Introduction

For the past two years, AI users focused on one skill above all else: writing better prompts.

Today, that advantage is disappearing.

With the release of Claude Sonnet 5, Claude Fable 5, and updates to Opus 4.8, Anthropic has fundamentally changed how Claude works. The biggest performance improvements no longer come from prompt wording—they come from choosing the right model, configuring effort levels, and matching AI behavior to the task.

Prompt engineering isn’t dead—but it has become just one small part of a much larger operating system.

The Three Big Changes in Claude 5

1. Configuration Matters More Than Prompting

Previously, users spent time refining prompts:

adding roleplay
writing lengthy instructions
using anti-laziness tricks
creating elaborate templates

Today, Claude performs best when you configure:

the correct model
reasoning effort
thinking mode
task specification

Instead of asking:

“How should I phrase this?”

The better question is:

“Which Claude model should solve this?”

This marks the shift from Prompt Engineering to AI System Design.

2. Claude Has Become Extremely Literal

Earlier AI models often tried to “help” by filling gaps.

Claude 5 behaves differently.

It follows instructions almost exactly as written.

That means:

Good

If you specify:

500 words
bullet points
executive tone
markdown format

Claude usually delivers exactly that.

Bad

Older prompting habits now reduce quality.

For example:

Instead of writing:

Don’t be vague.

You should write:

Provide three detailed recommendations with examples.

Positive instructions outperform negative instructions because Claude treats every sentence literally.

The lesson:

Tell Claude what to do—not what to avoid.

3. Intelligence Is Now Adjustable

Claude now introduces something similar to a reasoning budget.

Instead of every request receiving maximum thinking, users can choose different effort levels.

Think of it as selecting processing power.

Low effort:

quick answers
lower cost
faster response

High effort:

deeper reasoning
stronger analysis
higher token usage

Not every task deserves maximum intelligence.

Simple emails don’t.

Business strategy probably does.

The new skill becomes knowing when extra thinking creates value.

Prompt Engineering Is Becoming Prompt Architecture

Instead of creating one universal prompt, organizations now need systems that automatically decide:

Which model to use
How much reasoning to apply
When to spend more tokens
When speed matters more than perfection

This resembles cloud computing.

Developers don’t use the same server for every workload.

Likewise, AI users shouldn’t use the same model for every task.

The Seven New Prompting Rules

According to Joerg Storm, modern Claude prompting follows several new principles.

1. Be Explicit

Avoid ambiguity.

Specify:

audience
objective
output
format
length

Claude rewards precision.

2. Describe the Outcome

Rather than describing the process, describe the result.

Instead of:

Think carefully.

Say:

Produce a board-ready executive summary with three strategic recommendations.

3. Replace Negative Instructions

Don’t write:

Don’t repeat yourself.

Instead write:

Each recommendation should introduce a unique insight.

Positive language creates cleaner outputs.

4. Match Model to Difficulty

Not every task needs the most powerful model.

Routine work can use faster, cheaper models.

Complex reasoning deserves premium models.

5. Use Reasoning Only When Necessary

Deep thinking consumes tokens.

Reserve higher effort for:

research
planning
coding
strategy
scientific reasoning

Use lighter reasoning for routine communication.

6. Build Structured Specifications

Good prompts increasingly resemble product specifications.

Include:

objective
constraints
audience
expected format
evaluation criteria

The prompt becomes a design document.

7. Continuously Improve

Instead of storing static prompt libraries, organizations should maintain living AI workflows that evolve alongside new models.

The Economics of AI Effort

One of the most important ideas from the newsletter is effort economics.

Previously:

Higher-quality AI mostly meant using a larger model.

Now:

Quality depends on both:

model selection
reasoning effort

This introduces a new optimization problem.

Organizations must balance:

response quality
speed
token cost
business value

AI usage becomes a resource allocation exercise rather than simply prompt writing.

Why Old Prompt Libraries Are Becoming Obsolete

Many organizations spent months building prompt databases.

Those prompts were designed for older models.

Claude 5 changes the assumptions.

Common legacy techniques that may now reduce quality include:

excessive anti-hallucination wording
repeated “think step by step”
multiple warning statements
defensive prompt engineering
unnecessary verbosity

The recommendation isn’t to throw prompts away.

It’s to simplify them.

Modern Claude already understands much of what older prompts tried to force.

AI Teams Need a New Operating Model

Instead of teaching everyone prompt engineering, organizations should document:

preferred model for each workflow
default reasoning level
acceptable cost
quality expectations
output standards

Think of AI as infrastructure—not a chatbot.

This creates consistency while reducing experimentation costs.

A Practical Rollout Plan

Joerg Storm suggests treating the transition as an organizational upgrade.

Week 1

Audit existing prompts.

Identify outdated prompting habits.

Week 2

Categorize recurring tasks.

Examples:

writing
coding
research
analysis
customer support

Week 3

Assign:

best Claude model
reasoning level
expected response format

for every task category.

Week 4

Measure:

response quality
token costs
execution speed
user satisfaction

Then continuously refine the system.

Key Takeaways

The AI landscape has shifted from prompt writing to AI configuration.

Success now depends less on clever wording and more on designing intelligent workflows that combine the right model, reasoning effort, and task specification. Organizations that continue relying on old prompt libraries risk lower-quality outputs and higher costs, while those that adopt model-aware workflows will gain better performance, greater consistency, and improved efficiency.

Claude 5 signals a broader industry trend: future AI expertise will revolve around system architecture, orchestration, and optimization, rather than crafting increasingly elaborate prompts.

Final Thoughts

Prompt engineering isn’t disappearing—it’s evolving.

The competitive advantage is no longer the person who writes the longest or most sophisticated prompt. It’s the person who understands which model to use, how much intelligence the task requires, and how to design an AI workflow that balances quality, speed, and cost.

In the era of Claude 5, configuration has become the new prompt.

The Claude Operating System: Why Prompt Engineering Is No Longer Your Competitive Advantage