From boardrooms to manufacturing plants to hospitals, Artificial Intelligence (AI) is quickly reshaping not just how we work, but how we think about what we need to be successful at the work we do. Everyone is asking the same question: ‘What do I need to know to use AI successfully in my role?’ But I don’t think that is the right question. The question we should be asking ourselves is: ‘What do I need to protect while I use AI in my work?’ The only constant in the workforce today is change. And humans have become experts at pivoting and adapting to the constant change that the tech era has brought us. On one hand, AI is very unremarkable in its similarities to every other technological advancement in history. The principles of change management, the pockets of resistance, the pockets of early adopters — all of it is very familiar. But on the other hand, it is unique and formidable in its ability to impact how we think. AI is the first technological advancement that promises intelligence, and we need to reimagine what this means for humans in the workplace.
The Curious Case of Bob and the Corporate AI
At a public-facing institution, Bob, with over three decades in a highly technical oversight role, used his organization's enterprise-deployed AI tool to write a section of a critical report. When questioned on the underlying analysis for the AI-generated section, he responded that he’d gotten it from the enterprise AI, treating it as factual. The information turned out to be hallucinated. The report would have been produced for public consumption at institutional scale. The impact would have been swift, and the institution’s credibility immediately compromised. Yet for Bob, the AI wearing a corporate badge became an inherent voice of authority. After all, no organization would deploy a tool to its workforce that is known to produce erroneous information with confidence, right? What should keep every leader and manager awake at night about this incident isn’t simply that it happened, but that Bob operated with no malice. He was doing exactly what organizations across the world are telling their workforce to do: adapt and use the tools we give you. And the tool of today is AI. Bob isn’t a defector, and that’s exactly what made his trust a vulnerability.
The Impact of Unverified Acceptance
When organizations deploy tools, they envision greater efficiency, greater effectiveness, better service and products, higher revenue, and exceeding performance metrics, but what they don’t imagine is deploying a failure mode at scale. Failure Mode One: The branding effect. Bob didn’t simply trust the AI. He trusted the AI that was the organization’s AI. The enterprise wrapper, the logo, the branding, all conveyed authority. It performed a quiet trust validation in his mental model. Failure Mode Two: The knowledge gap. Bob didn’t just fail to verify; he failed to understand the thing he was talking to. He treated the output of the AI as settled fact because he lacked understanding of how that output was generated. Failure Mode Three: The verification cliff. Bob reviewed his report, checked sources for the portions he wrote, and then just didn’t for the portion the AI generated. He verified his own work, but not the AI's. Failure Mode Four: The scale math. The risk was never just the model’s error rate. It’s Error Rate x Unverified Acceptance x Impact = Exposure, and enterprise deployment maximizes the trust variable right at the moment it pushes out the tool to maximum users at maximum stakes.
It’s Not What to Offload. It’s What to Protect.
Fluency is the only lever an organization can pull to calibrate employee trust in the tool and increase validation where the expertise to validate resides, which is in the employee. When we think about tech fluency, we can’t simply think about how to use technology, and specifically AI; we must consider what the necessary KSAs are that survive working with a system that has its own intelligence and can directly influence our decision-making. Our thinking, judgment, and decision authority are the things worth protecting — and protecting them is the fundamental competency. Knowing how to evaluate and validate output from AI and other technology is how we not only protect our own cognition but also how we protect the organization.
You Can Only Validate What You Still Own
Evaluation isn't a skill you sprinkle on top of full delegation; it's structurally impossible once you've handed over the decision itself. Validation is a function of retained authority. Thus, when we think about tech fluency as a competency, it isn’t simply defining technical prowess. It’s defining the behaviors that build and strengthen the muscles of evaluation and validation.
What Tech Fluency Is Not
- Prompting without direction
- Tasking without oversight
- Using without understanding
- Occasional personal use
- Getting to the output without understanding how it was derived
What Tech Fluency Is
- Understanding the basics of what the technology does and what it does well
- Understanding where humans-in-the-loop are necessary
- Understanding where technology is deployed in your functional area
- Understanding technology in a workflow, not just abstractly
- Understanding how technology influences your behaviors and shapes your cognition
- Understanding how to strengthen the skills that protect your ability to evaluate and validate
- Understanding what verification tools like watermarks, detection results, and source checks do and don’t prove
The Persistent AI Agent and the Human Who Stopped It
In July 2026, the AI Security Institute (AISI), a research organization within the UK government designed to build understanding of AI risks and solutions, logged an incident where AI Agents exhibited unprecedented deception during a cyber evaluation. The agents created fake online identities, conducted research, planted prompt injections for other AI systems to execute, and then coordinated with other agents to attack an open-source project. The main safeguard that stopped the attack was a human. Not a researcher, but a human who saw suspicious activity and questioned it before acting. AISI stated that “In these cases, standard good practice, human judgment, and caution around AI-generated code stopped the worst outcomes. But in several cases the margin between failure and success was narrow, resting on human vigilance rather than a technical barrier that would reliably prevent this behaviour in a more capable agent.” The parallelism is striking. In this incident, the AI was doing exactly what AIs are designed to do at their best, which is to execute a goal with speed, scale, and tireless activity. The barrier that stood between their efforts and success was a human with enough fluency to look at activity that seemed plausible, who first paused to verify before proceeding. That human prevented unknown harm, and not because they thought they were doing something extraordinary, but because they did exactly what their trained muscles taught them to do. Bob didn’t have the trained muscle because he had never been taught.
The Science Says This Matters
Bob isn’t an anomaly. Bob is the product of a society that has pushed technology into daily life with little education about what the technology is or how it impacts us. From the internet to social media to algorithms and engagement culture, and now to AI, technology has moved from being calculators to behavioral influencers. We’ve been measuring this muscle for a decade, and the results should reframe how we think about what being “tech-savvy” means. When Stanford researchers assessed high school students, more than 96% failed to question why a fossil-fuel-funded website might be unreliable on climate. Growing up with technology, it turns out, builds comfort, not judgment. And organizations can’t buy their way past it either: the Organisation for Economic Co-operation and Development (OECD) found that countries investing most heavily in classroom technology saw no improvement in student performance. Access is not capability; investing in tools without building the underlying competency does not build functional expertise; it builds tool users. Most sobering for those of us in the workforce business: the gap doesn’t age out. In the latest international adult skills assessment (PIAAC), about a third of U.S. adults scored at or below Level 1 in adaptive problem solving — the ability to handle changing conditions and filter irrelevant information, which is the specific muscle that evaluating AI output requires. Microsoft Research found that self-confidence was predictive of whether critical thinking was used in an AI-assisted task, and the extent of such thinking during that task. In their study, higher confidence in AI was associated with less critical thinking, while higher self-confidence was associated with more critical thinking. The throughline in these studies is the starkest picture for why Bob is not the anomaly and underscores why interventions are needed today.
Practically Beta’s Tech Fluency Lens
Current research is showing a clear mandate: protect thinking and build the necessary skills that allow proper evaluation and validation of tech-generated output. In the future, most roles will require some level of tech fluency embedded as a competency; this is already happening today. When organizations develop or revise their tech fluency models, it should focus on the essential skills, knowledge, and behaviors that allow an employee to perform their role effectively alongside the technology they use daily. At every career stage, an employee should not simply be aware of available tools; they should be repeating behaviors that crystallize the skills that clearly separate where their judgement, nuance, knowledge, and expertise remain separate and distinct from those of the tools they use. And just as importantly, how those tools impact them. Employees should be asking questions like: ▪︎ How does repeated exposure to this technology change my perspective or my decision-making? ▪︎ What are the things I am not aware of because it exists in the negative space between the outputs? ▪︎ Is this tool still serving its purpose, or am I using it out of habit? ▪︎ Do I understand where this output came from, or am I assuming it’s correct because I don’t know how to get there myself?
Looking Through the Lens: Illustrative Questions by Career Stage
The concept of this muscle may feel abstract, but there is already precedent in the difference between foundational mastery and amplification with a tool. The use of the calculator is the simplest example of this muscle. When students enter kindergarten, they aren’t immediately given a calculator. Even in the most advanced schools with access to calculators, advanced software, and tools for analysis, educators first teach the fundamental theories of math. The system ensures that students understand why 2+2=4 before handing them tools that could simply provide the solution to them. The reason this is done is that if students were only ever provided answers, they would lack the foundational knowledge and skills to assess if the answer was right or wrong.
With AI, we ran the sequence in reverse: we handed the workforce the calculator first and told them to adapt. We are building dependency on the tool, and not reinforcing that foundational expertise matters even more now, not less. It is easy to forget there was a whole era before AI, and humans produced remarkable things during that time. The challenge is to now take the things that make us remarkable as humans and amplify them with the use of the technology available to us; not offload them and lose what makes us innovative and effective.
AI and employees in the workplace are a lot like students and their calculators. We want to use AI to be quicker, to standardize, to increase revenue, and to reduce errors, but we first need to understand what right looks like; and that can only happen if we define what knowledge, skills, and behaviors produce that competency.
We’ve Mapped the Case, Now What?
Hopefully by now the reason for the path ahead is clear. Developing tech fluency is not simply building learning architecture and innovative environments to encourage the use of technology; it’s reinforcing the core KSAs and expertise that predate the technology itself. After all, humans have been quite successful at analysis, advising, building, creating, developing, and more, since we became a species. Our challenge today is ensuring we don’t lose those inherent skills as technology progresses. One of the most critical steps is understanding where your workforce is currently. Using behavioral assessments like Palmer’s Workforce Adaptation Index and other needs assessment tools provides real data on where your organization is starting from. Do you have an organization that is very tech-savvy and tech-fluent? Do you have an organization where people are reticent because the work of evaluating and validating output on the back end feels daunting? Or something in the middle? Your best strategies are your targeted strategies. After you develop the tech fluency model that makes sense for your organization, you must map where your organization is today, and then you’ll be ready to implement.
Implementation looks like:
- Understanding why you must look at tech fluency through the lens of human impact, not just human use
- Building your competency model to the technological realities of today
- Assessing your workforce
- Mapping and implementing your learning and development strategies
- Developing continuous monitoring and reassessing intermittently
This work will be iterative because the advancements are continuous. If we look at this work as a single point in time, we’ll be behind in two years. Success looks like staying one step ahead, not five steps behind. Track the technology, measure the workforce, provide them with the foundation of understanding and development opportunities, measure and adjust, and you won’t simply build a more informed and adapted workforce; you’ll turn Bob from a liability into a compounding asset where his expertise is amplified by the technology he uses, and the organizational risk is minimized.
You’ll turn Bob from a good-faith failure into an inevitable success.
Practically Beta, by Karida