Emily Bennett · a16z
The First AI-Native University
Emily Bennett describes an institution built from the ground up around intelligent systems.
Open source ↗Our Campus, Your Compass
Building the Agentic Representation Layer for higher education.
A Message from the Founder
FOUNDER VIDEOVIDEO PLACEHOLDERARCAMPASS
The Learner Model
Static vs. Agentic
Continue scrolling↓
Our Campus, Your Compass
Building the Agentic Representation Layer for higher education.
We create a living model of each learner that understands how their reasoning takes shape, how their knowledge updates and how their capabilities develop over time.
01 / 03
Agentic Representation
In 2027, we are expecting to see the world's first AI-Native University.
Signals from the field
Satya Nadella · Video
A New University, A New Pedagogy
OpenAI Forum · Video
Building an AI Powered University
Arizona State University · Video
The Agent Learner Initiative
The building block of an AI-native university lies in the living model of every student.
Each living model carries a learner's knowledge, reasoning, goals and evidence. Together, they create a high fidelity A2A learner network where students communicate with each other at agentic bandwidth.
One learner. One continuous model. Every experience connected.The Core IP
Introducing
Arc Pyramid
The three levels of the Agentic Representation Layer as learning moves from the individual to collective intelligence.
Level 3Institution RepresentationTHE INSTITUTION
Level 2Cohort RepresentationONE COHORT
Level 1Individual RepresentationONE LEARNER
Signals and Ideas
The category is taking shape.
Products, research and our own working notes show where agentic higher education is moving.
01
Across the ecosystem
Independent public category signals, included here for context.
HBS Foundry
AI-native venture building
Harvard Business School now describes Foundry as an AI-native workspace that combines faculty-modeled AI mentorship, live experts and cohort community to help founders reach launch and funding readiness.
Read source ↗
Khanmigo
Guided AI tutoring
Khan Academy's AI-powered tutor and teaching assistant guides learners with questions instead of simply giving answers and supports educators with planning tools.
Read source ↗
Coursera Coach
Learner support at scale
Coursera reports that its pedagogy-grounded AI learning assistant has supported more than one million learners with interactive guidance, practice and real-time feedback.
Read source ↗
ChatGPT Edu
Institution-managed AI
OpenAI reports that secure, institution-managed ChatGPT deployments now support hundreds of college and university campuses across teaching, research and operations.
Read source ↗
Gemini for Education
Teacher-led AI learning
Google is bringing teacher-led Guided Learning, study notebooks and NotebookLM into Classroom and supported learning management systems.
Read source ↗01 / 05
02
Research Reference
Memory, world models and multi-agent simulation—the research foundations behind the Agentic Representation Layer.
From individual models to society simulation
Simile · 2026
Simile connects high-fidelity models of real people into multi-agent simulations designed to play forward decisions and emergent outcomes.
Open paper ↗
Sleep-time agents and evolving memory
Letta · 2025
The team behind MemGPT extended agent-managed memory with sleep-time compute, allowing agents to reorganize memory and form connections between interactions.
Open paper ↗
Generative agents
Stanford · UIST 2023
A complete experience record, retrieval and higher-order reflection produced coherent individual and emergent social behavior.
Open paper ↗
Generative simulations of 1,000 people
Stanford · 2024
Interview-grounded agents modeled 1,052 real individuals and reproduced attitudes and behaviors across multiple evaluations.
Open paper ↗
Concordia: grounded social simulation
Google DeepMind · 2023
Memory-equipped agents act inside a shared physical, social or digital world governed by an explicit simulation environment.
Open paper ↗
World Models
Ha & Schmidhuber · 2018
Agents can learn a compressed spatial and temporal model of an environment, then plan and train inside the model itself.
Open paper ↗Genie 3: a real-time world model
Google DeepMind · 2025
A general-purpose world model generates interactive environments that people and agents can navigate in real time while maintaining consistency over several minutes.
Open paper ↗03
Arcampass Heptagraphs
Arcampass Heptagraphs, published as a series of visual essays.
01
Company thesisThe Learner Is the Missing Digital Layer
Heptagraph · Blog02
Research noteWhat Makes a Learner Model High Fidelity?
Heptagraph · Blog03
Product architectureFrom One Learner to the Institution
Heptagraph · BlogArcampass