AI Workforce: Upskilling or Reskilling? How Companies Are Preparing Workers

Billions Spent. Skills Gaps Growing.

U.S. corporate training expenditures reached $102.8 billion in 2025, up 4.9 percent from the prior year, according to Training Magazine’s 2025 Industry Report. Companies spent an average of $874 per learner — with small firms spending over $1,000 per person and large enterprises averaging $468. Globally, the corporate training market is estimated at over $380 billion and growing toward $487 billion by 2030.

The spending keeps rising. The results do not keep pace.

87% of executives and managers say their organizations either face skill gaps already or expect them within the next five years, according to McKinsey research. Only 32% of business leaders report achieving healthy change adoption by employees, according to Gartner’s 2025 research. The World Economic Forum’s Future of Jobs Report 2025 estimates that 39% of workers’ key skills will need to change by 2030 — down from 44% in 2023, but still a massive transformation ahead.

Something is deeply wrong. Companies are spending hundreds of billions on training while simultaneously reporting that their workforce is not ready for the AI transition. The problem is not the amount of money. It is how the money is spent.

The Corporate Training Industry Is Broken

Traditional corporate L&D suffers from structural problems that money alone cannot solve.

Problem 1: Content Without Context

Most corporate training is generic — off-the-shelf courses from vendors like LinkedIn Learning, Coursera for Business, or Udemy Business. An employee takes a 4-hour course on “Introduction to AI,” checks a completion box, and returns to their desk. The course was generic, the examples were irrelevant to their actual job, and there was no connection between the training content and their daily work.

Research on memory retention shows that without reinforcement, people forget up to 70% of new information within 24 hours and up to 90% within a week, according to studies replicating the Ebbinghaus forgetting curve. A course completed in January and never reinforced is worthless by March.

Problem 2: Compliance Theater

A significant portion of corporate training spending goes to mandatory compliance training — anti-harassment, data privacy, security awareness, code of conduct. These courses are legally required but pedagogically weak: click through slides, pass a multiple-choice quiz, repeat annually. They satisfy legal requirements without changing behavior.

When AI upskilling is treated like compliance training — mandatory modules that employees complete to check a box — the result is the same: completion without comprehension, certification without capability.

Problem 3: Misalignment Between Training and Work

L&D departments often operate disconnected from business units. Training programs are designed based on general industry trends rather than the specific skills gaps of the organization. A retail company’s L&D team might purchase an “AI for Business” course library while the company’s actual need is for employees who can use the specific AI tools being deployed in inventory management, customer service, and demand forecasting.

85% of L&D leaders agree there will be a surge in skills development needs due to AI and digital trends in the next three years, according to a Gartner 2024 survey. Yet many L&D departments remain stuck delivering content that does not match what their organizations actually need.

Problem 4: The Time Problem

Employees do not have time for training. Industry research by Josh Bersin found that employees have an average of roughly 24 minutes per week for learning — in a 40-hour work week. That is 1% of working time. No meaningful skill transformation can occur in 24 minutes per week.

The time problem is structural: managers are evaluated on output (features shipped, targets met, projects delivered), not on their team’s skill development. Allocating time for training reduces short-term output. In performance-driven cultures, learning loses.

Upskilling vs. Reskilling: The Distinction Matters

The terms are often used interchangeably, but they represent fundamentally different challenges.

Upskilling means enhancing an employee’s existing skills to incorporate new tools and methods. A software developer learning to use AI coding tools, a marketing manager learning to use AI content generation, or a data analyst learning to build ML models — all of these are upskilling. The employee stays in their current role but becomes more capable.

Reskilling means training an employee for a fundamentally different role because their current role is being eliminated or radically transformed. A call center agent transitioning to a customer success role (because AI handles routine inquiries), a data entry clerk transitioning to data quality management, or a truck driver training for logistics coordination in anticipation of autonomous vehicles.

The WEF Future of Jobs 2025 report found that upskilling is the most common workforce strategy, with 85% of employers planning to adopt it. Over half of the global workforce is expected to undergo reskilling or upskilling to meet changing demands by 2030.

The scale challenge:

• Upskilling is relatively straightforward: build on existing knowledge, introduce new tools, practice in the existing work context. Organizations can upskill at scale with moderate investment.
• Reskilling is hard: it requires fundamental change in skills, identity, and career trajectory. It takes 6-18 months, requires significant investment, and has lower success rates. Most organizations significantly underestimate the difficulty and cost of reskilling.

What Actually Works: Evidence-Based Approaches

Research on effective skill development identifies several principles that successful programs share.

1. Learning in the Flow of Work

The most effective skill development happens in the context of actual work, not in separate training sessions. Josh Bersin (HR industry analyst) coined the term “learning in the flow of work” — the idea that training should be embedded in tools, workflows, and daily tasks rather than extracted into separate courses.

Examples:

• AI coding assistants that explain their suggestions, teaching the developer as they code
• In-tool tutorials that appear when an employee first uses a new feature
• Working alongside an AI tool that demonstrates techniques in the context of the employee’s actual project
• Just-in-time learning resources surfaced when an employee encounters a specific problem

When organizations adopt this approach, Gartner research shows employees apply 75% of new skills learned — far exceeding the retention rates of traditional training.

2. Cohort-Based Learning with Accountability

Self-paced online courses have completion rates of 5-15%. Cohort-based programs — where a group of employees learns together on a fixed schedule with assignments, discussions, and accountability — have completion rates of 70-90% and significantly better skill outcomes, according to multiple studies. Research also indicates a 69% greater chance of retaining information in cohort programs compared to self-paced learning.

Effective cohort models:

• Small groups (10-20 people) from the same organization or team
• Fixed schedule (weekly sessions over 6-12 weeks)
• Applied projects (solve a real business problem using the new skills)
• Manager involvement (managers attend kickoff, review final projects, and allocate time for learning)
• Peer accountability (group members support and challenge each other)

3. Applied Projects, Not Assessments

Multiple-choice quizzes measure recall, not capability. Skill development should be assessed through applied projects: build something, solve a real problem, demonstrate capability in context.

AT&T’s massive reskilling program — the “Future Ready” initiative that invested $1 billion to retrain 100,000 employees from legacy telecom roles to cloud and software engineering — succeeded in part because employees worked on real internal projects as they learned. By 2020, over 100,000 employees had completed retraining, allowing the company to fill crucial roles internally and reduce reliance on external hiring.

4. Managerial Support and Protected Time

The single strongest predictor of whether employees will develop new skills is their manager’s support. Managers who allocate protected time for learning (blocking calendar time, reducing sprint commitments, setting learning goals), who model continuous learning themselves, and who create psychological safety for skill experimentation enable skill development. Managers who see training as time away from “real work” kill it.

Microsoft’s growth mindset culture: Under CEO Satya Nadella, Microsoft explicitly shifted from a “know-it-all” culture to a “learn-it-all” culture, a transformation driven across 130,000+ employees in partnership with CHRO Kathleen Hogan. The idea was inspired by Carol Dweck’s research on growth mindset. The cultural shift — from managers who demonstrate knowledge to managers who demonstrate curiosity — created organizational permission for continuous learning. Under this transformation, Microsoft’s market value grew from roughly $300 billion in 2014 to over $2.5 trillion.

What Forward-Thinking Organizations Do Differently

Amazon’s Upskilling 2025

Amazon invested $1.2 billion to provide 300,000 U.S. employees access to free training programs through nine initiatives including:

• Machine Learning University: Internal courses teaching ML to software developers
• AWS re/Start: 12-week bootcamp transitioning non-technical employees into cloud operations roles
• Amazon Technical Academy: Program converting non-technical employees into software developers
• Career Choice: Pre-paying 95% of tuition for courses in high-demand fields

By the program’s reporting period, over 70,000 employees had actively participated in these programs. The programs that worked best were those with clear career pathways — complete this program and move into this specific role — and manager-supported time allocation.

JPMorgan’s AI Training

JPMorgan Chase mandated AI training for all new employees starting in 2024, with the broader goal of making AI work for every single employee across its 300,000+ workforce. The initiative included:

• Prompt engineering training integrated into onboarding for all new hires
• “AI Made Easy” program that engaged tens of thousands of employees in AI literacy
• LLM Suite deployed to 200,000 employees within eight months of launch
• 450+ AI use cases developed across the organization

The key success factor: training was tied to specific tools the company was deploying — particularly its internal LLM Suite — not generic AI education. JPMorgan acknowledged that AI is core to its $18 billion annual technology investment.

Singapore’s SkillsFuture

Singapore’s national SkillsFuture program provides every citizen turning 25 with SGD 500 in training credits, with an additional SGD 4,000 for mid-career workers turning 40. The program subsidizes employer training costs and partners with universities and industry to develop relevant curricula.

The program is government-funded, employer-integrated, and individually accessible — a model that addresses the collective action problem where individual companies under-invest in training because employees might leave. The SkillsFuture Level-Up Programme provides additional support for mid-career transitions.

The Micro-Credentials Revolution

One of the most significant shifts in corporate learning is the rise of micro-credentials — short, focused certifications that validate specific skills rather than broad degrees.

The Coursera CEO has declared that getting hired in 2026 is increasingly about micro-credentials. The data supports this: 90% of employers prefer candidates with a micro-credential on their resume, and 87% have hired at least one employee with micro-credentials in the past year, according to a 2025 Micro-Credentials Impact Report. Companies implementing micro-credential programs report 32% faster upskilling rates compared to traditional training approaches.

This connects directly to the broader shift toward skills-based hiring. While 85% of employers claim to prioritize skills over degrees, a Harvard Business School and Burning Glass Institute analysis found that fewer than 1 in 700 actual hires are affected by degree-requirement removal policies. The gap between stated intent and actual practice remains wide — but the direction of travel is clear.

The AI Catch-22

The deepest challenge in AI upskilling is that the skills needed are changing faster than training programs can be developed. A course on “ChatGPT for Business” developed in 2023 is already outdated in 2026. The specific tools, interfaces, and capabilities change quarterly.

This argues for teaching meta-skills rather than tool-specific skills:

• Learning agility: The ability to quickly learn and adapt to new tools and processes
• Critical evaluation: The ability to assess AI output quality, identify errors, and verify accuracy
• Problem decomposition: The ability to break complex problems into tasks that can be delegated to AI tools
• System thinking: Understanding how AI tools interact with existing workflows, data, and processes
• Ethical reasoning: Evaluating the appropriate and inappropriate uses of AI in professional contexts

The WEF’s 2025 report confirms this direction: the top growing skills are not just technical (AI, big data, cybersecurity) but also include creative thinking, resilience, flexibility, curiosity, and lifelong learning. The organizations that invest in building these meta-capabilities — rather than chasing tool-specific training — will be best positioned for the AI transition.

Frequently Asked Questions

Sources & Further Reading

• Training Magazine — 2025 Training Industry Report
• McKinsey — Beyond Hiring: How Companies Are Reskilling to Address Talent Gaps
• Gartner — 85% of L&D Leaders See Surge in Skills Needs (2024)
• Gartner — Only 32% Achieve Healthy Change Adoption (2025)
• World Economic Forum — Future of Jobs Report 2025
• Josh Bersin — Learning in the Flow of Work
• Amazon — Upskilling 2025
• JPMorgan Chase — Preparing the Workforce for AI
• Singapore SkillsFuture
• Ebbinghaus Forgetting Curve — Replication Study (PMC)
• Cohort-Based Learning Statistics — Learnopoly
• Coursera — 2025 Micro-Credentials Impact Report
• Harvard/Burning Glass — State of Skills-Based Hiring
• Fortune — Microsoft Culture Transformation Under Nadella
• CNBC — AT&T’s $1 Billion Reskilling Gambit
• LinkedIn — 2025 Workplace Learning Report