The Art of the Current: Navigating the Future of AI

Four legs. A seat. Maybe a back. Something designed for sitting.

But what about a stool? A beanbag? A yoga ball balanced on four legs? An art installation shaped like a chair that no one is allowed to sit on? A horse has four legs and can be sat on, but few would argue that makes it a chair.

The closer you look, the harder the definition becomes.

That idea reflects something engineer John Salvatier explored in his essay Reality Has a Surprising Amount of Detail. Salvatier explained that things that appear simple from a distance often become remarkably complex up close.

Consider the familiar claim that water boils at 100 degrees Celsius. That is true, sometimes. Purity, altitude, the material of the container, and even the addition of salt can influence the boiling point.

AI adoption has a similar problem.

 

When AI Adoption Hits a Dam

The Elwha River in Washington state offers an unexpected lesson in transformation.

 

 

Beginning in the early 20th century, dams disrupted the river’s natural flow and its salmon population. When the dams eventually came down, engineers expected the ecosystem to take decades to recover.

Instead, the river began, within hours, returning to its former channel. And after just two years, the salmon were returning as well.

The river did not need to be taught how to flow. It didn’t need a multi-step strategic framework for recovery. It just needed the obstruction removed.

That is a useful way to think about what is happening inside many organizations investing in AI.

Companies are purchasing thousands of licenses, securing executive support, developing rollout plans, and offering training. Yet employees may still struggle to integrate AI into their actual work. The technology is available, but the value is not flowing.

One Fortune 100 company encountered exactly this challenge after rolling out Microsoft Copilot across its team. Employees had access to independent learning resources and lunch-and-learns, yet adoption remained difficult. Some employees did not have time to explore. Others were overwhelmed by the technology and did not know where to begin.

The problem was not access. It was the environment around it.

AI had been introduced into workstreams that were more complicated than expected, with employees carrying different levels of experience, uncertainty, and resistance. Rather than creating a natural channel for AI to enter everyday work, the rollout had unintentionally created a dam.

Removing it required a different approach to learning.

 

Creativity Needs an Open Mode

In a 1991 talk about creativity, John Cleese argued that creativity is not a talent. It is a way of operating.

Drawing from psychologist Donald MacKinnon’s research into creative professionals, Cleese described where curiosity lives the “open mode”. This notion is conceptualized by the idea that creativity thrives in a state in which people have the space, time, confidence, and humor to explore possibilities without immediately needing to arrive at the right answer.

Those conditions are particularly relevant to AI.

When organizations introduce a new AI platform, employees are often given deadlines, compliance requirements, training materials, and expectations for adoption. What they are rarely given is permission to simply explore.

There is good reason for structure. Organizations cannot operate indefinitely in experimentation mode. Cleese’s corresponding “closed mode” is focused, purposeful, and execution-oriented. It is where ideas become decisions and work gets done.

But AI adoption requires both.

The most effective approach switches between them: open mode to explore, closed mode to build, open mode to assess, closed mode to scale. Back and forth.

That rhythm favors creativity over urgency and helps break free the organizations stuck behind the dam.

 

Permission to Play

That means AI experimentation cannot simply be another task added to an already full workload. People need permission to play.

And play, in this context, does not mean treating AI casually. It means creating organizational conditions in which experimentation is genuinely possible. Employees need to know that they can test ideas, ask basic questions, and encounter failures without feeling as though they are falling behind. Because fear is its own kind of dam.

People may worry about breaking something. They may worry about looking uninformed in front of colleagues. Some may worry about what AI means for their jobs. Others may simply feel overwhelmed by how quickly the technology is changing.

The first step, then, is not training. It is listening.

For the Fortune 100 organization, strategic discovery sessions brought together everyone from experienced users already experimenting with custom agents to employees who had barely opened the tool.

Those conversations exposed an important distinction: the challenge was not a lack of interest. It was a lack of clarity.

Employees described previous experiences with AI that had not met their expectations. Training did not always feel relevant to their roles. Messages from leadership could feel inconsistent. And many had never been asked what they actually needed from the technology.

So, the approach was to start with the people. Not the platform.

 

From AI Training to AI Enablement

The next step brought 150 employees together across 10 in-person sessions over four days and two locations. The goal was not to turn everyone into an AI expert. One workshop could never accomplish that.

Instead, the sessions were designed to build confidence, motivation, and curiosity. Copilot became less of an intimidating new technology to master and more of a practical tool that could help accelerate everyday work.

But the workshops also surfaced another problem: there was no single employee learning journey.

Some people needed foundational prompting techniques. Others were ready for advanced capabilities. Learning preferences differed. So did the amount of time people could dedicate to training. And role-specific questions appeared constantly.

The same organization, department, and technology could still produce completely different learning needs.

At the same time, AI itself was evolving too quickly for a traditional curriculum to remain current. A course built today could reflect a very different technological landscape only weeks later.

That led to a bigger question: What if every employee could have a learning partner that evolved alongside both the individual and the technology?

 

Building a Learning System That Can Evolve

The answer was an orchestrated agentic learning framework called the Capability Squad.

At its center was Kai, an employee-facing learning coach. Rather than simply pointing people toward a course catalog, Kai could learn about employees’ goals, preferred learning formats, available time, existing AI capabilities, and the workstreams they wanted to accelerate.

Behind Kai, several specialized agents also made up the Capability Squad and kept that learning experience current. Scout identified relevant learning materials from authoritative sources. Lex categorized them for the catalog. Archie created bridge modules as new technologies and capabilities emerged.

And Clyde played the important role of quality assurance, challenging whether materials were appropriate for their intended audiences, identifying redundancies, and questioning what needed to be updated or added.

Together, the five-agent system addressed one of the fundamental problems with traditional AI training: static curriculum cannot keep pace with a technology that is changing constantly.

Instead of creating a training project that would need to be refreshed again and again, the organization gained an asset designed to evolve.

Most importantly, the framework was grounded in what employees had already shared. Their questions, challenges, learning needs, and real workflows became the intelligence layer supporting the system.

 

Listen. Educate. Activate.

The approach ultimately came down to three ideas: listen, educate, activate.

Listen before deciding what employees need.

Educate in ways that meet people where they are rather than forcing everyone down the same path.

Activate by giving people the confidence and permission to apply what they have learned to real work.

For this organization, that rhythm resulted in 100% adoption, achieved organically rather than through a mandate or a metric tied to consequences.

It also transformed the role of use cases. Instead of sitting in a backlog of hypothetical ideas, they became a roadmap reflecting what employees needed now, what they would need next, and where the organization could go in the future.

That brings the story back to the river. From a distance, the Elwha looks like a simple blue line on a map. Up close, it is a living system shaped by countless variables.

A chair seems simple until someone tries to define exactly what makes it a chair.

And AI can seem straightforward until an organization tries to implement it at scale.

The instinct in moments of rapid technological change is often to move faster. Buy the tools. Build the curriculum. Launch the pilot. Identify the use cases. Measure adoption.

But sustainable AI adoption may demand something more patient.

Sometimes the most important work is not creating another structure to force transformation forward. It is understanding what is standing in the way, removing the obstruction, and giving people enough space to find the direction.

Because the organizations that make the most of AI may not be the ones moving the fastest. Instead, they may be the ones that trust the flow of the current.

 

 

 

The ideas shared here originate from Jennifer Lill’s Assembly 2026 presentation. To hear the complete conversation, watch the session, linked below.

 

 

Photo Credit: Susan Wilkinson | Unsplash