AI Literacy Was Never Enough
Most companies taught their teams to prompt. Almost none taught them to manage.
I have been training teams on AI for the better part of two years now. And for most of that time, what we delivered was good. We took businesses that had never touched these tools and showed them what was possible, walked them through use cases specific to their industry, got them comfortable with the interfaces, and left them with workflows they could run on Monday morning.
But something happened in the last few months that I think matters more than any model release or product announcement I could write about this week. The questions coming from my clients started sounding different. They stopped asking “how do I use this tool” and started asking something harder, something I did not have a ready-made workshop for.
How do I know what my team is actually running?
How do I evaluate whether an AI agent is doing what I think it is doing?
What happens when someone on my team builds a workflow I never approved, using a tool I have never heard of, on a personal plan that sends our client data to a server I cannot audit?
I sat with those questions for a while because they felt like they belonged to a different discipline entirely, less about learning a new technology and more about managing one that had already arrived and embedded itself into the work before anyone had written a policy for it.
The AI training industry bloomed in 2024, and I understand why. Companies were panicking, employees were curious, and the market responded with the only product it knew how to build quickly: courses. Prompt engineering workshops and “Introduction to ChatGPT” half-day sessions and AI literacy programmes that taught people what a large language model was, how to write a decent prompt, and why hallucinations happen. When the technology is new, the first instinct is always education, get people comfortable and remove the fear, and that instinct was reasonable.
But here we are in 2026, and Deloitte’s latest State of AI in the Enterprise report, based on a survey of over 3,200 business and IT leaders across 24 countries, says that insufficient worker skills remain the biggest barrier to integrating AI into existing workflows. The number one way companies have responded is education, broadening AI fluency across the workforce. And then eighty-four percent of those organisations did not redesign a single job or workflow around AI. The training happened and everything else stayed exactly where it was.
I think that the industry confused the floor with the ceiling. Literacy was always a starting point, and most companies treated it as the destination.
And while the courses were running, employees were drawing their own conclusions about the gap between what they were taught and what they actually needed. Netskope’s 2026 Cloud and Threat Report found that 47 percent of generative AI users in the workplace are still accessing tools through personal accounts that bypass enterprise controls. WalkMe’s 2025 survey found that nearly 60 percent of employees say it takes longer to figure out an approved AI tool than to complete the task without it. So people reached for whatever worked fastest, whether or not it had been sanctioned, and according to a CybSafe study of over 7,000 participants across seven countries, 38 percent of them shared sensitive company data with AI tools their employer did not know about.
I want to be careful about how I frame this because the instinct is to treat it as a compliance failure, a discipline problem, something to be policed. But I think it is more honest to say that employees were trying to keep up with the pace of their own work, and the training they were given did not equip them to do it safely. IBM’s 2025 Cost of a Data Breach report found that 63 percent of organisations either do not have an AI governance policy or are still building one. The Awareways Trend Report found that fewer than 11 percent of the AI applications being used in workplaces are visible to IT teams.
Leadership, in many cases, thinks things are going well because the output is faster, which it is, but faster output without oversight is a liability wearing the costume of progress.
Where I have been spending most of my time as a consultant is the next layer up, teaching teams how to use specific AI tools for specific use cases within their actual workflows. This is better because it connects to the work. But I have started to recognise that even this is limited, because it is still tool-level thinking, and the tools themselves change every few months. Training someone on a specific interface does not prepare them for the moment when that interface is deprecated, or when a new capability arrives that reorganises how the whole task should be structured.
What my clients are now asking for, and what almost nobody in the training market is providing, is something I would describe as AI management competence. This is the layer where a team lead understands what it means to orchestrate multiple AI capabilities across a department.
And the reason this matters so urgently right now is that AI is no longer a tool you open in a browser tab. It is becoming an operational layer that runs underneath everything, with agents executing tasks autonomously and models embedded inside platforms your team already uses, sometimes without a separate purchase or approval process.
The surface area of AI inside an organisation is growing faster than any training programme can cover, and it will keep growing regardless of whether your governance catches up.
Literacy is, by definition, a baseline. You would not call someone an effective manager because they can read. The ability to read is assumed, and what makes someone effective is what they do with it:
how they assess risk,
make decisions,
allocate resources,
oversee the work of others.
The same logic applies here. Knowing how to use ChatGPT, Claude or Copilot is where everyone starts, and it is not where anyone should stop. Knowing how to manage an organisation where AI is embedded in the work, where employees are using tools you may not have approved, where data is moving in ways you cannot see, where the roles themselves need to be redesigned around new capabilities, that is the competence that is missing.
And because it does not have a catchy name, it does not have a budget, which means it is not being taught.
I think that is starting to change. The clients I work with are asking what it means to be an AI-ready manager, what governance looks like at the scale of a 30-person firm that does not have an enterprise legal team, how to give their teams access to powerful tools without losing visibility into what those tools are doing. And I am honest enough to say that the training I was proud of eighteen months ago would not answer those questions either.
The language will catch up. The competence cannot wait.
Teach the orchestration.
All the Zest 🍋
Cien
Cien Solon is a founder and AI transformation strategist working at the intersection of people, platforms, and power. Through LaunchLemonade, she helps organisations design AI systems that are dependable, governable, and human-centred.
Sources and further reading
State of AI in the Enterprise 2026: The Untapped Edge, Deloitte AI Institute
Cloud and Threat Report: 2026, Netskope Threat Labs
AI in the Workplace Survey 2025, WalkMe
Oh, Behave! Cybersecurity Attitudes and Behaviors Report 2024-2025, CybSafe and National Cybersecurity Alliance
The State of Shadow AI 2026, Unseen Security (aggregating IBM and Awareways data)
This is super insightful, thanks for your thoughts