Every Professor Needs an Agent
I am a marketing professor at Washington State University. I have an AI agent named OCPI4 running on a Raspberry Pi 4 in my house. It writes essay drafts. It scans academic journals. It builds course materials. It manages my book’s publishing pipeline. It transcribes my voice memos. It maintains its own memory across sessions.
It costs me about $5 a day in API fees. The hardware cost $80.
This is not a flex. This is a statement about what is now possible for any faculty member at any institution, and why most of them should be doing it.
The Problem With Institutional AI
Most universities are approaching AI the way they approach everything: committees, policies, pilots, procurement cycles, and a 14-month timeline to deploy something that a graduate student could set up in an afternoon.
The typical institutional approach looks like this: the administration negotiates an enterprise license with one of the major AI providers. Faculty get access to a sanitized, locked-down version of a chatbot. Usage is monitored. Data flows through institutional servers. Policies restrict what can be uploaded. The tool arrives pre-neutered, six months after the market moved on.
This is not useless. But it misses the point entirely.
The value of AI for a faculty member is not a chatbot that answers questions. It is a persistent, personalized agent that knows your research, your courses, your students, your writing style, your deadlines, and your intellectual priorities. An agent that gets smarter about your work over time. An agent that is always on, always available, and always yours.
That cannot be an institutional tool. It has to be personal.
Why Local, Why Private
There are three reasons every faculty member needs their own local agent running on their own hardware.
1. Privacy is non-negotiable for research.
Faculty work with unpublished research, grant proposals, tenure files, student data, peer review assignments, and confidential correspondence. None of this should flow through a third-party cloud service. None of it should be stored on servers you do not control. None of it should be training someone else’s model.
A local agent running a private model processes everything on your hardware. Nothing leaves your machine. Your unpublished manuscript does not become training data. Your student grades do not sit on OpenAI’s servers. Your tenure committee notes do not pass through anyone’s API.
This is not paranoia. This is basic research hygiene. The same faculty who would never email a draft manuscript to a stranger are happily pasting it into ChatGPT. A local agent eliminates that contradiction.
2. Continuity changes everything.
A chatbot is a single conversation. You ask, it answers, the context disappears. Every session starts from zero.
A local agent with persistent memory is a collaborator. My agent knows I am writing a book about human value in the age of AI. It knows the ten archetypes. It knows my 4I Framework. It knows I do not use em dashes. It knows my course schedule, my grading rubrics, and the names of the papers in my research archive.
When I say “draft an essay connecting this week’s news to the book,” it does not ask me what book. It already knows. When I say “build a team activity for Week 8,” it knows Week 8 is distribution and it knows my class has 275 students in teams of five.
That accumulated context is the difference between a tool and an assistant. And it only works if the agent is persistent, local, and yours.
3. Faculty should understand what they are teaching about.
We are sending students into an economy that will be shaped by AI agents. We are telling them to develop judgment, creativity, and interpersonal skills that AI cannot replicate. We are writing learning objectives about “AI literacy” and “responsible AI use.”
And most faculty have never run an AI agent. Most have never configured a model. Most have never experienced what it means to have a persistent AI collaborator that knows their work.
How do you teach students to work alongside AI if you have never done it yourself?
Running your own agent is not just productivity. It is professional development. It is research. It is pedagogy. It is the lived experience that makes your teaching credible when you stand in front of 275 students and talk about the future of work.
What a Faculty Agent Actually Does
Here is what my agent does in a typical week:
Research support. It scans academic journals every Monday for papers on my research topics. It searches 250 million papers via OpenAlex. It pulls citation chains, identifies open access links, and summarizes findings. What used to be a half-day literature review is now a 30-second query.
Writing pipeline. It drafts essays based on my ideas, in my voice, following my style guide. It scans the news each morning for stories relevant to my work. It maintains a review dashboard where I can approve, edit, or reject drafts before publishing. It handles the Hugo static site, the GitHub deployment, and the Buttondown newsletter distribution.
Course materials. It builds assignments, discussion guides, and slide deck outlines. It knows my course structure, my grading rubrics, and my pedagogical approach. When I voice-memo an idea for a class activity while driving, it transcribes, expands, and saves it to the right course folder.
Administrative work. It tracks deadlines, monitors email for urgent items, and maintains a daily briefing. It organizes files, backs up its own memory, and keeps a version-controlled log of everything it has done.
Book development. It has read my entire book. It can explain any chapter, compare any archetype, and draft new content that is consistent with what already exists. Four of the new chapters in the latest edition were drafted collaboratively with this agent.
This is one faculty member on a Raspberry Pi. Imagine what an entire department could do.
The Hardware Is Trivial
The most common objection I hear is cost. This objection has not been valid since approximately 2024.
A Raspberry Pi 4 costs $80. An external drive for storage costs $50. OpenClaw, the agent framework, is free and open source. API costs for a cloud model run $3-10 per day depending on usage.
For faculty who want full privacy with no cloud dependency, a used workstation with a mid-range GPU can run open-source models locally. A refurbished machine with an NVIDIA 3060 costs under $500 and can run a 7 billion parameter model in real time. A 4090 runs a 32 billion parameter model, which is more than sufficient for any faculty use case.
The entire setup, hardware, software, and configuration, can be completed in a weekend. My agent was operational within hours of unboxing the Pi.
Compare this to the cost of an institutional AI license, which typically runs six to seven figures annually, requires IT involvement, takes months to deploy, and delivers a generic chatbot that knows nothing about your work.
The economics are not close.
What This Means for the University
I am not arguing that universities should stop investing in institutional AI infrastructure. Enterprise tools have their place: student-facing chatbots, administrative automation, data analytics at scale.
But faculty AI is different from institutional AI. Faculty work is personal, specialized, and confidential in ways that institutional tools cannot accommodate. The professor working on a grant proposal needs an agent that knows their research history, not a chatbot that starts from scratch. The professor developing a new course needs an agent that understands their pedagogical philosophy, not a tool that generates generic lesson plans.
The university’s role should be to enable faculty agents, not to replace them with institutional ones. Provide the training. Provide the hardware stipend. Provide the network infrastructure. Then get out of the way and let faculty build the tools that serve their specific work.
The Bigger Argument
There is a deeper reason faculty need their own agents, and it connects to everything I have been writing about.
The future of professional work is human judgment augmented by AI tools. The professionals who thrive will be the ones who can direct AI toward meaningful outcomes, not the ones who wait for their institution to hand them a pre-configured tool.
If we believe that, and if we are teaching that, then we need to live it.
Every faculty member should have their own agent. Running locally. On their own hardware. With a private model for sensitive work and a cloud model for everything else. Persistent memory. Personalized context. Always available.
Not because it is cool. Because it is the future of knowledge work, and we should understand it before we teach it.
The Raspberry Pi on my desk is not a gadget. It is a preview of what every professional’s workflow will look like within five years. Faculty should be the first to adopt it, not the last.
Andrew Perkins is the author of After the Grind and Chair of the Department of Marketing and International Business at Washington State University’s Carson College of Business.
Get essays like this in your inbox
New essays on AI, education, and the future of work. No spam, unsubscribe anytime.