The Agentic Student

I used to think the AI conversation in higher education was about cheating.

Professors worried about essays written by ChatGPT. Administrators debated detection software. Think pieces asked whether homework was dead. The entire discourse centered on a single question: How do we catch students using AI?

That question is already obsolete.

The Tool Stack Has Evolved

Here’s what I’ve observed working with university students and recent graduates over the past eighteen months: The conversation has moved far beyond “Did you use ChatGPT for this paper?”

The students I’m seeing now don’t just use AI. They orchestrate it.

A typical “power user” student in 2025 maintains a stack that looks something like this:

Research: Perplexity for citation-backed synthesis, Consensus for academic papers
Study: NotebookLM to convert dense PDFs into podcast-style audio reviews
Building: Claude Artifacts to generate interactive simulations and dashboards
Productivity: Notion AI for querying their own notes, Motion for AI-driven scheduling
Automation: Zapier workflows connecting their LMS, email, and task managers
Local/Private: Ollama running Llama models on their gaming laptops

This isn’t a chatbot. This is infrastructure.

The question isn’t whether students are using AI. It’s whether they’re using it to consume, integrate, or build.

Three Phases of Maturity

I’ve started thinking about student AI adoption in three distinct phases. Not everyone reaches all three, but the trajectory is clear:

Phase I: The Augmented Consumer

This is where most students start. They use AI for consumption—getting answers, summarizing readings, generating first drafts. The key insight here is that “search” behavior is dying. Students don’t want ten blue links. They want a synthesized answer with citations.

The viral example is NotebookLM’s “Audio Overview” feature. Students upload lecture slides and readings, then listen to AI-generated podcast discussions during their commute. They’re converting dead time into study time by shifting modalities.

Phase II: The Workflow Integrator

Here, AI becomes a logistics manager rather than just a content generator. Students at this phase use AI scheduling tools to “Tetris-block” their tasks into available time. They build Notion databases that they can query (“What are the key themes from my Econ notes?”). They automate the tedious—scraping LMS notifications, auto-filing emails, triggering follow-ups.

The pattern I see repeatedly: students treating their daily operations as a system design problem.

Phase III: The Tool Builder

This is where it gets interesting. Students at this phase leverage AI’s code-generation capabilities to build bespoke tools for specific, transient needs.

A physics student asks Claude to create an interactive simulation of supply and demand curves with adjustable parameters. A business student uploads a CSV and gets a working dashboard—no Python required. A CS student generates the boilerplate UI for a hackathon project in minutes instead of hours.

The critical shift: software engineering is no longer a specialized career. It’s a just-in-time capability unlocked by natural language.

The Shadow IT Dynamic

What strikes me most about this adoption isn’t the sophistication—it’s the independence.

While 92% of students report using AI tools, only 29% feel supported by their institutions. 40% say their schools actively discourage it.

This disconnect has created what enterprise IT would recognize as a “Shadow IT” culture. Students aren’t waiting for professors to integrate AI into the curriculum. They’re building their own parallel infrastructure—often in secret—to manage the demands of their coursework.

The irony is that this lack of guidance has paradoxically fueled innovation. Forced to navigate ethical gray zones themselves, students have become remarkably thoughtful about when and how to use these tools—more thoughtful, perhaps, than if they’d been given rigid rules.

The Anxiety Engine

Why the urgency? Why are students building automation systems instead of just Googling?

The answer is economic anxiety.

63% of Gen Z workers worry that AI may eliminate jobs. 61% believe AI skills are essential for career advancement. Students aren’t adopting these tools because they’re lazy—they’re adopting them because they’re terrified of obsolescence.

This manifests as an aggressive “arms race” of upskilling. Students believe that proficiency in agentic workflows—managing AI systems rather than just doing the work—will be the defining competency of their careers.

Whether they’re right is an open question. But the belief is driving behavior.

The Subscription Divide

Here’s the uncomfortable truth about this transformation: access is unequal.

There’s a widening gap between “power users”—often STEM-focused, financially capable of affording premium subscriptions—and those relying on free, rate-limited, hallucination-prone models.

While some students build custom GPTs with GPT-4o, others struggle with the free tiers. This creates a two-tiered system of academic assistance that institutions haven’t begun to address.

The hardware dimension matters too. In technical circles, the gaming laptop with a high-VRAM GPU has become a status symbol—not for gaming, but for inference capability. The ability to run a 70B parameter model locally is the new flex.

The Economic Spillover

The most surprising development isn’t academic—it’s entrepreneurial.

Students are monetizing their AI fluency. I’m seeing:

Automation agencies: Students approaching local businesses to automate appointment booking or customer follow-up, charging setup fees and monthly retainers
Digital products: AI-generated printables, wall art, and niche study guides sold on Etsy
Self-publishing: AI-assisted writing and illustration for Amazon KDP

The combination of Midjourney, ChatGPT, and no-code platforms has created a “Builder” generation that treats side hustles as software projects.

What This Means for Everyone Else

If you’re in higher education—whether as an administrator, faculty member, or parent—the challenge has shifted.

The question is no longer: How do we detect AI use?

The question is: How do we teach the skills that remain valuable when AI handles the rest?

Those skills are:

Curation: Knowing which tool fits which problem
System architecture: Designing workflows rather than completing tasks
Ethical oversight: Recognizing when AI output needs human verification
Deep verification: Going beyond the AI’s synthesis to primary sources

The General Purpose User

What we’re witnessing is the emergence of what I’d call the “General Purpose User”: a student comfortable switching between models (Claude for code, Perplexity for research, NotebookLM for listening) and connecting them via automation layers.

They’re not programmers in the traditional sense. But they’re not passive consumers either. They occupy a new space—one where natural language is the interface to capability.

The students are building their own future. The question is whether institutions will help them lay the foundation or merely watch from the sidelines.

This post is a companion to the full research whitepaper and executive presentation on this topic.