1

The Software 3.0 University

AI Agents, Human Intent, and the Future of the Academy

DUKE KUNSHAN UNIVERSITY

In China, For the World

Yisu Zhou

Interim Dean, Academic Strategy and Learning Innovation

China Director, Institute for Global Higher Education

Professor of Public Policy and Sociology

HEPA Forum 2026

2026-04-01

A Rapidly Shifting Ground

• On rapidly shifting ground for university administrators and academics.
• The declaration of a “new era” is common.

​

However…

This time is genuinely different due to AI.

2

The Story of Stanley

Stanley, a CTO of a mid-sized AI start-up and SJTU-trained engineer.

​

- How has AI changed your professional life?

- Completely.

​

Moving beyond the “Input-Output” model of simple prompts and responses.

3

From Writer to Architect

• The Agentic Workflow: �Managing multiple semi-autonomous “agents” simultaneously.
• The Agents: �Read task descriptions, search local codebases, scour the internet, and implement work.
• Stanley’s New Role: �From a writer of functions to an architect of context and an auditor of results.

4

Evolution of Programming: Software 3.0

5

The Battle for Jurisdiction

Jurisdiction: the right to control a domain of work.

​

• The Threat: �When core tasks are codified or routinized, technology threatens that jurisdiction.

​

• Destabilization: �AI is moving from simple execution to agentic problem-solving, challenging what it means to be an “expert”.

6

The Turf War

7

The Battle for Jurisdiction

Jurisdiction: the right to control a domain of work.

​

• The Threat: �When core tasks are codified or routinized, technology threatens that jurisdiction.

​

• Destabilization: �AI is moving from simple execution to agentic problem-solving, challenging what it means to be an “expert”.

8

Three dimensions

1. Epistemological Pivot�Where AI excels and where it hits a wall.
2. Crisis of Mastery�The destruction of traditional training grounds.
3. New Currency of Expertise�The high-value skills of the future human worker.

9

Dimension 1: The Epistemological Pivot

The Signal-to-Noise Ratio: �Different branches of inquiry face different challenges in the age of AI.

10

Hard Sciences

Built on formal chains of logic where “Signal” can be programmatically separated from "Noise".

Human Sciences

Operates in “noisy” knowledge spaces where quality indicators are subjective and interpretive.

Dimension 1: The Epistemological Pivot

The Hard Sciences: Accelerated by Verification

• The Frontier: AI solving 22 million algebraic problems in three months (Terence Tao) and generating novel quantum field theory frameworks (Steve Hsu).
• The Secret Weapon: Formal verification.
• Logic can be translated into machine-readable code (e.g., Lean) to instantly confirm or refute conjectures.

11

Tao’s talk: https://www.youtube.com/watch?v=mS9Lr43cIB4

Hsu’s paper: https://www.sciencedirect.com/science/article/pii/S0370269325008111

companion paper: https://drive.google.com/file/d/16sxJuwsHoi-fvTFbri9Bu8B9bqA6lr1H/view

online discussion: https://x.com/hsu_steve/status/1996034522308026435

12

Dimension 1: The Epistemological Pivot

The Hard Sciences: Accelerated by Verification

• The Frontier: AI solving 22 million algebraic problems in three months (Terence Tao) and generating novel quantum field theory frameworks (Steve Hsu).
• The Secret Weapon: Formal verification.
• Logic can be translated into machine-readable code (e.g., Lean) to instantly confirm or refute conjectures.

13

Dimension 1: The Epistemological Pivot

The Human Sciences: The Power of Idea

• Built on novel, compelling, and empirically supported accounts rather than formal proofs.
• Connections between ideas are theoretical and interpretive, not just logical.
• Reframing truth through new psychological concepts or cultural resonance rather than strict refutation.

14

Example: Why do employees leave a company?

https://www.jstor.org/stable/2780548

Image created by Nano Banana 2 from Google.

15

Why AI Progress is Slower in the Humanities

• No Automated Verifier: You cannot build a machine to judge the “resonance” of a sociological theory.
• Pattern Matching vs. Meaning: AI can mimic academic language but cannot organically generate novel sociological meaning.
• Verifiability vs. Novelty.

16

Verifiability vs. Novelty

AI models:

• perform massive cognitive compression of information.
• can synthesize existing academic consensus at lightning speed.
• struggle to organically generate novel sociological meaning.

​

• Truth in human sciences is interpretive, unlike formal refutation.
• The key challenge is Verifiability versus Novelty in knowledge.
• Expertise evaluation must pivot as AI affects departments unevenly.

17

Dimension 2: The Crisis of Mastery

AI creates structural friction:

• makes experienced professionals hyper-productive
• actively destroys the training ground required to build that very expertise in the next generation.

�Experts are good because they manually “did the reps”, but those reps are now being automated away.

18

“Squeeze it.”

The Claude Max 20x plan is a $200 monthly subscription designed for elite power users and developers.

19

A $200/month AI coding plan is much cheaper than an entry-level programmer:

​

• works 24*7
• can iterate & evolve across projects
• no long-term commitment

Dimension 2: The Crisis of Mastery

The “Ramen Chef Problem”

​

“If we don't train the chef how to make ramen from scratch, how will he ever know if the robot's broth tastes good?”.

​

• If mastery requires putting in the reps, but the economic incentive is to automate those reps, we face a pedagogical crisis.

20

Four Critical Questions for Higher Ed

1. Redefining Learning:�If “doing” is automated, what are we teaching?
2. The Reviewer Class: �How do we train “architects” instead of “executors”?
3. The Skill Gap: �Can a student evaluate a system without struggling with its manual components?
4. The Entry-Level Crisis: �What happens to internships when agents do the work faster?

21

The Pedagogical Crisis

• Higher education’s silos buffer us from rapid professional change.
• Students will not have this buffer against the velocity of change.
• The old educational paradigm, teaching students to be the workforce, is fading.
• We must prepare students to oversee an AI-driven digital workforce.
• Automating training “reps” creates a severe pedagogical crisis.

22

Dimension 3: The New Currency of Expertise

• Codified Knowledge:�Facts and formulas AI has already consumed.
• Embodied Knowledge: �Physical expertise that AI is rapidly approaching through robotics.
• Tacit Know-How: �The experiential, unspoken wiring of the human brain that AI cannot organically generate.

23

Dimension 3: The New Currency of Expertise

The Premium Human Skill

• Framing and Intuition: �The ability to frame messy problems and know if an AI output aligns with the "human condition".
• Architectural Vision: �Designing context and preparing the "ingredients" for AI agents.
• Evaluative Judgment: �Relentlessly verifying output against complex business or social needs.

24

Architect and Auditor

Image from Civilization 7

25

Dimension 3: The New Currency of Expertise

Architectural Vision and Evaluative Judgment.

• Human professionals must fiercely claim jurisdiction over the formulation of problems.
• The highest professional value will lie in orchestration: �leading a human-machine augmented approach to interpret, verify, and navigate social outcomes.

26

Orchestrating the Automated Orchestra

• Jurisdiction Shift: �We must claim jurisdiction over the formulation of problems, even if we lose the execution.
• The Human Value: �Leading a human-machine approach to navigate “noisy” social outcomes that computation alone cannot understand.
• The New Expert: �A curator of iterative prompts and a manager of agent fleets.

27

Conclusion: Redefining Human Expertise

AI has arrived as a foundational, general-purpose technology that will reshape the global landscape.

​

• The Task Ahead: �Build institutions that teach students to govern, interrogate, and make AI-generated knowledge meaningful.
• Courage over Fear: �Move forward not from fear of obsolescence, but with the courage to redefine expertise for Software 3.0.

28