I’m not the only one thinking about how context management is the key to good LLM applications. Since publishing our post detailing how long contexts fail, a conversation emerged regarding the term “context engineering,” compared to “prompt engineering.” (Let’s be clear…I had nothing to do with starting the debate, it’s just a happy coincidence…)

Today, Andrej Karpathy weighed in, supporting “context engineering”:

People associate prompts with short task descriptions you’d give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step.

As usual, Karpathy grasps a blurry idea and seems to snap it into focus, with with his authority, insight, and clear writing. Writing about long context fails, more than once I’d write the word “prompt” somewhere before quickly replacing it with “context.” When prompts are part of software, they’re context. As an ecosystem, I think we all arrived there at once.

I am a nerd for terminology and selecting the right word. Anytime I find myself debating terms, seeking to map out the detail they bring to our discussions, I end up making a table, like so:

'Prompts' versus 'Context'

Note that one isn’t better than another! They’re just different. As we map out our emerging domain, our language lags behind. We coin terms, negotiate them, and eventually it all settles down.

After all, as Stewart Brand said, “If you want to know where the future is being made, look for where language is being invented and lawyers are congregating.”