AI is spreading rapidly across workplaces, but access alone does not create productivity. The real divide is emerging between people who experiment deeply, people who use AI occasionally, and the much larger group that barely changes how they work.

You may already feel like you are surrounded by AI. New models appear constantly, companies are buying enterprise licenses, employees are being encouraged to experiment, and AI agents are becoming part of everyday conversations.

But there is a much more important question underneath all of that: Is AI actually changing how work gets done?

The article makes a compelling case that the answer is often not as straightforward as adoption numbers suggest. A company can have AI available to thousands of employees, report high usage, and still see very little improvement in how quickly work gets completed.

The reason is simple: using an AI tool and knowing how to use it well are different skills.

Running more campaigns won't move the needle. Building the right strategy will.

HubSpot Academy's Digital Marketing Certification gives you the foundation to make that happen. In just over 3 hours, you'll cover consumer behavior, ethical marketing, SEO, content, email, social media, paid ads, and AI — the full stack, in one course.

Walk away with:

Free. No credit card. Shareable on LinkedIn the day you finish.

Join 200,000+ professionals who have advanced their career with HubSpot Academy — get certified today.

Get Certified Free

There is an enormous difference between the most advanced AI users and the average employee.

At the frontier, some people are running multiple terminals, building agents, maintaining knowledge bases that update themselves, creating reusable skills, and experimenting constantly with new AI capabilities. They may use AI every day and already have strong opinions about how different models and workflows should be used.

That can create a misleading perception of how widespread advanced AI adoption really is.

The article argues that someone who regularly works with models, experiments with agents, and has developed a sophisticated AI workflow may already be among the most advanced users—even if they feel far behind the people constantly showcasing new experiments online.

The much larger gap exists behind them.

The article describes the median employee at a large enterprise as someone who may have opened an AI tool only a few times over several years. Someone might have tried an older model, decided AI was not particularly useful, and never returned to it.

That employee is operating in a completely different world from the AI power user.

Tip: Do not judge your own AI progress by comparing yourself with the most extreme users online. The important question is whether AI is becoming genuinely useful in your own work.

One of the strongest examples in the article involves a large, non-technical enterprise operations organization.

The organization introduced Claude Cowork, expecting the technology to improve productivity. Yet the leader found that the team continued moving at roughly the same rate.

The problem was not that the technology was incapable. The problem was how differently people used it.

The organization developed what the article describes as a barbell distribution. Around 5% to 10% became power users who worked with Claude Cowork regularly, used skill files, connected tools such as Outlook, and actively experimented with what the technology could accomplish.

Another roughly 20% used it a couple of times a day but extracted only limited value.

The remaining 70% barely used it at all.

So the company could point to a successful rollout, licenses could be active, and usage dashboards could show adoption, yet the underlying organization was not necessarily moving faster.

That is the central problem with treating adoption as a simple yes-or-no measurement.

Tip: When evaluating a new AI rollout, look beyond whether employees opened the tool. Ask what work actually changed because of it.

The article makes another important distinction: AI skill is a spectrum.

Consider two engineers given the same software task.

The first engineer can paste a Jira ticket into an AI tool and immediately ask it to make the changes. The model modifies several files, the tests pass, and the engineer quickly reviews the result before merging it.

The second engineer approaches the same task differently. They identify where the relevant code lives, tell the model which areas should and should not be touched, use existing skill files to encourage minimal changes, and make sure the work is properly tested.

Most importantly, they actually read the resulting diff.

That final step matters.

An AI-generated change can pass tests and still introduce unnecessary or incorrect modifications. In the article's example, the first engineer's change included a configuration change that had nothing to do with the original task. It was only discovered later when the production system began behaving differently.

The second engineer caught the unnecessary change before merging.

Same tool. Same task. Very different outcome.

The difference is not simply access to AI. It is knowing how to direct, evaluate, constrain, and verify what AI produces.

Tip: Treat AI output as work that requires judgment, not as work that automatically deserves approval because the system produced it.

The article also describes another enterprise where AI licenses had been deployed at enormous scale, involving an eight-figure annual commitment.

The surprising part was the distribution of usage: roughly 10% of employees consumed about 90% of the tokens.

That creates an unusual economic problem.

If the remaining 90% suddenly began using AI at the same intensity as the top 10%, AI-related spending could potentially increase dramatically. The article illustrates this with a hypothetical move from a $10 million commitment toward roughly $100 million in usage.

In other words, the organization can simultaneously have a highly successful group of AI users and a major adoption problem.

The power users demonstrate what the technology can accomplish, but their behavior cannot simply be assumed to represent everyone else.

This is why the article argues that a perfect rollout would still produce a barbell. Even if the technology is introduced properly, people will differ substantially in their willingness and ability to use it.

The article points to McKinsey's 2025 survey, which found that 88% of organizations reported using AI in at least one business function.

That sounds enormous.

But only 6% reported that AI was responsible for more than 5% of their EBIT.

Those numbers can exist together because "using AI" is an extremely broad definition.

An employee who asks AI to rewrite an email and another employee who operates multiple AI-powered workflows are both counted as users, even though their economic impact can be dramatically different.

The article also references MIT NANDA's GenAI Divide report, which found that 5% of integrated AI pilots were extracting millions of dollars in value, while the other 95% showed no measurable P&L impact.

The lesson is not that AI is ineffective. It is that AI experimentation, AI usage, and measurable business value are three different things.

Tip: Be careful with impressive adoption percentages. A high usage rate does not automatically mean a high productivity or financial impact.

This may be the article's most important distinction.

Adoption can be measured with questions such as:

Did someone log in?

Did they use the tool this month?

Did they send five prompts a day?

Those measurements are easy to collect, but they flatten a huge range of behavior into simple numbers.

They do not distinguish between someone who has never opened the tool, someone who occasionally pastes an email into it, and someone who has built several agents that perform meaningful operational work.

The article argues that what organizations should really understand is how skillful each person is at using AI and what return they are getting from the resources they consume.

That is much harder to measure than a login.

It is also much more useful.

Creators lock in holiday content calendars 90 days out, before brands figure out commissions. Waiting too long to launch an affiliate program means less runway to build demand and a missed shot at the best partnerships.

The 90-Day Holiday Sprint covers commissions, recruiting, and scaling a program at Day 30, 60, and 90.

Download the Free Guide

There is another uncomfortable dynamic in the article: every improvement in AI capability can potentially make the skill gap wider.

AI companies continue building increasingly capable agents and tools, but greater capability can also raise the skill required to take full advantage of those systems.

The most experienced users are often the people who experiment with new features first. They discover better workflows, create reusable skills, and find ways to connect AI with other tools.

That gives them even more leverage.

Meanwhile, people who have barely started using AI may find the increasingly sophisticated ecosystem even harder to navigate.

The article also makes an interesting point about internal AI experts. The people who have developed an advantage from their AI knowledge may not always have a strong incentive to eliminate the gap. If their expertise allows them to complete work much faster or handle the workload of several people, that advantage has value.

Closing the gap can therefore require deliberate organizational incentives rather than simply asking experienced users to teach everyone else.

Tip: Do not assume your best AI users will naturally spread their knowledge throughout the organization. Create a system that makes sharing useful skills worthwhile.

One of the article's criticisms is the heavy emphasis on teaching employees how to prompt.

Prompting matters, but the article argues that it represents only a small portion of what organizations actually need to understand.

The bigger question is which workflows should involve AI at all.

Some processes should never depend on a model when deterministic software can perform the task more reliably. Other processes may benefit from AI because they require interpretation, judgment, or handling of less structured information.

Understanding that difference requires looking closely at how work is actually performed inside an organization.

That is much harder to package into a generic training course because the right answer will differ from company to company.

Tip: Before improving prompts, improve the workflow. Determine what should be automated, what should use AI for judgment, and what should remain deterministic.

The article does not argue that training is unnecessary.

Quite the opposite.

Training can reveal who is genuinely interested in AI, who has the ability to become highly effective with it, and who may need a different approach.

For the strongest users, the article recommends creating a shared place where they can publish the skills and workflows they develop. Others can then discover, rank, reuse, and install those capabilities.

This creates a way for individual breakthroughs to become organizational knowledge.

But the article also recognizes that this will not turn everyone into an AI power user.

For the majority, a different strategy may be more effective.

This is where the article's solution becomes particularly practical.

Instead of expecting every employee to become proficient at prompting, agent creation, and workflow design, organizations can place AI directly inside the systems where employees already perform their work.

The article uses accounts-payable analysts as an example.

If analysts spend their days processing invoices, much of that repetitive activity may be suitable for automation. Rather than requiring every analyst to build an agent or repeatedly prompt an AI system, an organization can build automated processes into existing systems such as Salesforce, NetSuite, or Dynamics, depending on the workflow.

The employee's role can then become reviewing the AI's work, approving or rejecting it, and intervening when something requires human judgment.

That is a fundamentally different approach.

The employee does not have to become an AI expert just to benefit from AI.

Tip: If a process happens repeatedly, look for ways to make AI part of the process itself rather than making employees responsible for operating the AI manually every time.

There is a simple idea running underneath the entire article: people want the work done.

They are not necessarily looking for another application to learn, another dashboard to monitor, or another system requiring constant prompting.

That is why background automation can be so powerful.

When AI is embedded into the systems people already use, the technology can handle repetitive work while employees remain responsible for the decisions that require context, judgment, or accountability.

For someone already dealing with a crowded schedule, that distinction matters enormously.

The best AI experience may not feel like "using AI" at all.

It may simply feel like the work has become easier.

The article ultimately challenges organizations to rethink what they report as AI progress.

Instead of presenting adoption percentages to leadership, a more meaningful measurement would be the proportion of work that remains manual, hybrid, or fully automated.

That tells a much clearer story.

If 90% of employees have access to an AI assistant but almost all of their work remains manual, the organization has not transformed its workflow simply because adoption is high.

If a smaller number of AI systems quietly automate repetitive processes while employees focus on reviewing, deciding, and solving more complicated problems, the impact can be much more meaningful—even if fewer people are actively opening an AI chatbot.

The future of workplace AI, then, may not belong to the organization with the highest adoption percentage.

It may belong to the organization that understands where the work happens, where AI genuinely helps, where humans need to remain involved, and how to remove unnecessary effort without forcing everyone to become an AI specialist.

And for you, that may be the most useful takeaway of all: you do not need to chase every new AI capability. You need to find the places where it can quietly make the work in front of you better.

What’s your next spark? A new platform engineering skill? A bold pitch? A team ready to rise? Share your ideas or challenges at Tiny Big Spark. Let’s build your pyramid—together.

That’s it!

Keep innovating and stay inspired!

If you think your colleagues and friends would find this content valuable, we’d love it if you shared our newsletter with them!

As the world becomes increasingly digital, this question will be on the minds of millions of people seeking new income streams in 2026.

The answer is—Absolutely!

Thank you for taking the time to read today’s email! Your support allows me to send out this newsletter for free every day. 

 What do you think for today’s episode? Please provide your feedback in the poll below.

Share the newsletter with your friends and colleagues if you find it valuable.

Disclaimer: The "Tiny Big Spark" newsletter is for informational and educational purposes only, not a substitute for professional advice, including financial, legal, medical, or technical. We strive for accuracy but make no guarantees about the completeness or reliability of the information provided. Any reliance on this information is at your own risk. The views expressed are those of the authors and do not reflect any organization's official position. This newsletter may link to external sites we don't control; we do not endorse their content. We are not liable for any losses or damages from using this information.