Crossroad AI

Appearance

Session 6: Agent Skills & Global AI Trends Q3 2025 ​

Date: October 31, 2025
Duration: 2 hours
Participants: WiseTech Global Team

Session Recording ​

🎥 Watch Full Session

Recording Highlights:

📋 Overview ​

This comprehensive session explored the current state of AI in Q3 2025, diving deep into Agent Skills implementation, global AI market trends, and fundamental AI theory. We examined how companies are adopting agent technologies, analyzed the difference between MCP and Agent Skills, and discussed Kevin Kelly's vision of the Agent Economy. The session also included Eden Ye's insightful discussion on AI theory evolution, from supervised learning to modern large language models, and addressed fundamental questions about AI capabilities and limitations.

🎯 Key Topics Covered ​

1. Global AI Trends Q3 2025 ​

Q3 2025 AI Market Overview

The third quarter of 2025 marked a significant shift in the AI landscape, with intelligence and tool-calling capabilities becoming core competitive advantages across major tech companies.

AI Insights Dashboard

Key Market Trends:

  • Tool Calling Revolution: Browser and operating system level tool integration driving AI application diversification and automation
  • Major Players: Google, OpenAI, Meta leading in language, voice, image, and video models
  • Chinese AI Ecosystem: Baidu, Tencent, Alibaba, and ByteDance achieving breakthroughs in technical architecture and model performance
  • Hardware Evolution: NVIDIA Blackwell entering mass production, enabling local large model deployment

Industry Reports:

Notable Developments:

  • Enhanced agent capabilities across all major platforms
  • Shift from traditional chatbots to autonomous tool-calling agents
  • MCP (Model Context Protocol) servers enabling standardized tool integration
  • Agent Skills emerging as new paradigm for AI application development

2. Agent Skills: The Next Generation of AI Prompts ​

Agent Skills Overview

What are Agent Skills?

Agent Skills transform traditional AI prompts into executable, reusable skill modules that can be dynamically invoked by AI systems. They represent the evolution from static prompt engineering to dynamic, tool-enabled AI capabilities.

Core Concepts:

  • Natural Language Operations: Users describe tasks in plain language
  • Code Snippet Execution: Automated execution of predefined code modules
  • Task-Specific Modules: Specialized skills for document processing, data extraction, API calls, etc.
  • Dynamic Invocation: Skills are loaded on-demand based on task requirements

Agent Skills vs MCP:

AspectAgent SkillsMCP (Model Context Protocol)
DeploymentNo persistent backend requiredRequires continuous server operation
InvocationOn-demand, dynamic loadingAlways available, persistent connection
ComplexityLightweight, simple configurationMore robust, enterprise-grade
Use CasesSmall applications, flexible scenariosSystem integrations, large AI-native apps
DevelopmentAccessible to non-developersRequires technical expertise
FutureCo-existence with MCPCo-existence with Agent Skills

Key Resources:

3. Agent Skills Implementation Examples ​

Case Study 1: Taiwan Tech Learning Platform

The Taiwanese technology learning website tenten.co successfully implemented Agent Skills for educational content delivery:

  • Use Case: Interactive AI tutoring with code execution
  • Implementation Time: Minimal setup required
  • Benefits: Dynamic skill loading, context-aware responses, code demonstration capabilities

Case Study 2: Enterprise Customer Service System

Cloud Scale Intelligence built a complete enterprise customer service system using Agent Skills in just 40 minutes:

Key Implementation Principles:

  1. Modular Design: Each skill is self-contained with clear documentation
  2. Standardized Format: Consistent folder structure and naming conventions
  3. Task Routing: Automatic skill selection based on user intent
  4. Progressive Enhancement: Start simple, add complexity as needed

4. WiseTech AI Initiatives & AI Insights ​

Internal AI Prompt Standardization

WiseTech Global is adapting internal AI prompts to the latest Agent Skills standards, enabling:

  • Automated Loading: On-demand skill invocation
  • Cross-Department Reusability: Shared skill libraries
  • Product Integration: AI-powered features in CargoWise and other products
  • Documentation Automation: Intelligent document generation
  • Macro Commands: Complex workflow automation

AI Insights Publication

  • Frequency: Published every two weeks
  • Purpose: Share company AI developments and industry trends
  • Contribution: Employees encouraged to submit content and insights
  • Impact: Promotes knowledge sharing and AI adoption across teams

OCR in Production:

WiseTech has successfully deployed OCR (Optical Character Recognition) for:

  • Financial Invoice Recognition: 98.5% accuracy
  • Customs Declaration Processing: High accuracy with human-in-the-loop verification
  • Human-AI Collaboration: AI handles routine cases, humans handle edge cases

5. Agnes AI: Agentic AI Product Success Story ​

Company Background:

Agnes AI is a Singapore-based startup that has achieved remarkable success with their agent-based AI application.

Success Metrics:

Product Philosophy:

Rather than building a general-purpose agent, Agnes focused on specific, high-value use cases:

  • Vertical Integration: Deep integration with specific industries
  • User Experience First: Prioritizing usability over feature breadth
  • Practical Value: Solving real user problems effectively
  • Rapid Iteration: Quick feedback loops and continuous improvement

Key Lessons:

  • General-purpose agents face adoption challenges
  • Specialized agents with clear value propositions succeed
  • User experience matters more than technical sophistication
  • Market fit is crucial for agent application success

6. Kevin Kelly on the Agent Economy ​

Context:

On October 18th, Damien Li and Craig attended Kevin Kelly's interview in Shanghai, where K.K. discussed his vision for the Agent Economy with a prominent Chinese AI entrepreneur.

📄 Kevin Kelly Agent Economy Interview (PDF)

Key Insights from Kevin Kelly:

Definition of Agents ​

"An agent is an entity that does work on my behalf. It may not act directly, but it serves me and works for my benefit."

  • Agents don't have to be autonomous, but autonomy increases their power
  • We're moving toward sub-agents that work for other agents
  • An ecosystem with thousands of agent "species" with different capabilities

General Intelligence vs. Specialized Intelligence ​

K.K. challenges the concept of AGI (Artificial General Intelligence):

"I think the concept of 'general intelligence' is misguided. It's possible to have a general intelligence, but I don't think it would be very useful."

The Trade-off Principle:

  • Every engineering system involves trade-offs
  • Optimizing one aspect means sacrificing another
  • General-purpose AI is like a Swiss Army knife - mediocre at everything
  • Specialized AI agents will be superior for specific tasks

Intelligence is Multi-Dimensional:

  • Intelligence isn't a single scale or ladder
  • It's a high-dimensional space with infinite varieties
  • Human intelligence is just one point in this vast space
  • AI will occupy different positions with different strengths

AI is Different from Humans ​

"They're 'aliens.' They're over here; we're over there. There's a lot more out there."

Key Differences:

  • AI doesn't think like humans (and that's the advantage)
  • AI excels at tasks humans struggle with (memory, consistency, tireless operation)
  • AI struggles with tasks humans find easy (common sense, creativity, judgment)
  • Current LLMs are trained on human knowledge, not reality itself

The Need for Physical World Models:

  • Current LLMs lack true world understanding
  • They model knowledge about the world, not the world itself
  • Future: Large Physics Models and Large Chemical Models
  • Robots need world-grounded intelligence, not just text-based knowledge

AI and Productivity ​

"AI is very good at efficiency and productivity. Humans are very bad at efficiency and productivity."

Division of Labor:

  • Tasks requiring efficiency → Assign to AI
  • Tasks requiring innovation → Keep with humans
  • Innovation is inherently inefficient (mistakes, failures, pivots)
  • Science requires inefficiency to discover new knowledge

The 95/5 Rule:

  • 95-98% of AI will be invisible background infrastructure
  • Only 2-5% will have direct human interaction
  • Success means AI becomes invisible (like plumbing or electricity)
  • Human-facing AI should adopt human-like interfaces for comfort

Currency in the Agent Economy ​

Third Generation Stablecoins:

  • Crypto has been "looking for a job" beyond wealth storage
  • Bitcoin and early crypto too expensive and volatile for micropayments
  • New stablecoin models (e.g., Stripe's Tempo) enable:
    • Secure payments for tiny amounts
    • Embedded trust using blockchain
    • Cheap, fast transactions between agents

"I believe that will be the money for agents."

Examples:

  • Stripe's Tempo
  • Circle's stablecoin variants
  • Chinese companies developing "agent money"

The Future of Companies ​

More Companies, Including Giants:

  • Setting up companies will become as easy as clicking a button
  • Millions of new companies will emerge
  • One-person startups will thrive
  • But million-employee companies will also grow

Why Large Companies Can Grow Larger:

  • The limiting factor is managing complexity, not ambition
  • AI agents solve the "matching problem" - finding the right talent at the right time
  • Better collaboration tools increase efficiency at scale
  • With better matchmaking, companies can keep growing effectively

The AI Bubble ​

"Right now there is probably an AI investment bubble... Many early investors will likely lose money, but society as a whole will benefit later."

Historical Parallel:

  • Similar to the internet bubble 25 years ago
  • Massive investment in infrastructure (data centers, chips)
  • Early investors may not profit
  • Later investors and society will benefit
  • Overall positive for human progress

AI and Employment ​

Current Evidence:

  • After 3 years of operational AI, very few jobs lost
  • Employees become ~25% more productive
  • Occupations predicted to be devastated (radiologists, translators) still hiring
  • AI help desks expand markets rather than replace workers
  • No evidence of harm to lowest-skilled workers

Market Expansion Effect:

  • 24/7 AI support creates new possibilities
  • Companies that couldn't afford support now can
  • Human oversight still needed
  • Net effect: job expansion, not elimination

Optimism as a Choice ​

"I choose optimism because all the good things in this room and this city were created by optimists in the past."

K.K.'s Philosophy:

  • Optimism is a choice, not a personality trait
  • The world was built by past optimists
  • The future will be built by today's optimists
  • To shape the future, you need optimism
  • Vision + belief that it's possible = optimism

You're Not Late:

"You're not late. This is day one. In 2049, people will look back at 2025 and say, 'You didn't even have AI then.'"

7. AI Theory Evolution: From Supervised Learning to LLMs ​

Presented by: Eden Ye

Eden provided a comprehensive and accessible overview of AI's theoretical foundations and evolution, addressing fundamental questions about AI capabilities and limitations.

AI Theory Evolution

The Journey of AI Technology ​

Historical Progression:

  1. Supervised Learning → Foundation of modern AI
  2. Loss Functions → Measuring and improving performance
  3. Backpropagation → Training neural networks efficiently
  4. Convolutional Layers (CNN) → Visual understanding
  5. Self-Attention Mechanism → Contextual understanding
  6. Transformer Architecture → Parallel processing breakthrough
  7. Large Language Models → Emergent intelligence

Key Innovation Drivers:

Each technological leap solved specific limitations:

  • CNNs addressed spatial pattern recognition
  • Attention mechanisms enabled long-range dependencies
  • Transformers unlocked parallel training at scale
  • Scaling laws revealed emergent capabilities

Generalization Ability and Emergent Intelligence ​

Why Deep Learning Generalizes:

Eden explained generalization through symbolic reasoning theory and phase transition analogies:

Symbol Reasoning Theory:

  • Models learn to construct internal symbolic representations
  • These symbols capture abstract concepts beyond training data
  • Enables reasoning about unseen scenarios

Phase Transition Analogy:

  • Similar to water freezing into ice
  • At critical scale (model size + data), qualitative changes emerge
  • New capabilities appear that weren't explicitly trained

Factors Driving Emergence:

  • Model Complexity: Number of parameters and layers
  • Context Window: Ability to process long sequences
  • Training Data Scale: Diverse, high-quality datasets
  • Compute Resources: Intensive training regimes

Emergent Capabilities:

  • Few-shot learning
  • Chain-of-thought reasoning
  • Task generalization
  • Cross-domain transfer

Real-World Application Challenges ​

Laboratory vs. Reality Gap:

Eden used autonomous driving as an example of the "lab-to-production" challenge:

The Problem:

  • Long-Tail Distribution: Rare events are hard to capture in training data
  • Edge Cases: Extreme scenarios underrepresented
  • Dataset Bias: Training data doesn't reflect full operational domain

Example: Autonomous Driving

  • Handles common scenarios well (highway driving, normal traffic)
  • Struggles with rare events (unexpected obstacles, unusual weather)
  • Edge cases cause disproportionate accidents

Solutions:

  • Synthetic data generation for rare scenarios
  • Continuous learning from real-world operations
  • Human-in-the-loop for edge case handling
  • Balanced optimization between common and rare cases

Philosophical and Sociological Questions about AI ​

Eden addressed several fundamental questions about AI's nature and limitations:

Q1: Does AI Have Emotions?

Answer: Not in the human sense.

  • AI can simulate emotional responses
  • Lacks subjective experience (qualia)
  • No biological basis for feelings
  • Outputs are statistical patterns, not genuine emotions

Q2: Will AI Replace Human Jobs?

Answer: Transformation, not replacement.

  • AI augments human capabilities (25% productivity increase typical)
  • Creates new job categories
  • Humans remain essential for:
    • Creativity and innovation
    • Ethical judgment
    • Emotional intelligence
    • Complex decision-making
  • Historical precedent: Technology transforms work, doesn't eliminate it

Q3: Can AI Self-Evolve?

Answer: Not autonomously, but with human guidance.

  • Current AI lacks:
    • Self-awareness
    • Independent goals
    • Free will
  • AI evolution requires:
    • Human-defined objectives
    • Human-curated training data
    • Human-designed architectures
  • Self-improvement within human-defined parameters is possible

Q4: Why Does AI Hallucinate?

Answer: Statistical nature of language models.

Root Causes:

  • Probabilistic Generation: LLMs predict next tokens based on statistical patterns
  • No Ground Truth Verification: Models don't "know" facts, they predict likely continuations
  • Training Data Limitations: Gaps and inconsistencies in training data
  • Confidence Calibration: Models can't reliably assess their own certainty

Implications:

  • Hallucinations are a feature of the approach, not a bug
  • Human verification remains essential for critical applications
  • Techniques to reduce hallucinations:
    • Retrieval-Augmented Generation (RAG)
    • Uncertainty quantification
    • Human-in-the-loop validation
    • Multi-model verification

AI's Fundamental Limitations ​

What AI Lacks:

  1. Free Will: No autonomous decision-making capability
  2. Self-Directed Goals: Operates only within human-defined objectives
  3. True Understanding: Statistical pattern matching, not comprehension
  4. Consciousness: No subjective experience or self-awareness
  5. Common Sense: Struggles with basic physical and social reasoning

AI's Nature:

  • Sophisticated computational tool
  • Powerful pattern recognition system
  • Efficient information processor
  • Not: A sentient being or independent intelligence

8. Company AI Content Management & Automation ​

Automated Publishing Workflow:

Damien demonstrated the company's AI content management system:

Content Pipeline:

  1. Draft Creation: Notion used for initial content drafting
  2. Agent Processing: AI agents transform drafts into final format
  3. GitHub Action: Automated publishing to static website
  4. Multi-Format Support: Markdown, multimedia, presentations
  5. Search Integration: Full-text search across all content
  6. Translation: Automatic English translation for global accessibility

Benefits:

  • Content Preservation: All sessions archived and searchable
  • Easy Sharing: Static website accessible to all employees
  • Cross-Department Collaboration: Knowledge sharing across teams
  • Historical Reference: Complete learning resource library
  • Continuous Improvement: Feedback loop for content quality

📊 Presentations & Materials ​

Session Recording ​

🎥 Watch Full Session

Recording Highlights:

  • Q3 2025 global AI trends and market analysis
  • Agent Skills deep dive and implementation examples
  • MCP vs Agent Skills comparison and use cases
  • Kevin Kelly's Agent Economy insights and discussion
  • Eden Ye's AI theory evolution presentation
  • WiseTech AI initiatives and standardization efforts
  • Interactive quiz and team engagement

Main Presentations ​

📄 Q3-2025 AI Highlights Report (PDF)

Comprehensive analysis of Q3 2025 AI industry trends, market dynamics, and technological developments.

📄 Kevin Kelly: Agent Economy Interview (PDF)

Full transcript of Kevin Kelly's insights on the agent economy, future of companies, AI philosophy, and technological optimism.

Supplementary Images ​

  • 🖼️ Agent Skills Overview Diagram
  • 🖼️ AI Insights Dashboard
  • 🖼️ AI Theory Evolution Presentation

Agent Skills & Tools ​

  1. Claude Agent Skills Documentation - Official documentation
  2. Anthropic Skills GitHub - Open-source skill repositories
  3. TenTen.co Claude Skills Tutorial - Educational platform example
  4. Claude Skills 40-Minute Enterprise Customer Service - Implementation case study (Chinese)

AI Industry Reports & Analysis ​

  1. Q3-2025 Artificial Analysis State-of-AI Report - Comprehensive market analysis
  2. Kevin Kelly Agent Economy Interview - Future vision and insights

Agent Economy & Platforms ​

  1. Agnes AI - Singapore's successful agent platform
  2. Agnes AI Success Story (Chinese) - 200K daily active users case study

Payment Infrastructure for Agents ​

  1. Stripe Tempo - Stablecoin payment infrastructure
  2. Circle Stablecoins - Agent economy currency solutions

🎮 Quiz Activity ​

Quiz Details ​

Total Questions: 3
Duration: 15 minutes
Format: Teams chat rapid response
Participation: Active team engagement

Quiz Topics ​

The quiz covered the following areas:

  • ✅ Q3 2025 AI market trends
  • ✅ Agent Skills vs. MCP comparison
  • ✅ Kevin Kelly's agent economy concepts
  • ✅ AI theory fundamentals
  • ✅ Agent application case studies

🏆 Quiz Results & Winners ​

Congratulations to our Session 6 quiz champions!

🥇 First Place ​

Winner: Season Gu
Prize: Customized gift

🥈 Second Place ​

Winner: Elliott Li
Prize: Customized gift

🥉 Third Place ​

Winner: Walter Ma
Prize: Customized gift

📊 Full Results ​

RankNamePrize
🥇 1Season GuCustomized gift
🥈 2Elliott LiCustomized gift
🥉 3Walter MaCustomized gift

🔑 Key Insights ​

Technical Insights ​

  1. Agent Skills Evolution: The transition from static prompts to dynamic, executable skills represents a fundamental shift in how we build AI applications
  2. MCP vs. Agent Skills: These are complementary technologies - MCP for persistent integrations, Agent Skills for on-demand flexibility
  3. Standardization Benefits: Adopting standardized skill formats enables reusability, maintainability, and cross-team collaboration

Strategic Insights ​

  1. Specialization Over Generalization: Kevin Kelly's perspective on specialized agents aligns with market success stories like Agnes AI
  2. Infrastructure Investment: The AI bubble may hurt early investors but benefits society through infrastructure development
  3. Agent Economy Currency: Stablecoins and micropayment infrastructure are critical enablers for the agent economy

Philosophical Insights ​

  1. AI as "Alien Intelligence": Recognizing that AI thinks differently from humans helps us leverage its true strengths
  2. Multi-Dimensional Intelligence: Intelligence isn't a single ladder - different systems can be intelligent in incomparable ways
  3. Optimism as Strategy: Kevin Kelly's choice of optimism as a deliberate strategy for shaping the future

Practical Insights ​

  1. Human-AI Collaboration: The 98.5% OCR accuracy example shows AI augments rather than replaces human judgment
  2. Progressive Implementation: Start simple with Agent Skills, upgrade to MCP when complexity demands it
  3. Context Engineering: Quality context provision remains crucial for reliable agent behavior

💬 Session Highlights ​

Most Impactful Moments:

Kevin Kelly: "You're not late. This is day one. In 2049, people will look back at 2025 and say, 'You didn't even have AI then.'"

Eden Ye: "AI hallucination isn't a bug—it's the natural result of statistical probability distribution in language generation."

Damien Li: "Agent Skills transform our AI prompts from one-time use instructions into reusable, standardized capabilities that the entire organization can leverage."

Participant Engagement:

The session sparked lively discussions on:

  • Practical applications of Agent Skills in current projects
  • Strategic implications of the agent economy for WiseTech
  • Philosophical questions about AI consciousness and capability limits
  • Real-world deployment challenges and solutions

📚 Further Learning ​

Recommended Next Steps:

For Developers ​

  • Explore Claude's Agent Skills GitHub repository
  • Experiment with converting existing prompts to Agent Skills format
  • Study the Agnes AI case study for product insights
  • Review MCP vs. Agent Skills architecture patterns

For Product Managers ​

  • Analyze agent economy trends for product strategy
  • Consider Agent Skills integration opportunities
  • Study specialized agent success stories
  • Plan for stablecoin payment integration in agent scenarios

For Everyone ​

  • Read Kevin Kelly's full interview transcript
  • Understand the multi-dimensional nature of intelligence
  • Explore Q3 2025 AI industry report
  • Stay updated with AI Insights bi-weekly publications

Deep Dive Resources:

  • Anthropic's Agent Skills documentation and best practices
  • Research papers on emergent capabilities in large language models
  • Case studies of successful agent applications
  • Kevin Kelly's books: "The Inevitable" and "What Technology Wants"

🎯 What's Next? ​

Looking Forward to Session 7!

Potential upcoming topics:

  • Advanced Agent Orchestration: Multi-agent collaboration patterns
  • Production Agent Deployment: Monitoring, observability, and reliability
  • Agent Security: Safety considerations and risk mitigation
  • Economic Impact: Measuring ROI of agent implementations
  • WiseTech Agent Strategy: Internal roadmap and implementation plan

Stay Tuned:

  • AI Insights newsletter (bi-weekly)
  • Internal AI application updates
  • Agent Skills adoption progress
  • Community contributions and case studies

🙏 Acknowledgments ​

Special thanks to:

  • Damien Li for comprehensive Q3 AI trends analysis and Agent Skills deep dive
  • Eden Ye for illuminating AI theory evolution presentation
  • Aimee for facilitating the interactive quiz session
  • Kevin Kelly for sharing profound insights on the agent economy
  • All participants for engaging discussions and active participation

Want to Implement Agent Skills?

Start with the official Claude documentation and explore the GitHub examples repository. Begin by converting one of your existing AI prompts into a standardized Agent Skill!

Questions or Collaboration?

Reach out to the Crossroad AI team to discuss Agent Skills implementation, share your experiences, or propose topics for future sessions!

Session 6 | October 31, 2025 | Agent Skills & Global AI Trends Q3 2025 | Damien Li, Eden Ye

Internal WiseTech Global Resource

Copyright © 2025 WiseTech Global