What We Actually Mean by AI Literacy at Dixon AI

In our previous article, AI Literacy: The Foundation of AI Capability, we set out why AI literacy matters. The next question is the practical one: what do we actually mean by it?

At Dixon AI, we define AI capability as the combination of talent, purpose, AI literacy, AI tools and data infrastructure. Each matters, and each strengthens the others. An organisation can invest in tools and improve its data environment, but if its people do not understand how AI works, where it fits, and how to use it well, capability remains limited.

That is why AI literacy is such an important part of AI capability. It helps people make sense of the technology and apply it in real work. It helps them understand how AI works so they can make informed decisions about how and when to use it, and when not to use it. It gives people the language, context and confidence to judge where AI will help, where it needs oversight, and where a human approach is still the right one.

At Dixon AI, we think about AI literacy through seven connected areas:

Generative AI vs Retrieval AI, Models, Prompting, Bots, Agents, Applications and Ethics.

Together, these form the practical baseline for understanding modern AI in a business context.

1. Generative AI vs Retrieval AI

The starting point for AI literacy is understanding the difference between Generative AI and Retrieval AI, because that helps people determine the right type of AI for a particular task.

Generative AI creates new content. It can draft an email, summarise a report, suggest ideas, produce code, create an image or help write a first version of a proposal. It is often the first form of AI people encounter because it is flexible, visible and easy to experiment with. An important concept to understand is that Generative AI is probabilistic. That means the same model, given the same prompt, is unlikely to produce exactly the same output each time. The overall quality or direction may be similar, but the wording, structure or specific content can vary. That variability is part of how generative systems work.

Retrieval AI does something different. Rather than creating something new, it finds, surfaces and organises existing information from a defined source. That might be a company knowledge base, a set of policies, a contract library, a product database or project documentation. Its purpose is to bring back relevant data that already exists. When it is connected to the same underlying source and asked the same question, Retrieval AI is far more consistent and repeatable because it is returning information from a defined body of knowledge rather than generating a fresh answer each time. That makes it especially valuable when consistency and grounding matter.

This distinction matters because different tasks require different forms of AI support. If the task is to draft, simplify, brainstorm or synthesise, Generative AI may be the right fit. If the task is to find the latest approved policy, retrieve the correct wording from a contract, or ground an answer in trusted internal information, Retrieval AI is often the better choice. In many real business situations, the strongest setup combines both, with Retrieval AI supplying grounded information and Generative AI helping shape that information into something useful.

The difference between types of Ai is so important. People who do not understand it often use Generative AI when they need dependable retrieval, or they use retrieval-based systems when the task really calls for synthesis and creation. AI literacy begins with choosing the right form of AI for the job.

2. Models

Once people have determined the right form of AI, they then need to use the right type of model.

Most people do not need a technical explanation of how models are built, but they do need to understand that most of the Generative AI tools they use have different models, and that different models behave differently. The tool on the surface is not the whole story. The model beneath it shapes what the system is good at, where it struggles, how quickly it works and how much it costs to use.

Many organisations train people on tools rather than AI capability. Someone learns one tool and assumes they understand AI broadly, when in reality they may only understand the interface of that tool sitting on top of one particular model. That is a much narrower understanding than they realise.

Different models have different strengths. Some are better at writing and language tasks. Some are stronger at coding. Some are faster and cheaper. Others are slower but more capable. Some are better at reasoning through complex problems. Others are more effective for specific enterprise workflows. These differences matter because they affect output quality, speed, cost and suitability for the task.

For leaders, this is valuable because selecting the right model makes it more likely that people get the right output, get it more efficiently, and do so at the right cost and speed. A poor model choice can create frustration, weak outputs and unnecessary expense. A better model choice can improve quality, reduce time and make AI use much more practical at scale.

This also leads to better understanding about how to work with AI. The conversation becomes less about which tool is most fashionable and more about what capability is needed, what kind of model supports that capability best, and what level of quality, speed and control the work requires. That is a much stronger basis for decision-making, and it is also more durable as tools continue to change.

3. Prompting

Once people have selected the right model, the next question is how to get the best output from it. That comes down to Prompting.

At Dixon AI, we define Prompting as the skill of aligning the AI with your purpose. If you are clear about what you want to do, a good prompt helps the AI execute that for you more effectively.

A strong prompt carries the core ingredients the model needs to respond well. That usually includes the task itself, the context around it, the intended audience, the answer format, and any style or quality requirements that matter. Good Prompting is about being clear with your purpose, not clever tricks or technical ability.

One of the prompt structures we use most often is a simple one: Background, Role, Task, Format. In other words, what is the relevant context, what role should the AI take, what exactly is it being asked to do, and what should the output look like? This structure is useful because it forces the person using AI to think more clearly before they ask for anything.

Prompting is not just a tool skill, it is a crucial skill for working with AI. People who prompt well tend to define the purpose more clearly, give the right context, and review the answer more thoughtfully. They are not just asking AI for output. They are directing it towards a useful outcome.

Prompting also helps people understand the limits of AI. A vague prompt often produces a vague answer. A poorly framed task leads to a poorly framed response. A good prompt does not guarantee a perfect output, but it significantly improves the chances of getting something useful. More importantly, it helps the user stay in control of the intent behind the work.

4. Bots

Once people understand models and Prompting, the next part of the Dixon AI AI literacy model is Bots.

At Dixon AI, we define a bot as a combination of three things: a model, the prompt that tells the bot how to behave, and the data or knowledge base it is connected to. In other words, a bot is not just an AI tool in general. It is a more structured AI system designed for a more specific purpose.

The model provides the underlying capability. The prompt is a system prompt which shapes the behaviour. The data or connected knowledge base gives the bot relevant context or trusted information to work from. Put those three together, and you begin to move from one-off interactions into something more reusable and more operational.

Many people first encounter bots in practice by building or using Custom GPTs, ChatGPT Projects, Claude Projects or Copilot Agents. These environments make it easier to combine a model, instructions and knowledge into something that can support a recurring task.

That might be a bot that helps draft client updates in a defined tone of voice, summarises meetings into an agreed format, turns rough notes into first-draft proposals, or answers internal questions from a specific document set. The value is not just speed, but crucially the repeatability. Good design gets packaged once and then reused across many interactions.

Bot literacy matters because it helps people understand what they are really building. An effective bot can be simple to build. It is a designed combination of model, instructions and context. If that design is weak, the output will be weak. If the design is thoughtful, bots can make AI far more practical and far more relevant to everyday work.

5. Agents

After Bots comes Agents, and this is where the distinction between structured assistance and more autonomous execution starts to matter.

With a bot, people usually define the task quite explicitly. With an agent, the user is more likely to define the purpose or goal, and the agent makes decisions about how to achieve it. Rather than being told each execution step, the agent can work out a path towards the outcome.

It does this by autonomously calling skills and tools. That might mean searching for information, interacting with connected systems, retrieving data, triggering actions, checking progress, or moving through multiple stages of a workflow. This is what makes agents more powerful, but also more complex.

A useful way to think about it is this: a bot helps with a specific task, while an agent can help manage part of a process, or potentially even an entire process. That might involve coordinating steps, deciding what to do next, and using the tools available to achieve the goal with less constant human instruction.

Agents can be extremely valuable in the right setting. They can reduce manual effort, speed up routine workflows and help organisations manage multi-step work more efficiently. But they also raise the bar for judgement. The more freedom a system has, the more important it becomes to think about controls, escalation, accountability and oversight.

That is why Agent literacy is not simply knowing the term. It is understanding when an agentic approach is appropriate, when it is not, and what guardrails are needed around it. AI can support execution at an increasingly sophisticated level, but people still need to set the purpose, define the boundaries and remain responsible for the outcome.

6. Applications

Applications are where AI literacy becomes more practical and more market-aware. This part of AI Literacy is about having a good understanding of the wide range of AI applications available and being aware of AI capabilities across different products and categories so you can choose the right tool for any particular task.

In other words, AI literacy is not only about understanding the underlying concepts. It is also about knowing what kinds of applications exist, what they are good at, and how to choose the right product for the right use case.

That means being aware of far more than text generation alone. It includes things like digital avatars, voice generation, AI tools for creating PowerPoints, research assistants, design tools, coding assistants, AI video tools, workflow automation platforms and agentic systems. Different applications solve different problems, and organisations need a broad enough awareness to recognise what is relevant and what is not.

This matters because good AI use depends on matching the right application to the task. A team trying to create training content may need something different from a team trying to automate reporting, generate synthetic voice, or build an internal knowledge assistant. Without that awareness, organisations either default to one familiar tool for everything or miss useful categories entirely.

Applications also matter because they are where business value becomes visible. This is the point at which understanding turns into practical choices. Which product is best suited to this use case? Which application creates the most value for the least effort? Which category of tool is mature enough to use now, and which still needs caution?

We include Applications as part of AI literacy rather than treating them as a separate commercial issue because people need enough breadth of understanding to recognise what is possible, assess what is useful, and connect the right application to the right piece of work.

7. Ethics

We cover AI Ethics last because it underpins everything else. By the time people have worked through the rest of AI literacy, they have a much stronger grounding for thinking about AI Ethics in a practical way.

At Dixon AI, Ethics includes responsible and ethical use of AI, protecting data, thinking about bias, and considering AI safety. It means understanding not only what AI can do, but what risks come with using it and what responsibilities stay with the human user and the organisation.

That starts with data. People need to understand what information should and should not be entered into different tools, what happens when data is shared, and how to use AI in ways that protect commercially sensitive, personal or confidential information.

It also includes bias. AI systems can reflect biases present in data, in training patterns and in the way tasks are framed. People need enough literacy to recognise that outputs are not neutral simply because they sound authoritative. They need to question, review and apply judgement, especially in areas where fairness and consequence matter.

AI safety is part of this too. As AI systems become more capable, the risks are not only about inaccurate answers. They are also about over-reliance, poor oversight, inappropriate automation and misplaced confidence. A polished output can still be wrong. An efficient process can still be unsafe. Ethics helps people stay alert to that gap.

Most importantly, Ethics reinforces the principle that responsibility does not disappear when AI is involved. The technology may support execution, but accountability for purpose, choice and consequence remains human. That is why Ethics is not a final add-on. It is what keeps all the other parts of AI literacy grounded in responsible use.

Conclusion

At Dixon AI, AI literacy is the practical understanding that helps people get the best out of AI in a safe and responsible way. That is why it sits as a core part of AI capability. It helps people make informed decisions about how AI should be used, where it should be used, and where it should not. It gives organisations the shared language and judgement needed to move beyond surface-level experimentation and towards more confident, effective and responsible adoption.

Tools will keep changing. Models will improve. New applications will continue to appear. But organisations that build AI literacy deliberately will be better placed to adapt, make better decisions and create value from AI in a way that is sustainable.