Content bloat vs capability
Why AI course generation is failing enterprise proficiency
of enterprise organizations deploying generative AI authoring tools over the last 18 months experienced zero measurable improvement in employee role proficiency.
That is a tough pill to swallow for a lot of people. Content production is not learning, and course completion is not capability.
This edition looks closely at the gap between the sheer volume of AI-generated content and the actual transfer of human capability. It also examines the newly ratified IEEE agentic learning standard, and the frontline manager coaching deficit that is bottlenecking enterprise reskilling efforts.
If you are looking at your own learning management system, you probably know the tension. The data shows where the friction hurts the most.
The value of content has moved
An algorithm can generate a module in 30 seconds
Generative AI is reshaping how work gets done, and it is breaking traditional corporate training models. Industry analysis recently laid out a reality that many in the field are still avoiding: the commercial value of raw content creation has plummeted.
For decades, learning departments justified their existence by pointing to the volume of courses they produced. That reality requires a total role and value shift. The learning function has to evolve into a strategic adviser focused on the human side of AI transformation, which takes a completely different skill set. It is no longer about writing the course. It is about orchestrating how expertise is shared across an organization.
When employees can build their own performance support tools, the central training department loses its monopoly on knowledge distribution entirely.
Printing press. Learning departments justified their existence by the volume of courses they produced. The function writes the manual.
Air traffic controller. L&D orchestrates how expertise is shared across the organization, ensuring the right knowledge lands safely with the right person at the exact moment of need.
What faster authoring does and does not reach
The module exists. What has to happen next?
Content production is not learning, and course completion is not capability. Generation is the first stage of a longer path, and the edition's reporting attaches different evidence to each of the later stages.
Content generation
Established. An algorithm can generate a module in 30 seconds. The commercial value of raw content creation has plummeted, which is the point at which this path starts rather than ends.
Instructional design
Faster course authoring does not equal better instruction. Because of AI authoring tools, content catalogs across the benchmark organizations expanded by an average of 310 percent.
Learner experience
Learner fatigue and course abandonment spiked by 42 percent. Systems were flooded with generated noise that nobody had time to consume.
Effortful practice
Short, scenario-based decision prompts spaced over 30 days produced 73 percent retention at 60 days. Passive video clips showed an 81 percent decay in concept recall after 14 days.
Feedback
Capability is a product of deliberate application and feedback. A framework in Training Magazine asks for feedback on specific behaviors rather than personality traits, delivered in time, as a two-way dialogue that builds self-regulation.
Transfer to the job
Consistent adoption of new tools reaches 49 percent when managers model them, against 22 percent when they do not. Employees prioritize immediate work over a mandated protocol their own leadership does not use.
Performance
Platform engagement is not workplace performance. The 26 percent of organizations that put their investment into internal skills graph validation and contextual, on-the-job support saw a 38 percent increase in successful internal role mobility.
Content bloat versus real proficiency
The pile and the column
The Learning and Performance Institute and Deloitte Human Capital just published a benchmark report looking at 1,450 enterprises across 18 countries. What they found is the textbook definition of content bloat versus real proficiency. Because of AI authoring tools, content catalogs across these organizations expanded by an average of 310 percent.
If you know anything about learning science, you saw what would follow: learner fatigue and course abandonment spiked by 42 percent. Systems were flooded with generated noise that nobody actually had time to consume.
Here is the critical contrast. Twenty-six percent of the surveyed organizations ignored the content generation hype. Instead, they directed their investments toward internal skills graph validation and contextual, on-the-job support. That cohort saw a 38 percent increase in successful internal role mobility.
Choose where the investment goes and watch what each structure becomes.
Faster course authoring does not equal better instruction. If anything, it actively alienates the workforce by overwhelming people with modular trivia. Capability is a product of deliberate application and feedback; it does not come from library size.
These are figures reported by a benchmark of 1,450 organizations. They describe patterns across organizations, not a guaranteed effect of any single decision.
The technology has to track more than completions
Making the AI tutor measurable and governable
Technical standardThe IEEE Learning Technology Standards Committee, in partnership with The Learning Guild, just ratified the IEEE P2997 standard: the Standard for Agentic Learning Experience Architecture and xAPI-State Interoperability. It is a substantial upgrade to xAPI interoperability for AI tutors.
Before this, if an organization piloted an AI coaching bot, that bot lived in a silo, completely isolated, with proprietary chat logs that could not communicate with core enterprise HR systems. The new standard establishes a standardized telemetry vocabulary designed to capture prompt scaffolding depth, cognitive struggle indices, and task completion confidence, all while maintaining strict privacy boundaries. It logs the step-by-step diagnostic reasoning of the learner without storing proprietary personal data, and it gives instructional designers verifiable data on exactly where a learner hesitates and what concepts trigger confusion.
What does a cognitive struggle index actually look at?
A tutor session is a trail of questions heading toward a learning objective. Switch on the signals the standard's vocabulary can draw on.
It is not just about dwell time or how long someone stares at a screen. The AI analyzes the nature of the prompts the learner inputs, which makes conversational AI tutoring measurable and governable across platforms.
Where the human has to stay
It arrives at a critical regulatory moment. The European Commission just issued binding operational guidelines regarding the EU AI Act, formally classifying workplace algorithmic assessment tools as high risk under Annex III. If an organization uses AI to grade certifications or measure employee competence, it must maintain strict audit trails and mandatory human oversight.
Strict audit trail maintained across the whole path
Specialist skills and the physical cost of AI
Scope decides whether a tool is support or bloat
Product announcementGoogle just launched an AI for Healthcare Professionals certificate. Data indicates 80 percent of physicians are already using AI professionally. But healthy skepticism is warranted: how does a self-paced course that takes under five hours ensure genuine capability transfer for a medical doctor? That sounds suspiciously like the content bloat described earlier in this edition.
The distinction lies in the rigorous focus of the curriculum. The certificate does not attempt to teach clinical diagnostics or medical theory. It is strictly focused on practical administrative workflows.
Place each task where the certificate puts it
Drag a task into a zone, or select it and then choose a zone.
Inside the scope: extracting action items from communications, organizing schedules, and translating dense medical concepts into patient-friendly summaries. Outside it: clinical diagnostics and medical theory.
The cost nobody puts in the curriculum
A routine scheduling query, three ways to answer it
Guide and energy agency findingsThe physical cost of running AI models is almost completely absent from corporate digital literacy programs. Sustained Futures recently released a guide on climate fluency and AI skills highlighting inference: the actual process of a trained AI model producing an output from a new input. Generative AI does not merely retrieve a file. It calculates probabilities across billions of parameters for every single word or frame.
The International Energy Agency points out that while simple text generation is becoming more efficient, video generation and complex agentic reasoning require hundreds or thousands of times more energy per query.
L&D must teach computational energy awareness as a core digital skill, helping employees select the appropriate tool for the cognitive task at hand.
The frontline manager coaching deficit
Organizations are treating managers like network routers
Leadership studyIf manager behavior dictates employee adoption, any coherent learning strategy must examine the actual capacity of those frontline managers. Gartner just released its 2026 Global Leadership Study covering 2,400 enterprises. Eighty-two percent of executives expect frontline managers to lead the upskilling and reskilling of their teams. Yet those same managers spend less than 6 percent of their working time on talent development: roughly 2.8 hours per week. They are consumed by administrative duties and operational reporting.
If you push a gigabit data stream through a router that can only handle 10 megabits, the system crashes, regardless of how valuable the data is. The operational bandwidth simply does not exist. Organizations expect frontline managers to process complex reskilling demands and distribute capability to their teams without that bandwidth, and the signal drops.
82% of executives expect frontline managers to lead the upskilling and reskilling of their teams.
Less than 6% of working time goes to talent development, roughly 2.8 hours per week.
What reaches the team depends on how the intervention enters the manager's week.
Where does the coaching fit in the week?
Talent development, less than 6 percentThe weekly project check-in the manager already runs
The two outcomes are different measures. One is retention of the training after 60 days, the other is an increase in team retention. They are reported side by side, not as one comparison.
Traditional multi-day leadership workshops, pulling a manager off the floor for three days to study coaching theory, yield only an 11 percent retention rate after 60 days because the training is disconnected from daily reality. The behavior does not stick. Conversely, organizations implementing what Gartner calls in-flow enablement reported a 44 percent increase in team retention. In-flow enablement means short, weekly 10-minute coaching prompts linked directly to existing project check-ins. Rather than asking managers to schedule a separate coaching hour, you provide a targeted prompt to use during the meeting they are already running.
A framework outlined in Training Magazine explains how those brief conversations must be structured: feedback must focus on specific behaviors rather than personality traits, it must be timely, and it must foster a two-way dialogue that builds self-regulation in the employee. You cannot ask a stressed manager to recall a coaching model from an offsite six months ago. You must supply the exact conversational framing and behavioral cues in the moment of need, reducing cognitive load so they can focus on the human interaction. That is why event-based workshops have to be replaced by structured performance support.
The cognitive science of instructional design
Smooth to watch is not the same as hard to forget
Empirical studyBecause managers are pressed for time and generative AI is flooding enterprises with modular noise, the engineering of learning materials has to be scientifically grounded. An empirical study by MIT Sloan Management Review and Harvard Business Publishing tracked 1,800 corporate professionals to measure long-term knowledge retention across different models, comparing passive video microlearning against structured spaced retrieval practice.
This illustrates the fluency trap. Passive video snippets often function as corporate distraction masquerading as development. A learner watches a polished video, experiences cognitive ease because the delivery is smooth, but fails to do the effortful work required to encode the knowledge. AI personalization claims mean very little if they lack effortful retrieval practice. Cognitively, forcing the brain to retrieve information to solve a scenario strengthens neural pathways, signaling that the concept is critical for practical functioning. Passive video washes over short-term memory without leaving a lasting trace.
Match the medium to the job
A recent instructional framework published by eLearning Industry clarifies format utility based on cognitive load. L&D must stop defaulting to video simply because it appears premium. Choose a medium for each need.
An employee needs the four steps to escalate a customer issue.
A manager needs to see what a challenging managerial conversation looks like.
A team wants to explore how an expert reasons through a problem, and to reflect on it.
A new starter needs a walk through a software workflow.
How knowledge is structured and accessed across platforms
Skill credit has to cross a bridge built by colleagues
Product integrationA significant platform integration illustrates where enterprise architecture is heading. Degreed and 360Learning have established a cross-platform peer learning protocol. Degreed acts as the centralized skill taxonomy and credential tracking engine, while 360Learning serves as the collaborative platform where internal subject-matter experts author operational playbooks.
How does this prevent someone from faking a skill, or generating superficial course modules with AI to claim credentials? The integration includes a structured peer validation rubric. Operational playbooks created by practitioners must be reviewed and validated by verified colleagues within the organization before skill credit is recorded in Degreed's centralized skill graph. The system relies on social proof and technical peer review.
A playbook waits to earn skill credit. Send it across and see what happens at the bridge.
This structural approach addresses multiple challenges at once. It mitigates the skills validation crisis by relying on decentralized knowledge capture vetted by peers rather than multiple-choice quizzes. It supports time-constrained frontline managers by offloading training to experienced senior practitioners. And it anchors informal learning back into a validated enterprise skills taxonomy, preventing unchecked content bloat.
What holds it together
Capability does not come from the velocity of content
Looking at the broader landscape, one central conclusion stands out: organizational capability does not stem from the velocity of content generation. It does not matter if AI can author a course in 30 seconds. True capability develops through deliberate practice, rigorous peer validation, and effortful cognitive retrieval integrated directly into daily work.
Technology, whether an agentic AI tutor or an enterprise platform integration, is valuable only when surgically targeted at specific operational friction. If it fails to change behavior in practice, it remains expensive theater.
Are you building verifiable capability, or are you simply running an efficient printing press?