Dr Ravinder Tulsiani argues that generative AI demands more than basic tool training. L&D must map AI to real work, build judgement and validation skills, update competency frameworks and measure performance impact. The opportunity is to move beyond content delivery and redesign capability for an increasingly AI-enabled workplace at scale.
Generative AI is reshaping how work happens. L&D professionals need to look at what exactly is changing about work and how we should we respond. A major Microsoft Research study offers some clarity. In Working with AI: Measuring the Applicability of Generative AI to Occupations, researchers analysed 200,000 anonymised real-world AI conversations and mapped them to occupational work activities using the O*NET framework.
Rather than speculating about future disruption, the study measured where AI is already being used successfully. The findings should prompt a serious rethink in L&D.
The study found that AI performs best in activities involving:
- Writing and editing
- Explaining processes and policies
- Gathering and synthesising information
- Communicating technical details
- Preparing instructional or reference material
- Responding to enquiries
AI is highly effective at creating, processing and communicating information, which affects most roles as documentation, reporting, scheduling, compliance communication and customer interaction are widespread.
This is not just a technology trend but a structural shift in knowledge work
The researchers conclude that AI applicability cuts across sectors because most occupations contain some information component. This is not just a technology trend but a structural shift in knowledge work.
Not just automation but augmentation
One of the most helpful distinctions in the research is between:
- AI assisting workers in their tasks, and
- AI performing parts of those tasks itself
In some roles, AI functions as a collaborator, such as drafting, summarising or explaining. In others, it takes over discrete components of work, freeing humans to focus elsewhere. For L&D, this distinction is critical.
If AI collaborates, employees must learn how to:
- Prompt effectively
- Evaluate output
- Refine responses
- Apply judgement
If AI performs parts of a task independently, employees must learn how to:
- Delegate appropriately
- Oversee and validate
- Detect errors
- Intervene when necessary
The capability requirement shifts from “knowing how to use the tool” to knowing when and how to trust it. That is a much deeper learning challenge.
The measurement gap in AI training
The research assessed AI impact by examining:
- Task completion success
- Scope of AI capability within work activities
- Applicability across occupations
It did not measure course completions or satisfaction scores. Yet many organisations evaluating their AI programmes still rely on:
- Participation rates
- Learning hours
- Survey feedback
There is a disconnect. If AI is changing how decisions are made, how work is delegated and how outputs are produced, then impact must be measured at the level of performance.
For example:
- Are decisions faster without increasing risk?
- Has rework decreased?
- Are escalations more accurate?
- Is communication clearer and more consistent?
If L&D cannot demonstrate changes at this level, AI training risks becoming a tick-box exercise rather than a capability shift.
Democratising expertise or amplifying error?
The researchers suggest that AI may reduce performance gaps by broadening access to expertise. When employees can access high-quality explanations or structured outputs instantly, capability barriers lower.
However, this benefit depends on foundational knowledge. Without subject understanding, employees cannot evaluate AI output effectively. In those cases, AI can amplify misunderstanding rather than improve performance.
For L&D, this creates a tension. On one hand, AI can accelerate capability development. On the other, it increases the importance of:
- Critical thinking
- Risk awareness
- Ethical judgement
- Domain fluency
AI literacy without domain literacy is insufficient. Organisations must resist the temptation to treat AI as a shortcut around foundational skill development.
Where AI still struggles
The study also notes areas where AI is less capable:
- Physical or manual tasks
- Highly context-specific judgement
- Complex real-world decision-making with incomplete information
This highlights another training priority: boundary management. Employees must learn:
- When AI output requires verification
- When context makes generic responses unsafe
- When escalation is necessary
- Where legal or compliance risk increases
This is not about fear-based messaging. It is about calibrated confidence.
What should L&D do differently?
The implications are practical and strategic.
1. Map AI to work, not roles
Instead of launching broad “AI skills” programmes, start with work activity analysis. Where in daily workflows does AI meaningfully intersect? Examples:
- Drafting client proposals
- Producing reports
- Preparing training materials
- Answering standard customer enquiries
- Summarising research
Identify high-frequency and high-impact tasks. Design interventions around those moments. The goal is not awareness but workflow improvement.
2. Design for judgement, not just knowledge
If AI changes task execution, training must change cognitive capability. Replace passive instruction with:
- Scenario-based exercises
- Decision simulations
- Delegation boundary cases
- Output evaluation practice
Employees should practise:
- Comparing AI outputs
- Spotting hallucinations
- Adjusting prompts
- Assessing risk implications
Capability grows through practice, not explanation.
3. Update competency frameworks
Traditional frameworks focus on:
- Communication
- Technical skill
- Leadership
- Collaboration
Few explicitly include:
- Human–AI collaboration
- Prompt refinement
- Output validation
- Risk calibration
- Escalation judgement
These must now be embedded across professional standards, not isolated in digital literacy modules. AI capability is becoming a baseline expectation.
4. Reposition L&D strategically
AI can already:
- Draft training materials
- Summarise policies
- Explain procedures
- Create structured content
If L&D’s value proposition centres on content production, it will narrow. The opportunity lies in:
- Diagnosing capability gaps
- Designing practice environments
- Building simulation-based learning
- Measuring behavioural shift
- Advising on work redesign
The strategic role becomes architect of capability systems, not producer of courses.
The bigger organisational question
The Microsoft study does not claim that jobs will disappear wholesale. Instead, it shows that AI overlaps with meaningful portions of work activities across most occupations.
History suggests that when task composition changes, roles evolve. The question is whether L&D evolves with them. If AI shifts:
- How information is processed
- How decisions are supported
- How communication is generated
- How knowledge is accessed
Then L&D must shift from content delivery to performance engineering. The organisations that thrive will not be those that trained everyone quickly on AI tools. They will be those that redesigned capability around AI-enabled work.
A moment of choice for L&D
AI is not replacing learning, it is raising the bar. Employees must think more critically, not less and exercise better judgement, not less. The real competitive advantage will not come from AI access, but from how effectively people apply it.
L&D sits at the centre of that challenge. The question is whether we treat AI as another topic to cover — or as a catalyst to rethink how capability is built, measured and sustained. That decision will shape the profession for the next decade.
Dr. Ravinder Tulsiani is Principal, Interim Enterprise Capability & L&D Advisor at Catalyst Learning
