I use different AI tools depending on what I am trying to do, and I recently noticed something I hadn't paid much attention to before: I don't start with the same amount of context in each one.
With some tools, I am starting almost from scratch. With others, some context already exists from previous interactions or from the information available to the tool. That made me curious about how much of the quality of an AI interaction comes from the prompt itself, and how much comes from the context around it.
Showing instead of explaining
A lot of my early experimentation with AI focused on writing a better prompt. If the response wasn't quite right, I would add more detail or clarify what I expected. I still do that, but sometimes an example can explain what I want more easily than another set of instructions.
Take a program update for an executive. I can explain the audience, the level of detail and what the update should focus on. Or I can provide two or three examples of previous updates that worked well. Those examples give AI a better sense of what I am looking for without requiring me to describe every detail.
This is essentially few-shot prompting: giving the model a few examples to work from instead of relying only on instructions. Examples are one form of context. Previous corrections, decisions and the history around the work can add context too.
When previous context gets in the way
For a Program Manager, history matters. Knowing what was discussed before, what decisions were made and what has changed can make an AI interaction much more useful.
At the same time, that history needs some judgment. A discussion from a few months ago may have gone in a completely different direction since then, or something that looked important at the time may no longer matter. Sometimes I may also want to start fresh simply to get a different perspective.
What I want to find out
I want to experiment with this a little more by starting fresh, adding instructions, providing examples and then working with accumulated context. I am curious to see when context improves the work, when it mostly saves time and when starting fresh is more useful. If you use AI regularly, what has your experience been? Does more context usually help, or are there times when you prefer to start fresh?