AI literacy needs to match each employee’s role

Organizations can spend billions of dollars on AI technology and still lack the people who can deploy, govern, and operate it economically. That gap changes what AI literacy means for technology leaders because access to technology does not supply the skills to use it well. A basic course may give employees useful vocabulary, but organizations also need role-specific capabilities tied to the systems they plan to build, the costs they must control, and the rules governing employees’ use of AI.

AI investment does not create AI capability

AI literacy starts with a basic understanding of AI concepts and capabilities that lets someone use the technology responsibly and effectively. That baseline matters across an organization because AI increasingly affects decisions about tools, workflows, data, and spending. The required depth then depends on responsibility: an employee using an AI assistant makes different decisions from an engineer responsible for putting an AI system into production.

Those different responsibilities matter when organizations invest billions of dollars in AI technology. Buying access to models, development tools, or infrastructure creates technical capacity, but people still have to decide where AI is appropriate, configure and operate systems, manage risks, and understand their economic behavior. AI literacy is therefore an initial condition for extracting value from those investments, preparing for regulatory obligations, and keeping technology spending under control.

For technology leaders, these demands turn AI literacy into an organizational capability with different depths for different responsibilities. Employees need enough shared knowledge to make informed choices about the AI they encounter at work. People who build and run AI systems need considerably deeper expertise because their decisions determine how those systems behave in production.

Skills are a barrier between AI pilots and production

The need for deeper expertise becomes clear when AI work moves from experimentation into production. Generative AI pilots have been described as failing at a rate of 95%, while 88% of agentic AI pilots are described as never reaching widespread deployment. Those figures show the scale of the production problem, but they do not identify why projects fail or deployments stall.

Skills provide a separate, more direct measure of one part of that problem. The Tech Skills Report found that 48% of IT professionals have abandoned projects because they lacked sufficient technology skills. A capability shortage can therefore have an operational consequence beyond slower adoption: work that has already consumed engineering time and organizational resources can end because the people involved cannot take it further.

For an AI project, the capability required changes as the project develops. Experimenting with generative output requires one level of understanding, while testing behavior, establishing security controls, managing infrastructure, tracking token consumption, and scaling a production service require deeper skills. Agentic AI, where software can perform tasks through AI-directed actions, adds engineering and oversight requirements when those actions extend across tools and systems.

These changing requirements make skills shortages one credible barrier between an AI pilot and sustained deployment. Technical, organizational, economic, and governance problems can also contribute to the reported failure and deployment rates, so skills do not explain every failed or stalled pilot. For a CTO, the actionable point is narrower: an organization that funds AI experiments without checking whether its people can build and operate the resulting systems leaves a known production barrier unresolved.

That production barrier also shapes workforce decisions around emerging AI roles. New roles create demand for specialized skills, which organizations can obtain through external recruitment or develop among existing employees. Hiring for niche AI expertise can be expensive and time-consuming, while developing existing employees is generally faster and more economical, so leaders need to determine which capabilities already exist internally, which can be developed, and where recruitment is required.

AI literacy supports deployment, cost control, and governance

The production problem leads directly to two other operating concerns: cost control and governance. The same underlying knowledge affects all three because employees and engineers make choices about how AI systems are built, used, and controlled. Once those systems enter routine use, those choices can determine operating costs and compliance with organizational or regulatory requirements.

Cost is particularly sensitive to usage because AI consumption can accumulate through repeated model requests, employee activity, and autonomous agents. Some reports have claimed that an unnamed company received a $500 million Claude bill after failing to set employee usage limits. The unnamed attribution makes the figure an anecdote rather than a benchmark, but the control problem is concrete: organizations need people who understand how model usage creates costs and where limits can be applied.

For teams operating AI, that cost knowledge translates into practical controls. Guardrails can constrain how systems are used, token limits can restrict consumption, and visibility into agents can show teams what automated systems are doing and consuming. Prompt engineering also has an economic dimension because more efficient interactions can reduce unnecessary model work, making cost management partly a workforce capability.

The connection between knowledge and operating decisions also applies to governance. The EU AI Act formally defines AI literacy as the “skills, knowledge and understanding that allow providers, deployers and affected persons, taking into account their respective rights and obligations in the context of this Regulation, to make an informed deployment of AI systems, as well as to gain awareness about the opportunities and risks of AI and possible harm it can cause.” The definition ties literacy directly to informed deployment and awareness of opportunity, risk, and potential harm.

The EU AI Act also requires organizations to take measures supporting AI literacy among their staff. For affected organizations, workforce capability consequently becomes part of regulatory readiness. The knowledge required still varies by responsibility because staff need enough understanding to act within organizational policies and their own duties, while people deploying systems need technical knowledge of the systems they operate.

Deployment, economic control, and governance place different demands on the same workforce. Deployment requires people who can execute technical work; economic control depends on understanding how usage and design choices affect spending; governance requires informed decisions about use, risk, and obligations. Those differences require multiple levels of AI literacy within one organization.

One organization needs multiple levels of AI literacy

Those multiple levels can start with a common organizational baseline. Employees need a high-level understanding of how AI works, what it can do, when its use is appropriate, and what responsible use means under their organization’s policies. Shared concepts give teams enough common language to discuss AI decisions without requiring every employee to become an AI engineer.

From that baseline, employees can learn the concepts behind the systems they encounter. These include generative AI and large language models, or LLMs, the models behind many systems that generate and process language. They also include Retrieval-Augmented Generation (RAG), in which a system retrieves relevant information to support model responses, and agentic AI, where AI systems can take actions toward a task. Prompt engineering and AI ethics add practical knowledge about directing these systems and making responsible choices about their use.

The required depth then follows the employee’s role, existing skills gaps, the organization’s objectives, and its technology stack. Someone who uses an AI assistant in routine work may need to recognize appropriate use cases, follow internal policies, and evaluate outputs. Someone responsible for an AI service needs enough technical competence to understand its architecture, test its behavior, secure it, deploy it, and keep it operating at an acceptable cost.

Tool literacy follows the same role-based model because employees need to understand the AI services relevant to their work and choose among them appropriately. Possible starting points include ChatGPT, Copilot, Gemini, Google Bard, Anthropic Claude, and Midjourney. Familiarity with these services helps teams select a suitable service for a task and apply the organization’s policies to its use.

For technologists, the required depth rises because they are responsible for working systems. Engineers and operators need practical ability to build, test, deploy, and scale AI systems, which makes hands-on work essential. An AI-ready engineering team may need experience with agentic coding tools, token and cost management, AI infrastructure, and security because these concerns arise when operating AI systems.

The Pluralsight Tech Learning Pulse gives a more specific view of where technology teams are concentrating their learning. Pluralsight has a commercial stake in technology learning, so it benefits when organizations invest in developing these skills. Its reported focus areas are still useful for understanding the kinds of implementation knowledge involved:

Within those areas, Model Context Protocol (MCP) concerns how AI applications connect with external context and tools, making it particularly relevant to engineers working on agentic systems. Multi-agent systems add another implementation layer because multiple AI agents may participate in a workflow. Engineers working with these systems consequently need deeper implementation knowledge than employees who encounter agents primarily as users or decision-makers.

RAG shows the same difference between baseline literacy and implementation skill. A general employee can benefit from understanding that an AI application may retrieve external information to inform an answer. A developer working with RAG needs practical knowledge to construct and test that behavior within a working system, so the role determines what it means to know the concept well enough.

Once the required technologies are known, learning paths can supply that specialized depth. Named offerings include Anthropic Claude 3, Model Context Protocol (MCP), Prompt Engineering, LangChain, and Retrieval Augmented Generation (RAG) for Developers. Because learning providers benefit commercially when organizations buy or adopt such training, these offerings are examples of ways to develop the identified skills rather than evidence that every engineering team needs the same curriculum.

That distinction produces an intentionally uneven capability model. An organization benefits from common language around generative AI, LLMs, RAG, agents, responsible use, and policy because teams need to communicate about shared systems and risks. Technical depth follows responsibility, so people making architecture, deployment, security, and cost decisions receive the implementation skills those decisions require.

Build capability from assessment

Once capability varies by role, development has to start by identifying the skills each team already has. Teams should first take a skills assessment so the organization can locate existing knowledge and meaningful growth opportunities. The results can then shape customized upskilling programs around actual needs.

The same assessment makes the earlier workforce decision about development and recruitment more precise. Mapping current capabilities can reveal employees who already have adjacent engineering or operational skills and can develop into emerging AI responsibilities. Leaders can then compare those internal options with requirements that still call for niche external recruitment.

A customized program can preserve organization-wide standards while allowing role-specific depth. The shared layer establishes language and understanding across technical and non-technical employees, while separate tracks address individual roles, business objectives, skills gaps, and the technologies the organization has selected. Common concepts and organizational rules provide consistency; the work people are expected to perform determines specialization.

AI Academy is one example of that structure, described as an end-to-end program spanning AI literacy through agentic capability. Its progression begins by establishing shared AI language and understanding, then moves from individual AI productivity toward organization-scale capability and measurable business outcomes. As a commercial learning program, AI Academy’s provider benefits when organizations adopt this kind of training, so the program illustrates one implementation of the model rather than defining the model itself.

With that structure in place, assessment becomes a repeatable planning process. Leaders can identify current skills, compare them with the capabilities required by roles and the technology stack, and direct development toward the resulting gaps. The curriculum then follows from what the organization is trying to build and what its people need to learn to build it.

AI literacy has to remain current

A repeatable process matters because technologies and required skills keep changing. New AI tools alter employee workflows, while changes in models, agentic systems, infrastructure, and engineering practices change what technical teams need to know to complete projects and generate business value. A fixed curriculum therefore begins aging as soon as those requirements move.

Maintaining AI literacy requires organizations to reassess skills and update learning as roles and technologies change. Expertise can then deepen where new operating requirements create a specific need. Workforce capability remains one condition for making AI work in production, and continuous assessment keeps that condition aligned with the systems people are actually expected to use, build, and operate.

Main highlights

• Tie AI investment to workforce capability: AI tools and infrastructure create value when employees have the skills to deploy, govern, and operate them. Technology leaders can align AI literacy with each role’s responsibilities and the systems the organization plans to use.
• Close skills gaps before scaling pilots: Skills shortages can stop technology projects and create another barrier between AI experimentation and production. CTOs can identify the engineering, security, infrastructure, and operational capabilities required before committing resources to wider deployment.
• Connect AI skills to cost and governance: Knowledge of token consumption, usage limits, guardrails, security, and responsible use affects both operating costs and regulatory readiness. Organizations subject to the EU AI Act also need measures that support AI literacy among staff.
• Set different skill levels by role: A shared foundation in generative AI, LLMs, RAG, agents, responsible use, and organizational policy gives employees common language. Engineers and operators need deeper, hands-on skills in the technologies they build, secure, deploy, and scale.
• Start capability building with assessment: Skills assessments show which capabilities already exist, which employees can develop into emerging AI roles, and where external hiring remains necessary. Technology leaders can use those findings to create role-specific learning paths tied to business objectives and the technology stack.
• Keep AI literacy current: Models, tools, infrastructure, and agentic systems continue to change the skills required for production AI. Regular reassessment lets organizations update training as technologies and responsibilities evolve.