AI Is Moving From Answers to Action. Are We Ready?
The most important change in AI is not simply that more people are using it. It is that AI is being asked to do a different kind of work.
The question hidden inside “Who uses AI?”
Ask whether someone uses AI, and the answer is increasingly likely to be yes. One person may use ChatGPT to summarize a report or draft an email. Another may give an AI agent an objective, connect it to organizational systems, and expect it to complete several steps before returning with evidence, exceptions, or a recommendation.
Both use AI. But they are not developing the same capabilities, receiving the same leverage, or preparing for the same future of work.
That distinction is crucial because adoption is moving faster than measurement. A single number can make AI appear more mature, evenly distributed, and transformational than it is.
The more consequential questions are:
Who has access to AI, and how deeply are they using it?
What is AI helping them do—and who is learning to direct the work?
Generative AI Made AI Visible
Generative AI responds to a request by producing or transforming an output. It can draft, summarize, translate, analyze, create images, generate code, or answer questions.
It has lowered the friction of finding information, exploring an idea, and moving from a blank page to something usable. But the work remains organized around an answer or artifact:
The human asks. The system responds. The human evaluates and decides.
That model is beginning to change.
Agentic AI Wears Workboots
Agentic AI works toward an objective across multiple steps. An agent may retrieve information, use software, make intermediate decisions, monitor progress, recover from errors, and escalate when judgment is required.
The difference is not that generative AI is passive or that agentic AI is magically autonomous. It is the structure of the work:
Generative AI primarily helps a person produce an output.
Agentic AI can participate in a workflow.
That makes the shift more substantial than a product upgrade. Agentic AI is shaping what humans believe is possible for the way we design and decide what our work will be.
AI Adoption Is Broad. Agentic Work Is Not.
Enterprise research makes the distinction visible. In its 2025 State of AI report, McKinsey reported that 88% of respondents said their organizations regularly used AI in at least one business function. Yet only 23% said they were scaling an agentic AI system, while 39% were experimenting with agents.
These figures are not contradictory. They measure different levels of adoption—and different barriers to success.
Healthcare acceleration is worth examining
McKinsey identifies healthcare as one of the sectors where agent use is most widely reported. My working hypothesis is that workforce shortages create pressure to automate administrative, retrieval, routing, and documentation work, while care outcomes still depend on human judgment, trust, communication, and relationships.
The opportunity is leveraging automation to return human capacity to higher-leverage work.
The Exposure Divide Is Inside the Organization
Public availability is not the same as institutional preparation. In the Jobs for the Future study on AI and Black learners and workers, 31% of the overall sample reported that their employers offered AI training, while 16% reported access to paid AI tools provided by an employer or school.
The signal is clear: organizations may invest in AI skills while concentrating advanced access, training, and design experience among selected functions or leadership groups. Exposure to the different categories of AI—and to what each requires for success—is not universal.
That unevenness can fuel the perception that AI is hype. Leaders may use one broad word to describe systems with very different capabilities, requirements, risks, and rates of speed. They may expect the organization to move smarter and faster without seeing the investments each category requires.
A company can have hundreds of employees using generative tools while only a small group can connect agents to enterprise data, business systems, or customer-facing workflows.
Before evaluating progress, organizations need to ask where AI is being used, at what depth, by whom, and with what permissions.
“AI adoption” can describe anything from asking a chatbot for an answer to redesigning a business process around human-agent collaboration.
AI Is Moving Into the Flow of Value
As the technology moves deeper into workflows, it is increasingly connected to the flow of value: customer discovery, sales, service delivery, product development, research, supply chains, internal operations, healthcare administration, and commerce.
The progression is moving from:
Generate content.
Define tasks and use organizational data.
Recommend or execute workflows.
Coordinate work toward a strategic goal.
Deloitte’s 2026 State of AI in the Enterprise report reports that only 34% of leaders surveyed said they were truly reimagining the business. Organizations may be educating employees about AI without redesigning roles, workflows, and career paths around it.
The technology may be ready to participate in a workflow before the organization has created the standards, data, training, and context required for it to succeed.
The Leadership Job Changes When AI Enters the Workflow
Generative AI asks leaders where AI can help people work faster, how employees should use it responsibly, and how output quality will be evaluated.
Agentic AI adds another layer:
What systems must share data?
What goes to an agent, and what remains with humans?
What decisions can the system make without human intervention?
If AI moves information and work faster, leaders who hold every decision too long can become the bottleneck. Decision velocity is not carelessness. It is the ability to define what can be delegated, what requires review, and when the human-agent team must move.
That capability develops through exposure and deliberate practice. The leadership shift is from adopting a tool to designing a system of work—seeing the full workflow, defining standards, managing exceptions, and keeping humans in the loop of decision-making.
For leaders, progress starts with four moves:
Measure generative use separately from agentic deployment.
Identify workflows where automation can return capacity to higher-leverage human work.
Audit who has access to advanced tools, training, and design experience.
Define delegation, review, escalation, and accountability before agents enter critical workflows.
You do not need to master every tool. You do need a working mental model of the work, the decisions, and the people being prepared to lead it.
To Lead AI, Start With Where It Came From
In our next release, we’ll trace the technological evolution behind today’s AI. To lead the changes it drives—or advocate for resources credibly—you need to understand where it came from.
Healthy skepticism helps us test claims, identify limits, and decide where safeguards belong. But resistance becomes unproductive when we spend our energy debating whether AI is good or bad, as if organizations can simply put it back in the box.
They won’t.
The better we understand the evolution of the systems that made today’s AI possible, the better we can build mental models for what is safe, where additional guardrails are needed, and what we can create with AI—individually and together inside our organizations.
