Artificial intelligence is not a sudden invention of the 2020s. Its roots stretch back to at least 1956, when a small group of computer scientists gathered at Dartmouth College to ask whether machines could think. What has changed is not the existence of AI but its reach. Powerful, general-purpose systems have pushed the technology from research labs into everyday business conversations, and the result is a mix of excitement and anxiety that often obscures the practical questions leaders now face.
According to Paula Goldman, author of the forthcoming book Manage the Machine: How to Harness Human-AI Collaboration at Work, three shifts distinguish today's AI from earlier waves. The first is that AI has moved from specialist to generalist. For decades, AI operated behind the scenes in models trained for narrow tasks, such as predicting customer churn or flagging fraud. Today's foundation models are vast neural networks trained on oceans of data and adaptable across contexts. The same model that helps a developer write code can help a marketer draft copy or an HR leader compose a job description. That versatility pulled AI out of the back office and into nearly every corner of knowledge work.
The second shift is that AI now speaks human language. Users no longer need to code or navigate rigid menus; they can type or speak in everyday language and receive polished, humanlike prose in return. These systems also generate new content, including text, images, and code, which is why the current wave is often called generative AI. But that ease of use creates a new responsibility: workers must learn when to trust AI's output and when to challenge or reshape it.
The third and perhaps most profound change is agency. AI no longer just analyzes or predicts; it acts. So-called agentic AI can plan steps toward a goal, call for the right tools, check its own work, and continue without a human clicking send. A single agent can draft an email, open a ticket, schedule a delivery, and log the transaction. That autonomy promises significant productivity gains, but it also raises unresolved questions. How much independence should AI have? What does accountability look like when decisions are distributed across humans and machines?
These shifts blur the line between using AI as a tool and treating it as part of the team. Yet Goldman cautions that generative and agentic AI remain uneven in their abilities. An AI agent can draft a sophisticated legal memo but may confidently misread a contract's indemnity clause. It can write working code yet stumble on a small, unspoken requirement that a junior engineer would catch. Harvard researcher Fabrizio Dell'Acqua and colleagues describe this as the «jagged technological frontier» of AI: systems that amaze on some tasks and stupefy on others.
The practical task for organizations is to map that frontier. That means knowing when to rely on AI, when to supervise it, and when to step in with uniquely human judgment. Goldman argues that decades of lessons from earlier AI generations and from complex technologies such as aviation, automobiles, and power plants can be adapted to this moment. The challenge is not technical mastery alone but leadership and collaboration. How people choose to work with AI, and who they become in the process, remains a decision within human hands.