Next in Digital Health
AI can boost productivity, but workforce readiness depends on preserving judgment, accountability, and expertise.
As someone who writes a lot, there’s nothing more daunting than a blank page. A productive writing session requires time and focus to turn an idea into a finished product.
Generative AI can now condense that process into minutes (if I let it). That can be useful in certain contexts, but it also raises a larger question: What do we lose when AI fast tracks so much of our work?
AI entered the workplace with an irresistible prospect to help people move faster and spend less time on low-value tasks. For organizations facing workforce shortages and growing operational complexity, that promise is hard to ignore. Too many people spend too much of their day searching for and summarizing information, reformatting work, drafting routine messages, or moving data between systems. Across industries, workforce capacity is too constrained to deal with these unnecessary burdens.
This problem is amplified in healthcare.
AI should help reduce those burdens. The challenge for leaders is ensuring they automate the right type of work without automating the cognition surrounding the work.
As AI continues to move from being an occasional assistant to underlying every layer of the business—when a tool can synthesize the first analysis, suggest the first plan, create the first draft, or offer the first recommendation—it changes how people think (and in some cases, it creates a new dependency).
The workplace risk is easy to overlook because we’re still enamored with AI’s performance and how it has improved dramatically in a short period of time. Its first attempt is often more organized than a person's first attempt. These tools can also raise baseline performance across an organization by helping staff produce work beyond their traditional areas of expertise.
This means that “good-enough” work arrives much earlier in the development process, sometimes before a human has wrestled with the task deeply enough to have an informed opinion.
We’re starting to see early signs of unintended consequences. For example, some analyses suggest that our willingness to offload all the easy, mundane work may be increasing the intensity of work left over for humans. There’s also research showing how the output of an AI-enabled organization could start to converge. If everyone relies on the same core set of tools, prompts, and optimization logic, individual output may improve while collective output may start to sound more alike. Some organizations may view this in a positive light as a form of alignment, but others will see this as a pathway to constrained ideas. In other words, AI may “raise the floor” of workforce performance while simultaneously lowering the ceiling of uniqueness.
Part of offloading work to AI is distinguishing the types of work we still want humans to handle:
One challenge is that developmental effort often looks inefficient. A team debating competing strategies may appear slower than one accepting an AI-generated recommendation. An employee working through an early iteration may seem less productive than someone starting with a machine-assisted first pass. But these moments are often where expertise is built. If organizations automate them too aggressively, they may gain short-term productivity while weakening the human capabilities they hope to retain.
The future of AI-enabled work should remain developmental to optimally leverage AI’s capabilities while ensuring staff are retaining and building the skill sets they need to remain capable, distinctive, and accountable.
As AI begins to influence nearly every workflow, business leaders must define where human effort is valuable.
Five kinds of work are worth preserving:
1. The first thought
Initial brainstorming reveals assumptions, creates a baseline for learning, and forces staff to engage in a task before a machine constructs the answer. If AI always starts the decision-making process, people become accustomed to reacting to ideas without strengthening their own ability to generate those ideas.
Before employees turn to AI for an answer, ask them to articulate their initial hypothesis, recommendation, concerns, or decision criteria. The goal is to ensure people actively engage with the problem before technology begins shaping the solution.
2. The handoff
While it’s tempting to have AI take on a greater share of work, it also creates the risk that people become less vigilant over time in defining the boundaries of where AI operates and what should remain in human control.
This handoff decision becomes especially important as AI moves into high-volume workflows and organizations deploy more autonomous agentic tools. It’s during these handoffs where a human should determine whether a given task requires AI (believe it or not, AI is not the answer to every business challenge), the scope of impact AI should have on a task, and what parts of the work require escalation.
Leaders can support this type of work by ensuring staff remain skilled at deciding when delegation is appropriate and when it becomes a type of cognitive surrender.
3. The revision
As more work becomes AI-assisted, revision becomes one of the clearest signs of humans continuing to exercise judgment. Accepting AI-generated output with minimal change can be efficient, but it also weakens ownership as staff produce a higher volume of output while having only a shallow relationship to the thinking that went into it.
Meaningful revision allows staff to compare AI output against established goals, standards, or risk thresholds. This helps define what’s missing, what’s overstated, what sounds right but says very little, and if the output aligns with the original intent.
Leaders should assume AI will be involved in more work over time, and then make AI revision a visible and encouraged process. This can be done by fostering discussion around what changed from the AI-generated version, what the tool missed, or where human expertise altered the answer. This transparency builds a culture where AI use is expected, but so is active participation to help avoid always passively accepting AI-generated output.
4. The dissent
In some cases, AI’s ability to produce a balanced or reasonable sounding output on command can make consensus cheaper by synthesizing dominant themes. Returning to the “comfort of convergence,” an AI’s polished output can be useful, but it also restricts the ability to create differentiated strategies that push us out of our comfort zones.
Organizations need people who can challenge AI-driven consensus. They need staff who notice when a plan or recommendation is too generic, or when a faulty assumption is hiding behind the data. This is where the “constructive misfit” becomes essential— someone who readily uses AI tools but is inclined to challenge its output as a starting point. This person will want to know the data provenance, the local context, the model’s decision logic, or confidence in its output. In an AI-enabled organization, constructive misfits become a form of risk management by pressure testing AI and refusing to offload all effort to the technology. When business incentives are geared towards rewarding speed and productivity, dissent may feel like obstruction, but leaders should protect thoughtful pushback and principled disagreement.
5. The accountability moment
AI can help with decision making, but people and organizations still own the consequences. Someone must be able to explain why a recommendation was accepted, what alternatives were considered, what risks remain, and why the final decision deserves trust.
This matters because AI can make work feel more complete or valid than it is. A clean summary or confident recommendation can obscure unresolved ambiguity or make a weak premise appear mature.
Leaders can reinforce accountability by asking simple questions:
If staff struggle to answer these questions, that should be viewed as a red flag.
Developmental effort and the types of work outlined above become guideposts for AI governance and workforce strategy. Leaders should evolve AI conversations beyond just productivity and tool access metrics and instead focus on preserving valuable staff competencies and supporting the effort that builds new capabilities.
This means that for any AI-enabled workflow, there must be better clarity around key drivers of workforce enablement, such as whether staff are challenging assumptions and pushing new ideas if always resorting to using AI, whether early careerists are missing out on important skill building, if staff are maintaining appropriate autonomy in decision making, and more. Protecting this developmental effort is where organizations will start to carve out a distinct advantage in a marketplace where using AI is no longer a competitive differentiator.
Benchmark your AI readiness in six key domains—including culture and talent—and map a practical path to enterprise-wide impact with our free AI Maturity Assessment.
Before adopting AI into a workflow, ask:
Vizient Senior Director, Intelligence
As a senior director on the Intelligence team, Andrew Rebhan leads thought leadership and content creation for digital health research at Vizient. In this role, he keeps members up to date on the latest technology trends and how to plan for new, disruptive forces and innovation entering the health care industry. Areas of interest include artificial intelligence, consumer medical technology,...
