Don't bring a knife to a gun fight. As LMS flops at exam delivery, agentic AI’s time is getting closer

Budgets stretched, assessment itself under attack, and now a high-profile cyberattack on swathes of universities delivering exams via their learning management system (LMS).

So you’d expect me to say, ‘They should get a dedicated digital assessment system’. Which is true today. But tomorrow may see a very different answer.

Mis-matched learning systems used for exams can be a muddle - never designed for the rigorous demands of authentic summative assessment. The LMS/ VLE is the beating heart of many institution’s digital teaching environment. But it’s been oversold. Shoehorned into handling high-stakes, summative exams. The result? An overreach. Stifled innovation. Flaky exam security. And a clunky candidate experience.

An LMS groans at thousands of concurrent users typing or hitting ‘submit’ simultaneously. Those who use typed answers within high-stakes on-line exams know this. An LMS making occasional calls to a shared server is totally unsuited for high stakes exams.

Evolution for digital assessment, or something else? The assessment sector continues to navigate turbulence. Distressed procurement, imperfect choices, the perceived threat of AI, managing and sweating systems. This is compounded when admins still use quasi-analogue workarounds and processes meant for paper exams. Still working with spreadsheets?

Turbulence in the assessment sector is compounded by poor system choices, analogue workarounds, and ersatz processes with paper origins. 

The use of agentic AI radically changes the conversation. What if you (as an exam administrator) had one screen, type a prompt, and see your system setting up and delivering an exam session? An agentic AI system’s ability to parse multi-step goals, source data, create a prompt, and run task executions for the whole exam sector are almost upon us. We’re already seeing it deployed across item production, grading, and remote proctoring.

A paradigm shift – when agentic AI does the heavy lifting it means that the system (composed of AI agents) can decide actions, make reasonable assumptions, and decide with minimal intervention or supervision. Exam owner staff currently have to wrangle complex systems, juggling with multiple integrations, and shoe-horn manual/ analogue processes into a digital 2.0 world. Think of agentic AI as a problem solver. It makes decisions, takes actions, and can adapt to changing circumstances.

Exam owners wrangle complex systems, juggle multiple integrations, and shoe-horn manual/ analogue processes.

Here’s an example. Set up an exam session for a specific certification, delivered to a specific cohort, at a particular time. An agentic AI system could take that objective, and deconstruct it into stages, performing those stages by itself across multiple tools and systems. Such as an item banking tool, certification management system, test delivery software, remote proctoring system, candidate ID verification portal, examiner tracking system, etc.

The agentic AI will verify its progress, and continues until the objective is met, or arrives at a problem. It harvests (internet) data, interrogates and creates files, executes code, sends notifications, updates exam window capacity, corrects its own work, and fixes mistakes. And the exam administrator periodically checks in, without needing to manually intervene at almost every point, as they do today.

Exam administrators currently have to manually intervene at almost every point of the process. Agentic AI makes the decisions, takes action, and checks you are ready.

Agentic AI makes assessment supervision easier? AI is challenging the assessment sector to ramp up assessment supervision. Remember: assessment only holds currency if it is supervised and trusted. The huge irony is that agentic AI could liberate exam owner resources away from system juggling, and towards coherent supervisory activities.

This would be a game-changing leap for exam owners and awarding bodies. Much of exam production is workflow and processes. Agentic AI is great at this. Reducing the cognitive load of switching across all those mission-critical static systems, and risk of human error. No chance of a missed spreadsheet upload, or clumsy copy/paste.

The exam administrator is liberated to do higher value work. Nurturing learner centres. Sourcing and keeping examiners. Designing out malpractice opportunities. Making better, authentic, supervised assessments.

Assessment desperately needs more supervision. Agentic AI liberates exam administrators to provide that.

Exam owners must demand these efficiencies from suppliers. What does this mean for software and services companies serving the exam sector? Procurement changes. Empowered exam owners. Learners benefit from fewer human-cause errors and greater accuracy. Assessment currency and confidence will improve. The AI threat is flipped.

With apologies to the late Clayton Christensen, it’s time we stopped asking the LMS to do a job it was never hired for. The days of stitching together multiple-managed systems, (or using a compromised LMS), to deliver high-stakes exams are numbered.