Bringing Hybrid, Agent-based AI to Higher Education

Higher education institutions are actively exploring ways to expand access to student services while operating within practical constraints related to staffing, budgets, and infrastructure.

Historically, student services have been delivered through a mix of websites and portals, printed reference materials, and appointment‑based advising or help desks. As student expectations increasingly reflect preferences for more immediate, conversational and self-service interactions, colleges and universities are evaluating alternative delivery methods to extend access at scale.

As interest grows in agent‑based AI, institutions are increasingly exploring conversational interfaces as an entry point to digital services. These systems are designed to handle requests, surface information, and guide users through processes using natural language. Evaluating this approach at scale raises broader architectural considerations for IT teams, including how AI‑enabled services are delivered and where different classes of workloads should execute.

The evolution of the delivery of student services in higher education.

Considering Hybrid AI Architectures

A growing pattern in agent‑based AI delivery is the use of hybrid architectures that blend local execution on modern AI-enabled PCs powered by small language models (SLMs), with cloud‑based AI services built on large language models. The ibl.ai agentic platform is one example, combining agent‑based design with the flexibility to run selected AI tasks locally on Windows Copilot+ PCs—such as Microsoft Surface devices with NPUs—while relying on cloud‑hosted services when broader institutional context is required. This design enables institutions to dynamically determine where inference occurs, shaping how performance, data locality, and operational cost exposure are managed.

As discussed in Microsoft’s evaluation of small language models for retrieval‑augmented generation, SLMs can be appropriate for selected, well‑scoped tasks, particularly when model behavior and deployment context are tightly controlled. Examples include:

• Offline or local environments, including on‑device scenarios, where local inference may be required due to connectivity, policy, or architectural considerations.
• Latency‑sensitive interactions, where model placement and execution location are key design factors.
• Scenarios with constrained AI budgets or usage thresholds, where organizations choose to manage cloud‑based inference consumption by selectively running tasks locally.
• Resource‑constrained environments, where model size and deployment footprint are important considerations.
• Task‑specific workflows, where smaller models may be fine‑tuned to support narrowly defined use cases rather than relying on general‑purpose, out‑of‑the‑box models.

Importantly, decisions about where and how AI workloads execute are shaped by institutional design choices, governance requirements, and technical constraints—not by the AI models alone.

An Example of Hybrid Agent‑Based Design in Higher Education

The ibl.ai agentic platform provides one example of how hybrid agent‑based AI can be applied in higher education. In collaboration with the Surface engineering team, ibl.ai enabled support for on‑device inference and developed the student experience pack—a collection of preconfigured, task‑specific AI agents, designed to run on Surface Copilot+ PCs with NPU. These agents support common student‑facing scenarios, including Study Hub, Campus Connect, Career Launchpad, and Surface device‑related services such as onboarding, support, and care.

This collaboration exemplifies how services can be configured to operate either locally or in the cloud, based on how institutions dynamically determine where inference should occur. Under typical conditions, the agents may connect to cloud‑based services to access institutional systems, broader context, or cross‑service coordination. In other scenarios—such as offline use or institution‑defined usage thresholds—the same agents can be configured to run entirely on the device without requiring cloud connectivity.

Designing for Flexibility and Choice

As AI becomes more visible in student‑facing experiences, colleges and universities are balancing competing demands. Students increasingly expect conversational, always‑available support. Institutions want to meet those expectations responsibly, without sacrificing governance or sustainability. And IT teams need architectures that can flex across devices, networks, and deployment models.

Students want conversational access to services; institutions need a smarter way to deliver it—one that scales by design, preserves control, and uses hybrid AI execution to create value without unnecessary tradeoffs.