Generative AI refers to a class of AI models trained on very large datasets to learn the underlying statistical patterns in that data, then use those patterns to produce new, original content, text, images, audio, or code, rather than simply analyzing or classifying existing data. In an HR context, this shows up as tools that draft a job description, summarize a performance review, or answer an employee’s benefits question in natural language.
Generative AI vs predictive AI vs an AI agent
What it does
HR example
Predictive AI
Outputs a score, classification, or forecast from historical data
An employee-churn model predicting resignation risk
Generative AI
Produces new content that did not exist before
Drafting a job description or summarizing feedback
Autonomously plans and executes multi-step actions
Screening resumes and scheduling interviews without step-by-step approval
Prior generations of workplace AI are typically predictive or discriminative: given structured historical data, they answer “what is likely to happen, or what category does this belong to?” Generative AI instead answers “what new content should be created?” It does not classify existing data, it synthesizes new data that did not exist before. Both approaches can use similar underlying techniques, but predictive AI supports decisions, who to hire, who might leave, while generative AI supports content creation and communication.
An AI agent goes a step further than either. Generative AI produces an output, a piece of text, an image, a block of code, in response to a prompt, then stops, leaving a human to decide what happens next. An agent, by contrast, chains that same generative capability into a longer-running system: it decides which action to take next, calls other tools when needed, and moves through several steps largely on its own, which raises a meaningfully different governance question for HR than a tool that simply drafts something for review.
Two model families behind most of today’s generative AI
Large language models are trained on massive collections of text to predict and generate coherent, contextually appropriate language, and they power chatbots, writing assistants, and summarization tools. Diffusion models are trained to generate images, and increasingly video and audio, by learning to reverse a process of progressively adding noise to training images, letting them turn random input into a coherent picture.
Where the current wave came from
The modern wave traces to the transformer architecture, introduced in “Attention Is All You Need” by Vaswani and colleagues at Google, presented at the 2017 Neural Information Processing Systems conference. The paper proposed a self-attention mechanism that made it feasible to train much larger language models efficiently, the architecture underlying nearly all subsequent large language models. Generative AI moved from a research curiosity to mainstream workplace awareness with the public launch of ChatGPT by OpenAI in November 2022, widely regarded as the inflection point that triggered broad employee experimentation well ahead of most organizations’ formal AI governance policies.
Adoption is climbing, but value lags behind
McKinsey’s March 2025 global survey found 71% of organizations now regularly use generative AI in at least one business function, up from 65% in early 2024, yet more than 80% of respondents report no material, enterprise-wide financial impact from generative AI so far.
That same disconnect between adoption and measurable payoff shows up specifically in HR, not just at the enterprise level generally. Gartner’s October 2025 survey of 2,986 employees and 114 HR leaders found 88% of HR leaders say their organization has not realized significant business value from AI tools, even though 62% of employees said it had saved them time.
Peer-reviewed research suggests the gains, where they exist, concentrate among less experienced workers. A study published in The Quarterly Journal of Economics in April 2025 followed a staggered rollout of a generative AI assistant among 5,172 customer-support agents and found productivity rose by an average of 14%, with the largest gains among less-experienced agents and little effect on the most experienced ones.
What holds generative AI back in HR specifically
Skills, not access, are the bottleneck. Gartner found only 8% of HR leaders believe their managers currently have the skills to use AI effectively.
Unofficial use runs ahead of sanctioned tools. Research from MIT NANDA found 90% of employees at studied organizations use personal AI tools for work on their own, while only 40% of those organizations had purchased official subscriptions.
Guidance is often missing. Only 7% of organizations provide employees with guidelines on how to use the time AI has freed up, leaving genuine time savings unrealized.
Social adoption barriers matter as much as technical ones. More than a third of employees with access to AI tools do not use them because their coworkers are not using them either.
Generative AI FAQs
What is generative AI?
Generative AI is a class of AI models trained on large datasets to produce new content, text, images, audio, or code, rather than analyzing or classifying existing data the way predictive AI does.
How is generative AI different from predictive AI?
Predictive AI outputs a score, classification, or forecast based on historical data, such as a model predicting employee turnover risk. Generative AI creates new content that did not previously exist, such as a drafted job description or a summarized review.
Is generative AI the same as an AI agent?
No. Generative AI produces a single output in response to a prompt and then stops. An AI agent uses generative models as part of a system that autonomously plans and executes multiple steps toward a goal, with less human oversight per step.
Why hasn’t generative AI delivered more measurable business value yet?
Research points to several factors: a shortage of manager skills to use it effectively, widespread unsanctioned personal use running ahead of official tools, and a lack of organizational guidance on how to actually apply the time it saves.
Performance
