Your AI Is Hiring Too Many Bad Employees

I spent the early part of this week in Tennessee speaking to a room full of business leaders about AI, governance, leadership, and the NOW of work. Like so many conversations I have been having lately, it was energetic, thoughtful, and filled with optimism about what AI can make possible as well as fear about the job loss it will cause.

From there, I boarded a flight to Seattle to celebrate the wedding of a great friend. Somewhere over the Rockies, with terrible wifi and a few uninterrupted hours to think, I stopped thinking about AI and started thinking about leadership.

One idea from Tennessee refused to leave me alone:

We are spending an enormous amount of time talking about how AI works, but not nearly enough time talking about how work should work.

The more I reflected, the more I realized this was not really an AI insight. It was a leadership insight—one that has followed me through more than thirty years of helping organizations navigate change. By the time I landed, I was no longer thinking about the flight. I was thinking about work, and I knew I had to write.

During the last three decades, I have had the privilege of working with leaders through almost every major wave of organizational transformation. The technologies changed dramatically, from enterprise systems and the internet to cloud computing, mobile technology, digital transformation, and now artificial intelligence. Yet the conversations that mattered most were rarely about the technology itself.

They were about work—how it was designed, why it existed, whom it served, and whether it still created value.

Sometimes transformation meant eliminating an approval that had outlived its purpose. Sometimes it meant redesigning a process that had become more complicated than the problem it was supposed to solve. Other times, it meant questioning why twenty people were invited to a meeting that only required five.

The objective was never simply to make people work faster. It was to create more space for people to do meaningful work, make better decisions, and focus their energy where it could create the greatest impact.

That is why one question has remained so important throughout my career:

What work no longer creates enough value to deserve its place?

It is also why I find the current conversation about AI so fascinating. Everywhere I go, leaders want to discuss models, agents, reasoning capabilities, context windows, and which platform will eventually win. Those are worthwhile conversations, but I increasingly wonder whether our fascination with intelligence is distracting us from the more important question.

What work are we asking AI to do that never needed to exist in the first place?

Every leader has experienced a meeting that felt much larger than the decision it ultimately produced. Everyone in the room was intelligent, capable, and well-intentioned. Yet, as the conversation unfolded, it became clear that only a handful of people were materially shaping the outcome.

The meeting was not ineffective because people lacked talent. It was ineffective because the organization consumed far more time, attention, and energy than the decision required.

That kind of friction rarely appears overnight. One person is invited because they may have useful context. Another joins because they participated in a similar decision six months ago. Someone else is included because nobody wants them to feel left out. Before long, twenty people are in a meeting where five contribute and two make the decision.

The same pattern appears throughout our organizations:

• Reports continue to be produced long after anyone has stopped reading them.

• Approval steps remain in place even after the risks they were designed to manage have changed.

• Email chains expand until everyone is informed but nobody is accountable.

• Processes become more complicated because adding another step feels safer than removing one.

• Work survives not because it creates value, but because it has become familiar.

Organizations do not usually become inefficient because leaders deliberately create waste. They become inefficient because unnecessary work slowly becomes embedded in the way the organization operates.

As I began looking at AI through that same lens, I realized that we may be recreating digitally what we have spent decades trying to eliminate organizationally.

This brings me to a term that rarely appears in leadership conversations but increasingly should: tokens.

If you have used ChatGPT or another Gen AI tool, you have already used tokens—even if you have never heard the term. Tokens are the small pieces of information an AI system reads and produces. A token might be a word, part of a word, or even a punctuation mark. Your question consumes tokens, the context you provide consumes tokens, and the answer consumes tokens.

Think of tokens as the individual words in a conversation. Every word requires a small amount of processing. The right words move the conversation forward, while thousands of unnecessary words add cost, consume time, and make the point harder to find.

Consider a simple everyday example. You ask an advanced AI model, “What will the weather be tomorrow?” The model processes your instructions, interprets the request, and generates a polished response—all to retrieve information that a weather app could provide instantly.

It is not that the AI cannot do the job. It is that the job does not require that much intelligence.

The organizational equivalent would be asking your chief strategy officer to look out the window and tell you whether you need an umbrella. You will probably get an answer, but you have badly matched the capability to the task.

Wasted tokens are not always the result of bad technology. They are often the result of bad delegation.

This is why I have started thinking about tokens not simply as units of computation, but as units of organizational effort. Every instruction we give an AI system, every paragraph of background, every document uploaded “just in case,” every unnecessary example, and every request for deeper reasoning requires additional processing.

Some of that effort improves the result. Some of it does not.

That is when the analogy became impossible for me to ignore:

Every unnecessary token is another person in the meeting.

Intelligence without intention is just expensive activity.

Tokens have real consequences. They affect cost, processing time, and the amount of infrastructure required to operate AI at scale. The fact that AI providers offer prompt caching—discounting repeat input because processing the same context again has a cost—makes the economics visible.

More importantly, early research into agentic AI is beginning to show that higher token consumption does not reliably produce better results. A 2026 study of agentic coding tasks found enormous variation in token use across runs and reported that greater consumption did not translate consistently into greater accuracy. Another efficiency study found that models with comparable accuracy could differ sharply in the number of tokens they generated.

This does not mean that fewer tokens are always better. Some problems deserve deep reasoning, broad context, and meaningful computational effort. A critical person should remain in the meeting, essential context should remain in the prompt, and complex decisions should receive the intelligence they require.

But everything should have to earn its place.

The real leadership question is not, “How do we use fewer tokens?” It is, “How do we create the greatest value from the effort we are asking people and technology to contribute?”

The goal is not less intelligence. The goal is less wasted intelligence.

One of the greatest risks in the AI era is not that organizations will fail to automate enough work. It is that they will automate work that should have been eliminated.

For decades, leaders have been taught to ask how a process can become faster, cheaper, or more efficient. AI makes that temptation even stronger because it can perform many existing tasks more quickly than people can. But improving the speed of unnecessary work does not make the work more valuable.

A monthly report that nobody reads does not become useful because AI creates it in ten seconds. A confusing approval process does not become intelligent because an agent moves information through it automatically. A meeting that should not exist does not become productive because AI produces a beautiful summary afterward.

Automating waste does not create transformation. It creates faster waste.

This is where leaders need to pause and ask better questions:

• Are we using AI to redesign work, or merely to accelerate the work we already have?

• Are we eliminating friction, or digitizing it?

• Are we improving outcomes, or generating more activity?

• If we were designing the organization today, with AI available from the beginning, would we build the work this way?

• What should we stop doing before deciding what to automate?

These are not primarily technology questions. They are leadership and organizational-design questions.

AI can make an effective organization dramatically more capable. It can also make an inefficient organization dramatically faster at producing waste.

AI does not fix the way your organization works. It reveals it.

This challenge is also changing how we prepare our workforce for AI.

Most organizations begin with training: which tool to use, where to click, how to write a prompt, and what rules to follow. Training matters, but it teaches people how to operate today’s technology. It does not necessarily help them understand the work well enough to question it.

Education goes deeper. It helps people understand what AI is doing, where its effort and cost come from, when additional context improves an outcome, and when more activity simply creates more waste. Most importantly, education gives people the confidence to ask whether a task should exist before learning how to automate it. We call this “changefulness”.

Training teaches people how to use the tool. Education teaches them when—and why—to use it.

If we want our workforce to apply AI responsibly, we need people who can ask:

1. Does this work create enough value to continue?

2. Does this task require reasoning, information, or action?

3. Is the additional context improving the result or adding noise?

4. Should we automate this work, redesign it, or eliminate it?

5. Are we using AI because it is appropriate—or simply because it is available?

This distinction matters whether employees are building AI systems or simply using tools such as ChatGPT in their everyday work. A trained employee may know how to produce a longer, more sophisticated prompt. An educated employee understands that the best prompt is the one that produces the right outcome with the least unnecessary effort.

A workforce that knows how to use AI will increase adoption. A workforce that understands AI will increase value.

The goal is not to turn every employee into an AI engineer. It is to give every employee enough understanding to exercise judgment. Technology will continue to change, and today’s instructions will eventually become outdated but judgment travels.

That is why organizations should not treat AI readiness as a one-time training program. It should be an ongoing educational effort that helps people understand the relationship between technology, work, and value.

Tools change quickly. Good judgment compounds.

Before implementing a new AI workflow or even writing the next complex prompt, I believe leaders should apply what I call **The Work Value Test**.

1. Should this work exist?

Before asking how AI can perform the task, ask why the task is being performed at all. What decision does it support? What outcome does it improve? Who uses the result?

If nobody can clearly explain the value, automation may not be the answer. Elimination may be.

2. Does the work require intelligence, information, or action?

Some tasks require judgment and reasoning. Others require the retrieval of a fact, the completion of a routine action, or the application of a defined rule.

Using an advanced reasoning model for every task is the organizational equivalent of assigning executive-level talent to entry-level work. The capability may be impressive, but the allocation is poor.

The smartest tool is not always the right tool.

3. Does every piece of context improve the outcome?

More context can help, but it can also create confusion, increase cost, and make it harder for the system to distinguish what matters. Just because an AI system can process hundreds of pages does not mean it should, just as the ability to invite twenty people does not mean all twenty belong in the meeting.

The question is not whether more information is available. The question is whether more information improves the decision.

4. Are we measuring activity or value?

Organizations can easily count prompts, users, agents, documents processed, and tokens consumed. Those measures may demonstrate adoption, but they do not necessarily demonstrate value.

The measures that matter more may include:

- Work eliminated

- Time returned to people

- Decisions improved

- Friction removed

- Capacity created

- Customer or employee outcomes changed

Adoption tells you that people used the technology. Value tells you that something became better.

5. If every token were a person in the meeting, who could leave?

This question turns a technical discussion into a leadership discussion. It forces us to examine which instructions, documents, steps, and layers materially improve the outcome—and which remain simply because nobody has challenged them.

Again, the goal is not to be minimal for its own sake. The goal is thoughtful contribution. Everything should have to earn its place.

Great leadership is not about adding more. It is about knowing what deserves to remain.

As I thought back on the conversations from Tennessee, I realized that the leaders in that room were not really looking for another explanation of AI. They were looking for clarity.

Technology is evolving faster than any of us can comfortably absorb. Leadership has become an exercise in translation. Our job is not simply to understand what is changing; it is to help people understand what those changes mean for the work they do, the decisions they make, and the organizations they are trying to build.

That is one of the reasons we created Now to Next. Leaders do not need more noise. They need help translating complexity into clarity so that they can make better decisions, help their people adapt, and redesign work for a different future.

This is why the token conversation is not really about tokens. It is about work, value, and the discipline required to distinguish activity from progress.

The organizations that thrive in the next decade will not necessarily be the ones that deploy the most AI, consume the most tokens, or build the largest collection of agents. They will be the organizations that apply intelligence deliberately. They will question work before automating it, simplify before scaling it, and measure success by the value created rather than the activity generated.

The technology may be new, but the leadership lesson is not. Productivity has never been about asking people to do more work. It has been about creating the conditions for people to do work that matters.

AI gives us an extraordinary opportunity to extend that principle, but only if we are willing to ask difficult questions before reaching for easy automation.

Not every token belongs in the prompt, just as not every person belongs in the meeting. The goal is not simply to use less. The goal is to make every contribution count.

Before your next meeting or your next AI prompt—ask yourself:

What work am I asking people or technology to perform that no longer creates enough value to deserve its place?

Moving into the future is not about taking everything from the present with us. It is about having the judgment to decide what still belongs, the courage to leave unnecessary work behind, and the clarity to help people understand why.

That is how we move from Now to Next.

About Jason

Jason Averbook is the co-founder of Now to Next, an adjunct professor of business, and a globally recognized thought leader, advisor, and keynote speaker working at the intersection of AI, human potential, and the future of work. He spent the last few years as Senior Partner and Global Leader of Digital HR Strategy at Mercer, helping the world’s largest organizations reimagine how work gets done, not by implementing technology but by transforming the mindsets, skillsets, and cultures that have to come first.

Over the last two decades, Jason has advised hundreds of Fortune 1000 companies and their leaders, founded Knowledge Infusion and Leapgen, authored two books on the evolution of HR and workforce technology, and become a world renowned keynote speaker who has delivered hundreds of talks on the future of work. His work challenges leaders to stop treating digital transformation as an IT project and start treating it as a human strategy.

Through his Substack, Now to Next, Jason shares honest, provocative, and practical insights on what’s actually changing in the workplace, from generative AI to skills-based organizations to emotional fluency in leadership. His mission is simple: to help people and organizations move from noise to clarity, from fear to possibility, and from now to next.

You can reach him at jason@nowtonext.ai or connect on LinkedIn.