As AI transforms workplace productivity, traditional measures of learning success risk overlooking a growing skills gap. Dr Gleb Tsipursky explores why time to proficiency should become L&D’s defining AI metric, helping organisations assess independent judgement, preserve essential learning experiences and ensure employees develop expertise rather than increasing dependence on technology.
The most urgent AI problem for L&D may be hiding inside a productivity win. When software drafts the routine email, summarises the case, prepares the first analysis and answers the predictable customer question, employees finish faster. They also lose repetitions that once helped them learn the job.
AI skills are an L&D issue far beyond prompt writing
Stanford University’s paper on employment effects of AI found that employment among workers ages 22 to 25 in highly AI-exposed occupations now stands about 19% below where it would be if it had kept pace with less-exposed occupations, with the adjustment appearing primarily through reduced hiring. The same work also points toward a distinction between codified knowledge and experience-based knowledge. So, AI skills are an L&D issue far beyond prompt writing.
AI influence
Training Journal has already framed the profession’s measurement challenge well. Its 2026 readiness work asks whether L&D teams have access to decision-makers, usable data, permission to test and support to scale what works. Those are the right questions for L&D impact, but AI changes what impact should mean. If routine work previously doubled as practice, then measuring completions, satisfaction or even immediate productivity can miss a deteriorating capability pipeline. A faster employee today can still become a weaker independent performer six months from now.
The evidence does not support a simple story of AI making people less capable. In a study of 5,172 customer-support agents, researchers found that generative AI increased productivity by 15% on average, with especially large gains among less experienced and lower-skilled workers. Newer agents also moved down the experience curve faster. That is a real benefit for workplace learning.
A newer Harvard Business School analysis found a limit: AI helped novices with abstract, codifiable knowledge much more than with tasks requiring context-bound execution and lived experience. The researchers’ conclusion that AI can provide the map without supplying the terrain captures the new L&D problem: AI can compress parts of the learning curve while leaving other parts untouched.
Time for L&D
That is why L&D needs a competence clock. The clock measures elapsed time from first exposure to a task until an employee can make a sound decision independently, including when AI is available. Traditional time to proficiency measures how long until someone reaches acceptable performance. The competence clock makes that idea AI-specific by separating visible output from underlying judgment. It asks whether the worker can verify the result, recognise an exception, explain the reasoning and know when escalation is required.
The first hand on the clock should measure verification accuracy. Microsoft researchers surveyed 319 knowledge workers and found that higher confidence in generative AI was associated with less critical-thinking effort. They also found that critical thinking shifts when AI enters the workflow, toward verification, integration and task stewardship. That makes AI verification a skill that can be tested. Give employees AI-generated outputs containing realistic errors, omissions or weak assumptions, then measure whether they catch them, how quickly they catch them and whether they can explain why they matter.
The second hand of the clock should measure exception handling. AI often performs impressively until a task falls outside its effective frontier. In the well-known experiment with Boston Consulting Group consultants, AI users performed faster and better on tasks within the technology’s capabilities, but on a complex task outside that frontier they were substantially less likely to reach the correct answer. The practical lesson for human judgment is that workers need practice identifying boundary conditions. L&D should build simulations around cases that look routine at first but contain one detail that changes the decision.
The third clock hand should track dependence on coaching. A learner who produces acceptable work only after a manager repairs the prompt, spots the omitted constraint or explains why the AI output does not fit the situation has not yet reached independence.
Recent Anthropic research found that more experienced AI users attempted higher-value tasks and were more likely to elicit successful responses, suggesting that effective AI use itself develops through experience. That gives independent judgment an observable progression. Count how often a learner needs intervention, classify the intervention and track whether those interventions fall over repeated cases.
Employees increasingly supervise sequences of work rather than single outputs
The fourth hand should measure transfer. Can the employee handle a novel case that differs from the examples used in training? This matters more as AI workflows become more agentic. Anthropic’s June 2026 Economic Index describes a shift toward longer-running agentic tasks, which means employees increasingly supervise sequences of work rather than single outputs. Strong learning metrics should therefore include performance on unfamiliar scenarios, where the worker must decide what to delegate, what to inspect and where to interrupt the system.
These measures also solve a political problem inside L&D. Training Journal’s wider 2026 influence research found that only 3% of practitioners considered themselves fully prepared to turn evidence into action, while 41% reported low readiness. Access to decision-makers was identified as a major constraint. A competence clock gives L&D business performance metrics leaders already understand: risk, speed and independence. Instead of saying 900 people completed an AI course, the function can say the median employee now reaches independent judgment in eight weeks rather than twelve, catches 92% of seeded errors rather than 74%, and needs half as many manager interventions.
Those numbers should never become generic enterprise benchmarks. Different roles face different failure costs and learning curves. A marketing employee reviewing draft copy does not need the same verification threshold as a financial analyst, engineer or compliance professional. The point is to define acceptable independent performance with the business before training begins. LinkedIn’s 2025 Workplace Learning Report found that 49% of learning and talent professionals said executives were concerned employees lacked the skills needed to execute business strategy. A competence clock connects early-career development directly to that strategic concern.
It’s not ALL about AI
L&D should also resist the temptation to turn this into an AI-only curriculum. The World Economic Forum’s Future of Jobs research says 59 out of every 100 workers would need training by 2030 under employers’ expectations, while skill gaps remain the most frequently identified barrier to transformation. The OECD’s July 2026 paper emphasises critical thinking and broader human capabilities alongside digital and AI literacy. The competence clock should therefore include subject-matter judgment, communication and escalation behaviour, not simply tool fluency.
There is another reason to move quickly. OECD analysis of training supply found that only a small share of available courses in several countries contained AI-related content, even as a large share of jobs faced meaningful AI exposure. At the same time, Microsoft’s cross-industry field experiment found that generative AI changed some individual work patterns, such as time spent on email, more readily than activities requiring coordination. That suggests AI-assisted work can change faster than the systems around it. L&D cannot assume job design, manager coaching and developmental assignments will automatically catch up.
The solution is to redesign learning at the same time organisations redesign work. When AI removes a routine task, identify what learning function that task served. If it taught pattern recognition, create rapid comparison exercises. If it exposed juniors to exceptions, build an exception library. If it prompted manager feedback, preserve scheduled review. If it helped employees see the consequences of a decision, use case simulations that show downstream outcomes.
Levelling the playing field
OECD research on training for AI transitions makes clear that access to relevant upskilling remains uneven. Employers therefore need to create deliberate practice inside work, rather than waiting for the training market to solve the problem. A practical rollout can start with one role and one high-volume workflow. Establish a baseline before changing training: error detection, exception decisions, manager interventions and time to independent sign-off. Then introduce AI, replace the lost repetitions deliberately and measure the same indicators every two weeks. The objective is a shorter competence clock without a rise in hidden dependence. That gives L&D a testable hypothesis rather than a broad promise about AI capability.
The broader goal of AI adoption at work should be to move people toward higher-value judgment faster without hollowing out the experience that makes such judgment possible. That requires L&D to stop treating every automated task as pure efficiency. Some tasks are also classrooms. When those classrooms disappear, the practice has to reappear somewhere else by design.
A competence clock makes that design measurable. It gives L&D a way to prove that employees are becoming faster without becoming more dependent, more productive without becoming less discerning, and more capable of using AI without surrendering the judgment the organization will still need when the easy cases are gone.
Dr Gleb Tsipursky is CEO Disaster Avoidance Experts and author of eight books, including The Psychology of AI Adoption at Work
If this article piqued your interest, you’ll probably find value in our conference!
Find out about keynote speaker Erica Farmer, her opening session on the
AI Fire Hydrant and what it means for productivity, performance and L&D
Don’t miss 50% off your ticket for the first 30 only!
