Context engineering / AI operations

So most people have heard of prompt engineering by now, but there's a new term everyone's talking about context engineering. And so I wanna talk about that.

If you don't know, prompt engineering is just the skill of writing effective prompts in ChatGPT or other large language models. In the early days of ChatGPT, there's a lot of skepticism about whether or not this was really a skill.

At this point, I think anyone who spent at least 1,000 hours talking to an L L m will agree not only is prompt engineering a skill, there's actually a very high skill cap. So what is context engineering?

Well, as you might have guessed, this is just the skill of being able to provide all the context necessary for the L L m to perform the specific task. My understanding is that the CEO of Shopify is the one that coined this term, but I think the reason we didn't have a name for this until recently is because only recently have L L m's been able to handle a large amount of context.

A year ago, if you gave an L m more than 1,000 words, it would just end up diluting the most important parts...

he better I've gotten at working with LMS, the better I've become at working with humans. I think we have this natural tendency to just be very lazy when we're given instructions.

This makes sense from a biological point of view, right? We never want to expend more energy than we need to.

What I found, whether you're working with a human or a large language model, is that it's worth the five to 15 minutes it takes to provide all the instruction and context they'll need to do the task well. And if this is a task that you plan on doing multiple times, it might well be worth it to spend 30 minutes, an hour, or even an entire day putting together very specific instructions, context, and examples for the lolm or human.