The Origin of the ‘Industrial Waste’ Label for Skills

Recently, a piece titled “So-called Skills are Just Industrial Waste in the AI Era” by Titanium Media stirred significant discussion in the AI community and developer circles. The author sharply criticizes the current push by major platforms for ‘skill stores’ and ‘plugin marketplaces’, arguing that they represent a commercial strategy aimed at turning AI’s dynamic intelligence into static plugins.

The core criticisms of the article focus on several points:

• Technical Superficiality: Skills are essentially just ‘glue code’ along with API documentation (JSON Schema), and their technical value has diminished to nearly zero in the face of large models’ ability to auto-generate code.
• The Myth of Universality: In the complex and messy reality of business, internal systems, data standards, and permission barriers vary widely. A standardized skill (like ‘querying financial reports’) cannot meet the specific needs of different enterprises and becomes ineffective outside its preset environment.
• Replacement by Protocols: The article argues that standardized underlying protocols, like MCP (Model Context Protocol), are the future. They provide a unified ‘highway’ that allows AI to generate calling instructions dynamically based on specific scenarios, rendering pre-packaged static skills obsolete.

The author’s conclusion is striking: developers obsessed with packaging ’exclusive skills’ and platforms creating ‘AI skill supermarkets’ are heading towards mediocrity, as they are constraining AI’s ‘ubiquitous fluid intelligence’ with ‘pre-industrial era thinking’.

Is the Critique Overly Extreme?

This critique acts like a scalpel, precisely dissecting the bubbles and restlessness present in the current AI application ecosystem. The criticism of ‘skill inflation’ and the platform’s ‘rental’ mentality indeed warrants caution across the industry. As large models can understand natural language and generate code, simply packaging API calls as ‘cool skills’ raises long-term value concerns.

However, dismissing skills outright as ‘industrial waste’ is an overly absolute and radical assertion that confuses the ’end of technology’ with the ‘development process’, underestimating the scaffolding value of skills in the popularization and implementation of AI.

1. Skills as Scaffolding, Not Buildings
   For most non-professional developers and even business personnel unfamiliar with coding, skills (or similar reusable AI tool packages) serve as a shortcut to access and utilize AI capabilities. They lower the entry barrier, allowing users to complete specific tasks without understanding complex MCP protocols or writing prompts themselves. This is akin to how driving school cars (even manual ones) remain necessary learning tools before the widespread adoption of autonomous driving. Skills act as a productivity lever during this transitional period.

2. Protocols and Skills: A Relationship of Roads and Vehicles
   The MCP protocol praised by the author addresses the issue of safely and standardly connecting various resources, building the ‘highway’. In contrast, skills (at least in their evolved form) address ‘what to do after connecting and how to do it more efficiently’, functioning as the ‘vehicles’ on the road. No matter how good the road is, various vehicles are needed to transport value. The future direction is not to eliminate ‘vehicles’ but to make their production and use extremely cheap, dynamic, and intelligent.

3. The True Value Lies in Refined Knowledge, Not Glue
   The article criticizes generic skills like ‘checking the weather’ or ‘searching the web’. However, truly valuable skills in the industry are those that deeply encapsulate specific domain knowledge, workflows, and business logic. For instance, a ‘contract review skill’ that integrates legal knowledge, case libraries, and document drafting logic has value far beyond just calling APIs; it embodies ‘structured business cognition’. This is not industrial waste but a refined digital asset.

The Future of Skills: From Stockpiling to Instant Generation and Integration

Thus, we need not mourn the death of skills but rather redefine their evolutionary direction. Skills will not disappear, but their forms and production methods will undergo fundamental changes:

• From Pre-fabricated to Instant Generation: Future AI agents will dynamically combine or generate necessary micro-skill modules based on real-time tasks, disposable after use, rather than relying on a pre-installed, rigid skill store.
• From Functional Description to Cognitive Packaging: High-value skills will no longer merely describe ‘how to do’ (API calls) but will encapsulate ‘under what circumstances, why, and how to think before doing’, incorporating decision logic and domain cognition.
• Deep Coupling with Protocol Layers: Skills will be deeply integrated with underlying protocols like MCP, becoming a flexible, executable ‘intelligent instruction set’ rather than isolated applications.

Conclusion

“So-called Skills are Just Industrial Waste in the AI Era” serves as a timely wake-up call, reminding us not to repeat the platform hegemony fantasies of the old era in the face of new technological waves, nor to indulge in packaging ’technological MSG’.

However, we must also avoid the trap of ’technological radicalism’, dismissing necessary tools in the development process. Skills, as an important bridge for democratizing AI capabilities, still have a vital historical mission ahead. Their endpoint is not a garbage heap but an evolution into more agile, intelligent, and seamlessly integrated ‘intellectual components’ with underlying protocols.

What we need to discard is the ‘rental’ mentality and the complacency of ‘glue code’, while embracing a deep reverence for ‘real business problems’ and ‘dynamic intelligence’. This is the path that developers and entrepreneurs in the AI era should take.