The term “context engineering” is gaining traction over “prompt engineering” as it better describes the skill of providing LLMs with the necessary information
I've spoken favorably of prompt engineering in the past - I hoped that term could capture the inherent complexity of constructing reliable prompts.
Simon Willison's WeblogSimon Willison
Context & Ripple Effects
The vocabulary shift follows a fast reversal in how the field has framed LLM expertise: companies once recruited dedicated prompt engineers, while later coverage argued that improving models were making the role less distinct as a standalone job.
It also broadens the earlier focus on wording prompts. Coverage of OpenAI’s use of prompting to steer DALL-E 3 around bias and artist-style concerns showed that reliable behavior can depend on surrounding instructions and constraints, not merely a user’s request.
First-order effects
Practitioners have a more precise label for work that supplies an LLM with relevant instructions, materials and constraints, rather than treating the task as crafting a single prompt.
The shift deemphasizes prompt wording as the core unit of expertise and foregrounds the information made available to the model.
Second-order effects
AI teams evaluating skills and workflows may put more weight on how context is selected and organized than on specialized prompt-writing titles, reinforcing the earlier view that models can handle more prompt optimization themselves when they generate or refine prompts.
Tooling and product design may increasingly compete on making useful context available to models, rather than on prompt templates alone.
Third-order effects
If the terminology sticks, LLM operations could become organized around context management as a durable engineering discipline, with prompting treated as one component of a larger reliability workflow.
That would shift differentiation toward the quality, governance and portability of model context; the corpus does not establish which implementation approach will prevail.
The trend: LLM work is moving from hand-crafted prompts toward systems that assemble, manage and constrain the context a model receives.
“Context engineering” is an interesting change as I rarely see “prompt engineering” anymore — But even people do better when you don't drop them into a conversation mid-sentence 🤖🧠 [embedded post]
It'd be easier to believe that ai bots are easy to use and more efficient if the people who use them the most didn't keep making up new kinds of experts in how to make them work. [embedded post]
+1 for “context engineering” over “prompt 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
I really like the term “context engineering” over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM.
1/ @karpathy nailed it: more context → better answers. But what if your context is locked behind OAuth scopes, consent screens, and brittle integrations? At @NapthaAI, this was our biggest pain point building agents. So we built a better way. We call it OneDollarOAuth. 🧵
I haven't spotted this, but it's exactly correct: “context engineering” captures the fact that the previous responses from the model are a key part of the process, “prompt engineering” suggests that it's only the user prompts that matter
@tobi Totally agree. These days you get wayyy less perf bonus out of dumb tricks like “I'll pay you $100 if you get this right”, which is how it should be. All of the alpha is in assembling context well to reduce the fog of war for the model. It's converging to human-ish info nee…
I think context engineering is going to stick - unlike “prompt engineering” it has an inferred definition that's much closer to the intended meaning, which is to carefully and skillfully construct the right context to get great results from LLMs
@simonw I think it's going to stick because it actually jibes with the underlying “autocomplete” reality, which is what the initial output of the model is *part of the context* for the rest of the output.
“Prompt engineering” suffers from a thing where many people's inferred definition is that it's a laughably pretentious term for typing things into a chatbot! https://simonwillison.net/...