarizing meeting notes
-
Creating social media captions
-
Editing text
-
Translating content
can often be completed just as effectively using smaller, more affordable models.
The goal isn’t to use the smartest model all the time.
The goal is to use the smartest model only when its intelligence creates measurable value.
The Routing Rule Every AI User Should Learn
One of the most practical lessons from the newsletter is a simple routing strategy.
Step 1: Use the frontier model for thinking.
Ask it to:
-
Analyze the situation
-
Challenge assumptions
-
Build the strategy
-
Identify risks
-
Create the roadmap
Step 2: Switch to a more efficient model.
Once the strategy is complete, let a lower-cost model handle:
-
Writing
-
Editing
-
Formatting
-
Expansion
-
Documentation
-
Follow-up tasks
This simple habit can significantly reduce AI costs while maintaining high-quality output.
Give AI Goals, Not Tiny Tasks
Older AI models often required users to break work into small instructions.
Today’s frontier models perform best when they understand the overall objective.
Instead of saying:
“Write an email.”
Provide context like:
“The client has missed two invoice payments, communication has stopped for three weeks, and I want to recover payment while preserving the business relationship. Review the conversation history, identify the best approach, draft the email, and ask me any clarifying questions before proceeding.”
The second prompt allows the AI to think strategically instead of mechanically completing a task.
The difference in quality is substantial.
Four Simple Prompting Habits That Improve Results
The newsletter also highlights four techniques that consistently improve AI responses.
1. Ask for the answer first.
Rather than reading pages of reasoning, request the conclusion before the explanation.
2. Require evidence.
Instruct the model to distinguish verified facts from assumptions and clearly identify any uncertainty.
3. Separate thinking from execution.
Tell the AI whether you want advice, brainstorming, or complete execution.
4. Avoid analysis paralysis.
When sufficient information exists, ask the model to make a recommendation instead of presenting endless possibilities.
These small adjustments produce clearer, more actionable outputs while reducing unnecessary back-and-forth.
AI Spending Needs Governance
As AI agents become capable of working autonomously for hours, usage-based pricing introduces a new challenge.
An unattended AI workflow can continue consuming tokens—and therefore money—long after you’ve stepped away.
This is why organizations should implement:
-
Monthly spending limits
-
Daily usage caps
-
Per-user budgets
-
Approval workflows for expensive models
-
Model routing policies
Managing AI usage is beginning to resemble cloud infrastructure management. Governance is no longer optional—it is essential.
A Simple Framework for Choosing the Right Model
Before starting any AI task, ask yourself two questions:
How ambiguous is the problem?
Is it straightforward, or does it require deep reasoning?
What happens if the AI is wrong?
Can mistakes be corrected easily, or would they have significant business consequences?
Tasks with low ambiguity and low consequences can be delegated to fast, inexpensive models.
Tasks that are both highly ambiguous and high-risk deserve your most capable AI model.
Everything else falls somewhere in between.
AI Success Isn’t About Bigger Models—It’s About Better Workflows
One of the most powerful messages from this week’s newsletter is that AI maturity isn’t measured by using the newest model.
It’s measured by building systems.
The organizations that succeed with AI won’t necessarily have access to better technology than everyone else.
They’ll simply know:
-
which model to use,
-
when to use it,
-
how to control costs,
-
how to design repeatable workflows,
-
and how to combine human judgment with AI capabilities effectively.
As AI continues to evolve, prompt engineering alone won’t be enough.
The next competitive advantage is AI orchestration—treating different models as specialized members of a team rather than a single tool.
Those who master that mindset will achieve better results, lower costs, and far greater scalability.
The future of AI isn’t about always choosing the smartest model. It’s about consistently choosing the right one.
drstorm.substack.com
Stop Using Claude Like This. #171
One routing mistake can turn powerful agents into runaway AI spend
-

