What constitutes context?
Context may consist of the question itself, previous messages, documents, instructions, examples, roles, company information or desired output. In practice, anything that helps the model provide a more targeted response forms part of the context.
Why is context so important?
A model without context usually provides a general answer. As soon as you add your objective, source information and constraints, the same AI becomes much more specific. The quality of an answer therefore often depends less on the model alone and more on the context you provide.
What types of context are there?
You might consider task context, such as exactly what the user wants; document context, such as relevant sources; and organisational context, such as tone of voice, policy or process agreements. The better these layers align, the more useful the output becomes.
What often goes wrong?
Many applications provide too little, too much or the wrong context. Too little context results in superficial answers. Too much context makes the answer slow or muddled. The wrong context causes a model to go in the wrong direction, even if the question itself is logical.
A practical rule of thumb
Provide only what is necessary to carry out the task properly. Think in terms of the decision or answer you wish to receive, rather than the urge to dump all the information at once. Good context is focused, relevant and up to date.
From concept to practice
How AITJE applies this
With token optimisation, we investigate what information the model actually needs at each step. Code can retrieve customer data, select fields or perform a calculation before a targeted instruction is sent to the model. This builds context for the task, rather than sending a complete archive each time. We compare the result, token usage and correction work with the original approach.
Token management & optimisationFrom explanation to application
Would you like to know what this means for your work?
