The Great Online Learning Reset?
How agentic AI is forcing us to reimagine online learning -- for the better
If you work in L&D, you've likely seen the viral videos this week: Manus and other agentic AI tools completing asynchronous training courses in minutes. These AI agents don’t just assist learners, they autonomously complete courses by responding to questions, submitting assignments and even contributing meaningfully to online discussions.
In the wake of the rise of increasingly powerful AI agents, many educators are now scrambling to update security measures, block access to AI Agents and create new consequences for this new breed of “violation”.
In this week’s post, I’d like to put forward a spiky point of view: the agentic AI problem isn't an integrity or security problem—it's a learning effectiveness problem. If an AI can "learn" your material without actually learning anything, what exactly are your human learners getting from the experience?
The TLDR of my argument is that AI agents haven’t broken online learning — they have inadvertently exposed what most of us have known for decades: the traditional "content + quiz" model of asynchronous learning is fundamentally broken.
Let’s dive in!
The Structural Problems AI Has Exposed
Tonnes of research exists to show that traditional asynchronous learning platforms fail for three key reasons:
Cognitive Decoupling: Online async training separates theoretical knowledge from practical application, creating what learning scientists call a "transfer gap." Research shows skills developed in isolation require up to 57 reinforcement cycles versus just 1-2 when learned through applied practice.
Assessment Through Proxy Metrics: Quizzes measure recall, not competency. A recent Salesforce study found employees scoring 90%+ on compliance tests demonstrated only 34% protocol adherence in actual workplace situations.
Temporal Discontinuity: The forgetting curve is brutal—without application, 60% of knowledge vanishes within 48 hours without application. Pre-emptive training without immediate practice is largely wasted effort.
Agentic AI tools aren't cheating the system: they're revealing something that we have known but ignored for decades - that our system was never truly measuring learning in the first place.
So, what does this mean for the future of online learning?
Reimagining Async Learning in the Agentic Era
What if rather than viewing agentic AI as a threat, we embrace it as a catalyst for creating more effective learning experiences? What could this look like in practice?
Cognitive Recoupling
The structural problem of cognitive decoupling—i.e. the separation of theoretical knowledge from practical application—has plagued online learning since its inception. Agentic AI forces a reckoning with this decades-old challenge.
One potential solution lies in embedded experiential learning architectures that merge instruction with real-world application. Emerging learning platforms will likely adopt more context-aware AI “coaches” that guide learners through complex problem-solving scenarios.
For instance, systems similar to those used by Khanmigo and Duolingo which deliver adaptive content and tutoring will likely evolve into dynamic workplace support & training systems, where AI agents generate dynamic practice scenarios and performance support in the flow of work while adjusting difficulty based on learner performance.
These systems will leverage real-time performance analysis, using computer vision and natural language processing to provide instant feedback on practical tasks. Imagine a cybersecurity training module where an AI mentor observes a learner’s navigation through a simulated network breach, offering targeted guidance when missteps occur—a concept previewed in Articulate’s real-time AI integration demos.
From Proxy Metrics to Competency Measures
In the agentic AI era, quizzes and completion metrics will give way to AI-powered competency mapping. Rather than assessing immediate recall, we will like see a shift towards continuous skill verification through:
Behavioural analytics: Tracking decision-making patterns in both simulated and real environments.
Work product analysis: AI evaluation of authentic artefacts like code repositories or design portfolios.
Collaborative validation: Peer review systems augmented by AI quality control.
This future is already a reality for some. Microsoft’s AutoGen v0.4 framework, for example, shows how asynchronous agent networks could assess competency through distributed evaluation processes powered by AI.
In practice, this might involve the multiple AI agents:
A simulation agent to generate realistic work scenarios for supported practice and feedback
A support agent to help employees learn in the flow of real world tasks
An assessment agent to track performance against previous performance and industry standards
This multi-agent approach eliminates the so-called "proxy metric" problem by evaluating actual job readiness rather than content recall.
The Death of the LMS
The forgetting curve’s devastating impact on asynchronous learning demands AI-driven reinforcement systems. Future platforms will employ:
Learning in the flow of doing: Agentic AI delivers learning-as-support in the flow of work
Just-in-time learning: Agentic AI delivers bite-sized refreshers triggered by calendar events or other workflow contexts
Predictive intervention: Algorithms anticipating knowledge gaps and decline based on industry standards and individual performance patterns
Social reinforcement: AI-facilitated discussions & support sprints timed to reinforce key concepts
For example, an AI agent might detect a learner struggling with statistical process control and schedule:
A 90-second video refresher before their next production meeting
A simulated machine calibration exercise
A discussion & support session
This approach transforms learning from isolated “learning events” into continuous improvement cycles in the flow and completion of work.
“Stop and learn” learning platforms as we know them will likely disappear into the back end and be replaced by agents which help us not just to be more productive (delegation) also learn new skills (develop) in the flow of work.
This may seem like a far stretch from the status quo of video + quiz LMS, but change here is already well underway. Agentforce for Slack, for example, offers a suite of preconfigured autonomous AI agents that enable employees to increase productivity and enable learning in the flow wok via of interaction with AI agents.
The Blurring of Learning Modalities: Online, In-Person & Beyond
For years, in-person and live-online training sessions have held a crucial role in learning because they offer what traditional asynchronous courses lack—immediate feedback, real-time coaching, and hands-on practice. The assumption has been that some aspects of training simply cannot be replicated in an online format without a human facilitator. However, the emergence of agentic AI is beginning to challenge this very premise.
As AI-powered learning ecosystems evolve, the distinctions between online, in-person, and blended learning models will become increasingly fluid. If AI can provide real-time guidance, adaptive feedback, and dynamic practice opportunities, then the rigid separation between e-learning and instructor-led training starts to dissolve.
In an agentic AI future, what might in-person training look like? Rather than serving as a default for complex skill development, it will likely become more experiential, social, and high-stakes. Think of in-person learning not as the primary delivery mode but as a capstone experience—a space for high-impact collaboration, mastery validation, and deep human connection. For example:
Scenario-Based Mastery Assessments – After completing AI-driven simulations and real-world practice sessions, learners might come together in physical settings to apply their skills in high-stakes, team-based problem-solving exercises.
AI-Augmented Coaching Sessions – Rather than replacing human coaching, AI will enhance it by offering rich performance data and insights, allowing in-person sessions to focus on nuanced skill refinement rather than basic instruction.
Hybrid Work-Based Learning – Imagine AI coordinating a seamless blend of asynchronous preparation, live virtual collaboration, and hands-on, in-person execution—optimising when, where, and how different learning experiences occur based on learner performance and needs.
Rather than existing as separate silos, online, in-person, and blended learning will likely converge into a continuous, AI-enhanced learning experience that dynamically shifts between digital and physical environments. The real question is no longer whether online learning can replace in-person training but how the two will intertwine to create more effective, intelligent learning ecosystems.
Rethinking & Rebuilding Online Learning Systems
The disruption caused by the rise of agentic AI, while uncomfortable, creates an unprecedented opportunity to build scalable online learning experiences that actually work.
To be clear: agentic AI hasn’t broken online learning—it’s exposed systems that have been broken for decades. The path forward requires us to focus not on returning to the classroom or creating new consequences for using AI agents to complete courses.
Instead, the future lies in redesigning online learning around real-world skill application, authentic competency assessment and continuous reinforcement. This means moving beyond static, content-driven models toward AI-powered, workflow-integrated learning ecosystems that ensure knowledge is not just consumed, but actively used and refined in real work contexts.
As these AI-powered systems evolve, they’ll transform not just how we learn, but what it means to be an educated professional in the AI age. The future belongs to platforms that recognise AI agents not as adversaries to block, but as mirrors revealing our pedagogical failings—and partners in building something genuinely better and potentially transformative.
Happy innovating!
Phil 👋
PS: If you want to get hands-on with AI supported by me and a cohort of people like you, apply for a place on my AI & Learning Design Bootcamp.