Prompting is a skill that develops through practice. A practitioner starting out with AI tools produces adequate results with simple prompts. As they develop the skill, they become able to produce excellent results with more sophisticated prompt structures, and the range of work they can accomplish with AI expands substantially. The progression can be understood as a series of levels, each building on the previous ones.
Level One: Simple Prompting
At the first level, the practitioner writes basic instructions that direct the AI to a task. "Summarise this document." "Draft an email to the client explaining the delay." "Extract the main points from these meeting notes." Simple prompts are appropriate for simple tasks, and they work well when the task is unambiguous, the desired output is standard, and the practitioner is willing to accept whatever shape the output takes.
Simple prompting is where every practitioner starts, and it remains useful for low-stakes quick tasks throughout a practitioner's working life. A draft email to a colleague about a scheduling conflict does not require an elaborate prompt. Most practitioners spend a substantial portion of their AI use at this level because the tasks substantively are simple, and the cost of a slightly imperfect output is low.
Level Two: Structured Prompting
At the second level, the practitioner applies the five-component structure from Section 2 deliberately. Role, task, context, format, constraints. The output quality improves substantially compared with simple prompting on anything beyond the most trivial tasks. Structured prompting is the working level for most meaningful professional use of AI, and it is the level the practitioner should aim to develop within the first month or two of deliberate AI use.
The shift from Level One to Level Two is where practitioners typically report the largest productivity gains. The additional time invested in writing the prompt is modest. The improvement in output quality is substantial. The ratio of output-editing time to prompt-writing time shifts sharply in the practitioner's favour, and the work starts to feel substantively accelerated rather than marginally helped.
Level Three: Few-Shot Prompting
At the third level, the practitioner provides examples of what they want as part of the prompt. Instead of describing the desired output in abstract terms, they show the AI one or two examples of the kind of output they're looking for, and then ask for similar output for the new case.
Few-shot prompting works particularly well for tasks that have a specific structure or voice that is hard to describe but easy to demonstrate. A practitioner who wants client emails in the firm's specific voice can include two or three examples of past firm emails as part of the prompt, and the AI will produce a new email that matches the voice of the examples. A practitioner who wants deal summaries in a specific format can include a few examples of prior deal summaries, and subsequent summaries will follow the same format.
The examples work because the AI's training included producing similar output when given examples, and examples carry more specific information than abstract descriptions typically can. The practitioner who says "write in the firm's voice" gives the AI something vague to act on. The practitioner who pastes three examples of the firm's voice gives the AI something concrete. The concrete version produces better output.
Level Four: Chain-of-Thought Prompting
At the fourth level, the practitioner asks the AI to reason through the task step by step rather than jumping directly to the output. "Think through this carefully, explaining your reasoning at each step, before producing the final recommendation." Chain-of-thought prompting consistently produces better results on tasks that require multi-step analysis, because it forces the AI to work through the analysis in a visible, checkable way rather than producing a conclusion that may not reflect careful reasoning.
Chain-of-thought prompting is particularly valuable for analytical tasks where the reasoning matters as much as the conclusion. A practitioner evaluating whether a particular course of action makes sense can ask the AI to reason through the considerations, weigh the trade-offs, and explain how the recommendation follows from the analysis. The resulting output is substantially more useful than a bare recommendation, because the practitioner can inspect the reasoning and identify where the analysis was strong, where it was weak, and where their own judgment should adjust the conclusion.
The trade-off is that chain-of-thought outputs are longer and take slightly longer to read. For tasks where the practitioner only needs the final output, the additional reasoning adds noise rather than value. The skill is knowing when chain-of-thought is worth requesting and when a direct answer is more useful.
Level Five: Role-Chaining and Multi-Step Prompting
At the fifth level, the practitioner structures complex tasks as sequences of prompts rather than single prompts. The output of the first prompt becomes input to the second. The output of the second shapes the third. A consultant producing a strategic analysis might use a first prompt to generate a list of relevant frameworks, a second prompt to apply the most relevant framework to the specific situation, a third prompt to identify weaknesses in the analysis, and a fourth prompt to produce a final recommendation that addresses the weaknesses.
Role-chaining extends this pattern by having the AI play different roles at different stages. The first prompt might instruct the AI to act as an analyst producing an initial view. The second might instruct the AI to act as a senior reviewer critiquing the initial view. The third might instruct the AI to act as the analyst again, revising the view to address the reviewer's critiques. This pattern produces output that has been improved through internal iteration within a single session, and it often reaches quality that a single-prompt approach cannot reach.
Role-chaining is an advanced skill. Most practitioners do not need it for most of their work, and trying to apply it too early tends to produce complexity that is not earned by the results. It becomes useful when the task is substantively complex enough that a single prompt cannot capture everything that needs to happen, and when the practitioner has developed enough judgment about what the AI can and cannot do to structure the sequence effectively.
What Progression Looks Like
A practitioner who starts deliberately developing prompting skill typically moves from Level One to Level Two within a few weeks, reaches competence at Level Three within a few months, and develops Level Four and Level Five capability over longer periods of deliberate use. Not every practitioner needs to reach Level Five. Level Two is the working level for most professional use, and a practitioner who stays at Level Two competently produces substantial value from AI tools across their working life. The higher levels expand what is possible when the task warrants the investment.