From Custom GPTs to Skills: What Is Changing in ChatGPT

17.11.2026 from CA Redaktion | Information Management
Person steuert digitale Workflows und Prozessdiagramme an einem Touchscreen als Symbol für den Wandel von GPTs zu Skills.

At a Glance: What’s Changing in the Way We Work with ChatGPT

  • Custom GPTs are being phased out: OpenAI is moving to plugins and skills. Existing Custom GPTs can be migrated.
  • Using skills: Define recurring workflows, subject-matter rules, and quality standards once and reuse them every time.
  • Standardizing controlling processes: Make variance analyses, management reporting, and forecast reviews more consistent and efficient.
  • Working across systems: In principle, skills can also be used in other AI systems. Adjustments may be needed depending on the tools used.

If you want to become confident using AI in controlling, you can find the right seminars here.

For a long time, Custom GPTs offered a simple way to tailor ChatGPT to specific tasks. Controlling professionals could add their own instructions, knowledge, and capabilities to create, for example, an assistant for reporting, data analysis, or internal processes.

OpenAI is now fundamentally changing this approach, as Custom GPTs are being phased out. Going forward, existing workflows are expected to be handled through plugins instead. What are known as skills play a central role in this transition. OpenAI has already announced plans to provide a migration path from existing Custom GPTs to plugins.

What Are Skills?

Skills are reusable workflows that tell ChatGPT how a specific task should be performed.

For example, a skill can specify:

  • what information is needed,
  • which steps should be carried out,
  • which subject-matter rules apply,
  • how the output should be structured,
  • which quality checks should be performed,
  • and, if needed, include scripts, templates, or reference files.

This turns a good prompt into a standardized process.

In controlling, for example, you could create a skill for a plan-versus-actual variance analysis. Instead of describing in detail each time which price, volume, and mix effects need to be calculated, you define that logic once in the skill and then reuse it.

What Is the Difference Between a Skill and a Plugin?

Skills sind Bestandteile von Plugins.

A plugin can bring together different capabilities, including:

  • Skills – which define instructions and reusable workflows.
  • Apps – which connect to external services and data.
  • Other extensions – for example, for interacting with files or applications.

A plugin can consist entirely of skills or combine skills with externally connected applications.

Why Are Skills Valuable?

The major advantage is that recurring ways of working can be standardized.

For example, a company can define how to write commentary on a monthly report, perform a variance analysis, or review a business case. That approach no longer needs to be explained again with every request.

As a result, skills enable:

  • Consistent results
  • Reusable subject-matter logic
  • Uniform quality standards
  • Easier sharing of best practices
  • A combination of natural language, subject-matter expertise, and programming code

This is particularly relevant for those of us in controlling. Many tasks follow recurring rules, while interpreting the results still requires human judgment and subject-matter expertise.

Can Skills Be Transferred Between Different AI Systems?

At the core of a skill is typically a file called SKILL.md. It contains the workflow instructions in Markdown, a simple text-based format. OpenAI explicitly describes this format as portable and as an open or standardized approach that can also be used in other AI applications and platforms.

This makes AI workflows less dependent on a single user interface. In principle, instructions, process logic, and quality rules can be transferred, version-controlled, and reused in other systems more easily.

However, this does not automatically mean that every skill will work with every AI provider without modification. If a skill contains Python scripts, specialized tools, external apps, or provider-specific features, adjustments may be necessary.

The standardized structure therefore makes portability significantly easier, but it does not guarantee complete vendor independence.

What Does This Mean for Existing Custom GPTs?

Existing Custom GPTs are not automatically lost. They can be migrated to a plugin through the interface in just a few clicks. However, this is also a good opportunity to reassess the Custom GPTs created so far: Which features do we no longer need, what is redundant, and where can we automate additional steps?

Typical examples of skills include:

  • Reporting and commentary guidelines
  • Data analysis
  • Review and approval processes
  • Research workflows
  • Company policies
  • Templates and quality checks

Going forward, these elements can be transferred into skills and plugins.

Based on the information available as of early October, Custom GPTs are scheduled to be discontinued in affected Enterprise workspaces on December 11, 2026. For other plans, timing and availability may vary depending on the account or workspace. After that, Custom GPTs will no longer be accessible.

Conclusion: From Individual Assistants to Reusable Workflows

With skills, the focus of automation and processes is shifting. Until now, the emphasis was often on the individual AI assistant: “I’ll build my own GPT.” Going forward, the more important question will be: “Which process do I want to standardize and make reusable?”

This is an interesting development for companies. Subject-matter knowledge, process logic, and quality standards can be documented in a structured way and used directly by AI systems.

My Tip:

Anyone currently using Custom GPTs or extensive prompt templates should identify the workflows they contain and assess which of them could be implemented as skills.

In controlling, recurring tasks such as variance analyses, management commentary, forecast reviews, data validation, or business plan analyses are particularly well suited. In this way, individual prompts gradually become a reusable, AI-supported way of working.

Florian Bliefert

Author

Florian Bliefert

Manager at CA controller akademie

Florian Bliefert is an expert and the key point of contact for all topics related to AI and RPA in Controlling.

share the post