2025.05 MCU. Managerial AI skill stacking keynote (1).pptx

73 % of workers report productivity gains from AI ... while writers, coders, and designers are already being replaced by AI-augmented staff.

This talk borrowed descriptions from the business and tech world that define AI-augmented staff as "agent-bosses"— knowledge workers who stack cognitive, technical, and managerial skills to coordinate teams of AI agents, not just use working with a single chatbot tool.

1. Managerial AI skill stacking— A new skillset for worker readiness for the AI age Nigel P. Daly, ITI@Taipei , TAITRA May 10, 2025 MCU Conference: Language in the Post-AI world Some trends and speculations from the business world

2. Abstract Managerial AI skill stacking—A new cognitive skillset for worker readiness in AI age As language educators helping students prepare for a new era in the job market in EFL countries like Taiwan, we should know that AI-driven transformations are reshaping knowledge work and demanding a new form of workflow —managerial AI skill stacking. This involves combining cognitive, technical, and managerial competencies to effectively leverage AI systems, especially with emerging genAI technologies that consist of several legacy Large Language Models (LLMs) of different AI personalities (ChatGPT, Claude, Grok, Gemini), newly emerging reasoning LLMs (GPT-o1 and -3, Gemini 2.0 Flash, and DeepSeek R1), and AI agents (OpenAI’s Operator and ChatGPT Operator, MS Copilot, Google DeepMind’s Project Mariner) that will come to prominence in the next year or two. In short, this marks the emergence of a new cognitive and knowledge worker paradigm. Skilled knowledge workers will need to use more than just ChatGPT—they will draw upon different AI tools for different tasks and will need to communicate and coordinate these AI tools as a manager would a team. Recent research indicates that widespread AI adoption is increasing—not reducing—the need for human judgment, leadership, and strategic decision-making. To thrive, knowledge workers must cultivate 21st-century skills—critical thinking, communication, collaboration, creativity, and digital literacy—alongside AI fluency. This paper integrates studies on skill stacking and emerging evidence of how AI shifts knowledge workers into managerial roles. It also explores the “cognitive bleed” between human and AI reasoning and suggests paths and practices for AI fluency training to help our students prepare for and compete in an AI age of knowledge work.

3. Plan I. The impending c risis? II. GenAI 2025: From Tools to Agents II.5 Will humans be relevant? III. “Frontier Firms” and “Agent Bosses” – knowledge w orker 2.0 IV. What are agent boss skills? Managerial AI Skill Stacking V. Implications for education Trends from the business world: AI advances, “frontier firms”, new “ agent-boss” workers

4. I. An impending crisis ?

5. AI + people = productivity The 2025 AI paradox: - Understanding consumer perceptions in APAC - F5, https://www.f5.com/content/dam/f5/corp/global/pdf/report/the-2025-ai-paradox-understanding-consumer-perceptions-in-apac.pdf PwC's 2025 Global CEO Survey https://www.pwc.ro/en/press-room/press-releases-2025/pwc-s-2025-global-ceo-survey-reveals-that--while-most-executives.html Generation AI in Asia Pacific | Deloitte Insights, accessed May 6, 2025, https://www2.deloitte.com/us/en/insights/topics/emerging-technologies/generative-ai-adoption-asia-pacific-region.html https://news.microsoft.com/zh-tw/work-trend-index-2025/ 73% of individuals  personal productivity gains [1] 56% of CEOs saw more employee time efficiency [2] Daily GenAI users saved nearly a full workday per week (6.3 hours) [3] employees in "AI Forward" companies: handle increased workloads (58% vs. 21%) [4] Ok, from interested parties … but still ….

6. “The AI jobs crisis is here, now.” Brian Merchant, May 2, 2025: the AI jobs crisis isn’t coming—it’s already here. Eg1. Duolingo going “AI-first”: replaced writers and translators with genAI -- not better quality, but cheaper and easier to control. 150 new courses in 1 year (vs 12 yrs before) Eg2. Mark Zuckerberg: AI will write most of Meta's code in just 12-18 months (AI will outdo top engineers) Eg3. Artists, illustrators, and journalists displaced by AI (Voice actors on strike for 9 months) Eg4. USA Department of Government Efficiency (DOGE), led by Elon Musk -- mass layoffs across federal agencies, displacing tens of thousands of workers Merchant: The deeper issue isn’t the tech—it’s power. AI is a management strategy to cut labor, centralize control, and reshape work. Creative and entry-level white-collar jobs are quietly disappearing.

7. The Grad-Gap in Employment Grad unemployment vs general population (from Derek Thompson’s April 30, 2025 Something Alarming Is Happening to the Job Market - The Atlantic ). W hen firms hire 1 employee with ChatGPT instead of 3 junior staff

8. and Taiwan …? The unemployment early 2025: 3.35% - lowest levels for the same period in the past 25 years BUT High Youth Unemployment : ages 20–24 is 11.33% ( national avg = 3.35% ) Why? S kills mismatch —grads often lack the practical skills/experience needed in high-demand industries, especially tech [ Levine, B. (2024, January 10). Beyond degrees: Taiwan’s youth unemployment crisis . Global Taiwan Institute. https://globaltaiwan.org/2024/01/beyond-degrees-taiwans-youth-unemployment-crisis/ ]

9. II. GenAI 2025: From Tools to Agents

10. The rapid evolution of GenAI Tech advances everyday!!! AI-news fatigue … hard to keep up! Timeline GPT-3.5 → GPT-4 → DeepSeek R1 → GPT-o3 → Gemini 2.5 / Claude 3.7 2025 is the year of reasoning and agentic AI.

11. Reasoning LLMs Arrive Large Language Model designed for complex, multi-step problem solving More accurate and reliable answers Key capabilities Performs step-by-step logical reasoning ("chain of thought") Handles multi-layered analysis and decision-making Understands and adapts to context for accurate responses Training methods Combines large-scale neural networks with fine-tuning and reinforcement learning Applications Mathematical problems, coding, puzzles, complex question answering DeepSeek R1 OpenAI GPT-o3 Gemini 2.5, Claude 3.7 Grok 3 (“Think mode”)

12. Reasoning … sort of … sometimes? Can still hallucinate … Anthropic study … Reasoning models don't always say what they think \ Anthropic

13. Enter the Agents Definition autonomous software system that perceives its environment, makes decisions, and takes actions to achieve specific goals-often with the ability to learn and adapt over time For example, Deep Research Autonomous AI agent performing complex, multi-step research independently VS 2. Tasks (ChatGPT) Automation feature enabling scheduled or repeated actions, not fully autonomous or multi-step  not a standalone agent

14. Enter the Agents Commercial examples 1. Customer Support Agents ( eg , Trengo , Zendesk, Intercom) H andle customer inquiries, process refunds, update account info, and even solve technical support issues automatically 2. E-commerce Agents ( eg , Amazon, Shopify platforms) Automate order processing, track shipments, send reminders about abandoned carts, offer personalized shopping suggestions Use image recognition for visual product searches, making online shopping smoother and more engaging Company Key Agents / Platforms Capabilities Interoperability Microsoft Researcher, Analyst, Copilot Research, data analysis, workflow automation Supports Google’s A2A protocol OpenAI Operator, Deep Research, Agents SDK Autonomous web interaction, multi-step tasks Integrates with developer tools and APIs Google Agentspace , Deep Research, NotebookLM Plus Enterprise automation, no-code agent creation Developed and promotes A2A protocol

15. Tech acceleration, slow uptake ... (?) AI is like a “normal technology” [not a superintelligence ] , like electricity [ Narayanan & Kapoor, 2025 ] SO An infrastructural general purpose tool and will take a number of years for the tool to find its way into corporate, education, and personal life … (AI applications mostly used in data-intensive areas like marketing, customer service, and supply chain/operations … so still "department-limited,“ [ Stanford 2025 Human-centered AI report]) BUT Mollick : because of the advances in agentic ai ( ie ai that can do stuff for you ) -- “ might actually accelerate diffusion dramatically compared to previous technologies. ”

16. Taiwan companies and CEOs 88% Taiwanese CEOs: 2025 is critical for re-evaluating core strategies/operations (vs global 82%). MS work trend index 2025 Agentic AI systems capable of autonomous task execution = major trend for 2025 ( IBM 2025 report 2025 Work Trend Index 82% companies want to add agents as "digital team members" within the next 12-18 months 58% already in process of fully automating departmental or team work/business processes using agents (significantly higher than the global average of 46%) Taiwanese CEOs overly optimistic? PwC 2025 report 8-10% of Taiwanese enterprises had clear AI applications, with 90% still exploring entry points. 31% reported already seeing AI-driven revenue growth, and 60% expected GenAI to enhance profitability in the coming year. [low adoption vs high optimism= optimism gap]

17. AI-savvy student workflow – for a report Agentic AI Reasoning LLMs Reasoning LLMs Assuming … … healthy PAIRR Human-first human-AI interaction: P – PLAN A – ASK I – INVESTIGATE R – REVISE R – REITERATE

18. II.5 Will humans be relevant?

19. Not really … According to Bill Gates …

20. AI will continue to have 3 limits … Daniel Susskind (2025): humans are still needed with AI limits: 1. General Equilibrium Limits: humans more efficient for some tasks, eg human dispatchers managing unexpected disruptions in logistics (AI good for routine route optimization) 2. Preference Limits: people simply prefer human interaction, eg clients who may still desire a trusted human advisor in finance (despite AI's capabilities) 3. Moral Limits: some tasks need human moral judgment, eg high-stakes negotiations require human judgment and people to be responsible for consequences ; decisions impacting people, such as life-or-death medical judgments

21. III. “Frontier Firm” and the “Agent Boss”, or knowledge w orker 2.0

22. Predictions for 2028-2030 – AI disruption 47% of companies expect AI to change 30% of work in 2025   AI in the workplace: A report for 2025 | McKinsey By 2030 ( WEF Future of Jobs Report 2025 ) 86% of businesses: AI tech to transform their operations 39% of current skills: expected to become outdated 85% of employers: up-/re-skilling is top priority 170 million new jobs by 2030 -- for those who adapt AI in the workplace: A report for 2025 | McKinsey

23. “Frontier Firm” MS report: 2025: The year the frontier firm is born New type of organization, “AI-first”, “AI-enhanced” workplace early adopters of AI at scale powered by "intelligence on tap," "human-agent teams," new role for everyone: “ agent boss ” structured around on-demand intelligence Expectations scale rapidly, operate with agility, and generate value faster Generate higher productivity, innovation, and employee optimism Employees at Frontier Firms: report higher rates of company thriving (71% vs. 37% globally) feel more able to take on more work (55% vs. 20% globally) more optimistic about the future of work (93% vs. 77% globally) less likely to fear AI taking their jobs (21% vs. 38% globally) OK, it’s Microsoft …

24. Frontier firms Why is this r ole emerging? Frontier firms (top 10% in productivity) -- twice as likely to adopt advanced digital technologies like genAI and multi-agent systems 68% of frontier firms: strong increase in task complexity when AI tools are introduced—not a decrease BUT AI reduces routine execution time, but raises the cognitive and managerial burden of planning, oversight, ethical review, and alignment Microsoft and LinkedIn (2024): 75% of knowledge workers use AI, only 31% have adapted their workflows to use it effectively—most are “experimenting ”

25. Frontier firm memo – what is AI-first?

26. Evolution of change Frontier firm’s 3 phase AI Integration: Phase 1 : AI as an assistant ( now or soon ), Phase 2 : AI agents joining teams as digital colleagues, Phase 3 : "Human-led, agent-operated" model Phase 3, humans set direction for agents running entire workflows, human as “agent boss”

27. What Is an Agent Boss? Definitions Agent Boss = knowledge worker who manages AI agents Prompt AI, define goals, monitor processes, handle breakdowns, evaluate results, and iteratively improve workflows . 2. Agent = AI-powered system that can reason, plan, and act autonomously to complete tasks or entire workflows, with human oversight at key moments In Frontier Firms, agent boss is a new role for “everyone ” [ MS report: 2025: The year the frontier firm is born ] Limits + specialization Low Human:Agent ratio (maybe up to 1:5)

28. IV. So, what skills do “agent bosses” need? - Managerial AI Skill Stacking

29. The AI M anagerial Skill Stack 1. Base – Cognitive Skills Critical thinking, judgment, evaluation [WEF’s Future of Jobs Report 2025: Critical thinking and complex problem-solving as top skills by 2027] Why It Matters: These are the human fundamentals—knowing when to trust, revise, or reject AI outputs. Eg Spotting bias, asking “Is this output valid?”, and evaluating relevance or tone.

30. The AI Managerial Skill Stack 1. Base – Cognitive Skills Critical thinking, judgment, evaluation Why It Matters: These are the human fundamentals—knowing when to trust, revise, or reject AI outputs. Eg Spotting bias, asking “Is this output valid?”, and evaluating relevance or tone. 2. Middle – Technical Skills Prompting, tool selection, output tuning Why It Matters: Enables control of AI behavior and fluent tool use across platforms. Eg Knowing which model to use (Claude, GPT-4, Gemini), structuring prompts with role, task, format, and tone.

31. The AI Managerial Skill Stack 1. Base – Cognitive Skills Critical thinking, judgment, evaluation Why It Matters: These are the human fundamentals—knowing when to trust, revise, or reject AI outputs. Eg Spotting bias, asking “Is this output valid?”, and evaluating relevance or tone. 2. Middle – Technical Skills Prompting, tool selection, output tuning Why It Matters: Enables control of AI behavior and fluent tool use across platforms. Eg Knowing which model to use (Claude, GPT-4, Gemini), structuring prompts with role, task, format, and tone. 3. Top – Managerial Skills Delegation, coordination, strategic oversight, comminication Why It Matters: Transforms AI from tool to teammate —human becomes the orchestrator of hybrid workflows. Eg Mapping a task across agents, supervising and synthesizing outputs, aligning with org. goals.

32. The AI Managerial Skill Stack 1. Base – Cognitive Skills Critical thinking, judgment, evaluation Why It Matters: These are the human fundamentals—knowing when to trust, revise, or reject AI outputs. Eg Spotting bias, asking “Is this output valid?”, and evaluating relevance or tone. 2. Middle – Technical Skills Prompting, tool selection, output tuning Why It Matters: Enables control of AI behavior and fluent tool use across platforms. Eg Knowing which model to use (Claude, GPT-4, Gemini), structuring prompts with role, task, format, and tone. 3. Top – Managerial Skills Delegation, coordination, strategic oversight, communication Why It Matters: Transforms AI from tool to teammate —human becomes the orchestrator of hybrid workflows. Eg Mapping a task across agents, supervising and synthesizing outputs, aligning with org goals. This pyramid reflects how a 21st-century knowledge worker will evolve … not just using AI tools, BUT managing them with the same skills used to lead human teams.

33. Human Managerial skills 1. Task Analysis & Delegation Break down tasks and give to right people Communicate expectations, formats, and deadlines 2. Strategic Goal Setting Align tasks with broader team or organizational goals 3. Coaching & Development Give feedback to improve individual and team performance. Support learning through iteration and encouragement 4. Quality Control Review and verify work for accuracy, relevance, and standards Catch errors before they become problems 5.Stakeholder Communication Translate work into clear insights for internal and external audiences Ensure messaging is aligned, timely, and persuasive

34. AI Managerial skills

35. The “agent boss” skill stack 1. Task analysis and delegation (Task-to-agent mapping) Good managers assign the right task to the right teammate , with clear expectations. When your team is human + AI, which tasks should be done by people? which by machines? which machines? which require a collaborative loop? EXAMPLE (Student AI boss) Agent? LLM? or Reasoning LLM?

36. The “agent boss” skill stack 2. Strategic Goal Setting (Prompt Framing, ie “intentional prompting”) Good managers connect everyday tasks to larger goals to make the team’s work contribute to real outcomes With AI, this means framing prompts that are purposeful, outcome-driven, and aligned with strategic intent i e intentional about context and consequence EXAMPLE (Student AI boss)

37. The “agent boss” skill stack 3. Coaching & Development (Feedback and Iteration) Effective managers don’t settle for first drafts—they guide teams by feedback loops that refine and improve performance Working with AI is no different: it requires reviewing outputs critically, identifying what works, and prompting clear revisions EXAMPLE (Student AI boss)

38. The “agent boss” skill stack 4. Quality Control & Risk Management (Editorial Oversight) Managers safeguard quality by spotting errors, off-brand messaging, or risky decisions before they reach the client or audience. As an agent boss, you’re the final filter—responsible for identifying hallucinations, bias, or irrelevant content in AI outputs, and cleaning them up before use EXAMPLE (Student AI boss)

39. The “agent boss” skill stack 5. Stakeholder communication (Human–AI Output Framing) Great managers translate technical or messy work into insights for the right audience (internal team, external execs). With AI, your role is to shape its outputs into persuasive, accessible, and stakeholder-ready narratives whether for your (hybrid) human team or executive decision-makers EXAMPLE (Student AI boss)

40. V. Implications for education ….

41. AI & education Students need to prepare somewhere … DEC Global Student Survey (2024 digitaleducationcouncil.com ) 50% of students believe that over-reliance on AI would negatively impact their academic performance 48% of students feel unprepared for an AI-enabled workplace Should there even be implications for educators …?

42. Challenges for youth + education in next 3-5+ years 1. Demand for New Skills Employers seek candidates with skills in AI, machine learning, data analysis, and digital tools soft skills : creative thinking, problem-solving, adaptability, leadership, and EQ are becoming more valuable as routine tasks are automated Continuous learning : 39% of current skills may be outdated by 2030 (WEF), making upskilling and reskilling a top priority. Need for adaptability and resilience 2 . Hybrid Human-AI Work Models Need to work alongside AI “agents” or automation tools, so need to be comfortable managing and using these technologies

43. Challenges for youth + education in next 3-5+ years 3. Changing Organizational Structures Traditional hierarchies are giving way to project-based, outcome-driven teams -- young workers may need to take on more leadership roles – over people and AI tools Digital collaboration and remote work  increases opportunities but also needs strong self-management and communication skills 4. Increased Competition and Opportunity Remote work and digital tools  increased competition, and more opportunities “Frontier Firms” and innovative employers are quickly adopting AI and looking for adaptable , tech-savvy young talent

44. Are we just training students to become cogs in the corporate machine? Well, yes. Otherwise, what is the alternative? Thank you

45. Agent boss example The Agent's Job end-to-end logistics in a supply chain: agent places orders, monitors inventory, tracks shipments, and flags issues The Agent Boss's Oversight Human supply chain manager = "agent boss“ – not manually performing these tasks, G uiding the agent system , resolving exceptions that the agent flags, and managing key relationships , such as with suppliers E nsuring AI agent meets objectives and steps in when human judgment or negotiation is required Managing Workflow The emerging role of the agent boss involves a human manager overseeing one or more AI-powered systems that can reason, plan, and act to complete tasks or workflows autonomously O versight is dynamic process that provides continuous guidance, resolving exceptions, intervening at key moments, and periodically evaluating output, rather than simply checking upon task completion

46. Advice for youth Invest in digital and AI skills : Take online courses, certifications, or university modules in AI, data science, and digital tools. Develop soft skills : Practice communication, teamwork, leadership, and creative problem-solving. Embrace lifelong learning : Stay updated with industry trends and be proactive about upskilling. Gain practical experience : Seek internships, projects, or part-time roles that offer exposure to AI and digital transformation. Build a global network : Participate in online communities, hackathons, and professional groups to expand your reach. Success = Adaptable + digitally fluent + proactive about learning + strong human + thinking skills