AI Is a Tool. Are We Learning How to Use It?

Powerful AI tools have become remarkably easy to access. Open ChatGPT, Claude, Gemini, or any number of specialized products, and capabilities that would have seemed impossible only a few years ago are sitting right in front of you. Companies are giving employees these tools and asking them to innovate. Product teams are building AI into their experiences. Schools are figuring out what learning means when a student can generate an essay in seconds.

But access to a tool and knowing how to use it are two very different things.

That distinction sits at the center of our conversation with Andrew Lippert, founder of Operating Thesis and former CTO at Edmentum and McGraw Hill, on the Product Builders Podcast. Andrew has spent more than two decades in education technology, watching the industry evolve from paper products to CD-ROMs, the web, and now generative AI. His perspective on this latest shift is less about what AI can do and more about whether we are developing the judgment to use it well.

We know these tools are powerful, and it's increasingly clear they'll change how we work. The more interesting questions now are where they belong, what problems they should solve, and what responsibilities we should keep.

Unlimited Freedom Doesn't Create Innovation

For much of the last few years, the prevailing approach to AI has been experimentation. Give people access, encourage them to play, and see what happens. There's real value in that. It's difficult to understand a new technology without getting your hands dirty. But experimentation alone doesn't necessarily lead to innovation.

“Unlimited freedom doesn’t beget innovation.”

Andrew's point runs counter to some of the language surrounding AI, where limitless possibility is often positioned as the benefit. In practice, innovation tends to happen within constraints. A limitation gives us something to push against. A clearly defined problem gives us something to solve. Without either, a powerful tool can produce plenty of output without producing much value.

AI is exceptionally good at giving us something. It can answer a question, generate a concept, write code, summarize information, or create an image in seconds. That speed can make progress feel almost automatic, even when we're solving the wrong problem. Constraints force us to slow down enough to ask what we are actually trying to accomplish.

This does not mean teams should stop experimenting. Andrew draws an important distinction between familiarity and mastery. Someone holding a piece of sandpaper can accomplish a task, but a master craftsperson understands how that tool fits into a much larger process. The same is true of AI. We need the freedom to experiment enough to understand what these tools can do, where they struggle, and how they might fit into our work. Mastery starts when we develop the judgment to know what they should do.

The goal is not to experiment forever. At some point, the question needs to shift from What can this thing do? to What should we actually ask it to do?

Purpose Before the Prompt

This shift matters most inside organizations, where experimentation eventually has to translate into business value.

Andrew sees companies giving employees access to Copilot, Claude, and other AI products without giving them a framework for deciding where to apply those tools. Teams naturally start looking for tasks to automate or existing workflows they can make faster. That's not inherently bad, but it can keep us from asking whether the workflow itself should change.

Purposeful adoption requires a little more discipline.

• Start with the problem, not the technology. “Where can we add AI?” is rarely the best opening question. Start by understanding what someone is trying to accomplish, where the friction exists, and what a better outcome would look like. AI should earn its place in the solution rather than becoming the reason a solution exists.
  ‍
• Don't confuse efficiency with value. Making an existing process faster can be useful, but faster isn't automatically better. Sometimes the bigger opportunity is to rethink the process entirely. If AI gives us new capabilities, applying them only to old workflows may leave much of their value on the table.
  ‍
• Experiment to learn, then narrow the focus. Playing with new tools helps teams understand their possibilities and limitations. Eventually, exploration needs to become application. That means moving toward a defined problem, a desired outcome, and a shared understanding of what the technology is being asked to accomplish.
  ‍
• Define the human role. AI can generate options and accelerate execution, but context, prioritization, collaboration, and judgment still matter. Teams need to be clear about where AI contributes and where a person remains accountable for the result.
  ‍
• Know when to stop. Faster execution makes it easier to keep going, even when an idea is headed in the wrong direction. Mature teams need to recognize when to abandon a prototype, reframe a problem or when AI isn't the right tool. Andrew sees this kind of decision-making as an area where many organizations still need to mature.

Taken together, these ideas point toward a different definition of successful AI adoption. It is not about how frequently a team uses AI or how many workflows have an AI component. It is about becoming more deliberate about what we ask technology to do and why.

Education provides an unusually clear example of what happens when that distinction becomes unavoidable.

‍Education Is Already Rethinking the Rules

When generative AI entered the mainstream, schools found themselves at the center of the conversation almost immediately. If a student can ask ChatGPT to write an essay, what happens to the essay assignment?

The first instinct was understandably defensive. Schools explored AI detection, plagiarism controls, restrictions, and outright bans. Andrew is skeptical that this approach can work for long. AI is increasingly becoming part of the world students are preparing to enter, and trying to keep it outside the classroom risks preparing them for a reality that no longer exists. The more interesting question is not how to preserve every existing assignment in the age of AI. It is what those assignments were supposed to teach in the first place.

For decades, education has relied heavily on tests, essays, and knowledge recall because they provide practical ways to evaluate students at scale. Those methods are imperfect. Plenty of smart students have struggled with traditional tests, while succeeding when asked to apply their knowledge in different ways.

AI makes that tension harder to ignore, but it may also give educators new ways to address it.

Andrew points toward more experiential forms of learning where students demonstrate their abilities through problem-solving, collaboration, communication, decision-making, presentations, and real-world projects. Instead of asking only whether someone knows the right answer, educators can look at how they work through a problem and apply what they know. Historically, those experiences have been difficult for one teacher to facilitate and evaluate at scale. Emerging technology could help change that.

This changes the role AI can play in education. Instead of simply producing the assignment for the student, it can become part of a richer learning environment, assisting with the work while allowing educators to evaluate a broader set of human skills.

It also raises a question that extends well beyond the classroom: How do we use AI to enhance someone's abilities without letting it replace the thinking they need to develop?‍

‍What Businesses Can Learn From the Classroom

At first glance, education's struggle with AI can feel pretty specific to education. Look a little closer, and the parallels with business are hard to miss.

A school cannot give students access to ChatGPT and assume they will instinctively understand when to use it, how much to trust it, or when relying on it undermines the skill they are supposed to develop. A company cannot give its employees AI tools and assume those things either.

In education, meaningful adoption requires more than access. Administrators have to understand new capabilities, educators need professional development, and those ideas eventually have to translate into classroom practices that set students up to succeed. Andrew describes this as a process that moves through the educational system, not a tool that simply appears in a classroom and transforms learning overnight. Businesses need a similar level of intentionality.

That starts with frameworks. Teams need a shared understanding of what good output looks like, what context AI needs, what work can be delegated, and where human responsibility remains. Security and governance matter, but those guardrails alone don't teach people how to make good decisions. Andrew argues for an approach where teams agree on the context they're working within rather than everyone independently experimenting in different directions.

Businesses can also borrow from experiential learning. We probably won't build meaningful AI capability through a presentation about prompting or a company-wide memo encouraging everyone to experiment. People learn by applying tools to real problems, seeing where they help, recognizing where they fail, and adjusting along the way.

A product team might experiment with AI while working through an actual research synthesis. A designer might test where generative tools accelerate exploration without handing over the creative decision. A business leader might use AI to surface options while retaining responsibility for choosing which option makes sense in context. The specific use case will change, but the learning happens through doing.

That's important because the mechanics of AI will keep changing. Today's preferred model or prompting technique may not matter much in a few years. Defining a problem, evaluating an answer, providing context, collaborating, and knowing when not to use a tool are much more durable skills.

The goal isn't necessarily to build a workforce that's good at AI. It's to build a workforce that's good at solving problems with AI available to them.

That is a subtle difference, but an important one. It puts the problem back at the center and keeps technology in its proper role as a tool for solving it.

‍The Best AI May Eventually Disappear

There is another possibility that makes all of this focus on learning how to “use AI” feel somewhat temporary.

The best technology often becomes invisible.

Andrew uses the light switch as an example. An enormous technological system underpins the simple act of illuminating a room, but nobody walks over to a switch and marvels at the infrastructure that makes it possible. We flip it, the light turns on, and we move on with our lives. AI may eventually follow a similar trajectory.

That possibility is particularly interesting from a product perspective because much of today's AI experience still asks users to understand the technology. We open a dedicated tool, learn how to prompt it, review its output, and consciously think about the fact that we are interacting with AI.

A more mature experience could look very different. Instead of requiring everyone to become an AI power user, technology could quietly remove complexity, anticipate needs, and make sophisticated capabilities accessible to people who have little interest in understanding what is happening underneath.

“How do I make this invisible? How do I make this so intuitive? How do I make this so delightful?”

‍

Andrew sees those questions as worthwhile objectives for product managers. The person on the other side does not necessarily need to notice the technology. They need the product to work. A mature AI product may not make people better at using AI. It may eliminate the need for them to think about AI at all.

The Tools Aren't Magic

For all the attention given to what AI can automate, generate, or accelerate, Andrew's final takeaway from our conversation puts the responsibility back on us.

The tools aren't magic.

They can help us explore possibilities, execute faster, and solve problems in ways that were previously impractical. What they cannot do for us is decide which problems deserve our attention, provide every piece of context that matters, or take responsibility for whether the outcome is actually good. As Andrew put it at the end of the episode, AI can help solve a problem, but it is no replacement for doing the work.

Learning to use AI well may therefore have less to do with mastering every new model than with strengthening the human skills around it. We need to get better at defining problems before solving them, creating constraints that give experimentation direction, evaluating what comes back, knowing what to delegate, and recognizing the thinking we should retain for ourselves.

If we do that well, an interesting irony might emerge. The better we become at deciding where AI belongs, the less visible AI itself may eventually become. We may stop measuring progress by how many AI features we have added and start measuring it by something much more useful: whether the products, services, workplaces, and learning experiences we create actually work better for the people using them.

The next phase of AI will not be defined by who has the most powerful tools. It will be shaped by who develops the judgment to know which tool to use, which problem is worth solving, what work to delegate, and what thinking they should never give away.