The days of the once-coveted prompt engineer role appear to be numbered. In July 2023, Anthropic made headlines with a $300,000 offer for prompt engineers, while entry-level roles fetched around $85,000 and senior positions averaged $200,000. But as AI systems evolve into agents powered by retrieval, memory, policies, application programming interfaces (APIs), and workflows, prompts are becoming just a cog in a larger machine.
Context engineering is already eclipsing prompt engineering, much like the fleeting role of chief AI officer. But what role will humans play as context engineering itself gets automated? Mint explains
Is prompt engineering already past its prime?
Prompt engineers acted as translators between humans and early AI models, turning natural language queries, or prompts, into structured instructions that produced reliable outputs. For instance, instead of one-off user queries, they could design a template that consistently summarises legal contracts with the right tone and accuracy.
Their value lay in blending strong writing and reasoning skills with some technical know-how, like APIs and frameworks such as LangChain. But as AI models matured and advanced in reasoning, prompting became faster, cheaper, and more standardised. Companies now prize system-level design that involves context, memory, retrieval, and workflows over pure prompt crafting, reducing the exclusivity that once commanded six-figure salaries.
Is it giving way to context engineering?
Context is the set of tokens an LLM processes when generating a response. In English, a token is roughly 75% of any word. A context window is an AI model’s short-term memory, or the number of tokens, or chunks of text it can process at once. It includes past messages, responses, and the current query. Once the limit is reached, older tokens drop off, which can reduce accuracy.
Larger context windows let models handle more information and reasoning at once, but context engineering makes the difference between generic outputs and precise, relevant answers in complex tasks by deciding what goes into that memory. Thus, the focus is moving from wordsmithing to managing tokens and context. For instance, a travel-booking agent might automatically pull past preferences, loyalty numbers, and budget limits into its reasoning without the user retyping them each time.
So, what happens to prompt engineers?
Prompt engineering will not vanish for now. It will survive as a subset of broader AI systems’ work. However, as AI agents evolve, prompts will no longer be crafted manually for each task but generated dynamically through context pipelines. The field is already shifting toward “agent engineering” or “AI systems design”. In these roles, prompts remain crucial building blocks but are embedded within larger contexts of retrieval, memory, workflows, and safety layers.
This shift will reduce the need for specialized staff who only refine prompts. Just as HTML coders gave way to full-stack web developers, context engineering will absorb prompt engineering, leaving little room for prompt-only specialists as enterprises demand end-to-end system expertise.
What are the dangers of automating such roles?
While automation improves efficiency, removing skilled engineers entirely could weaken safeguards and creativity, both of which remain critical to ensuring reliable and ethical deployment of AI agents. Automating prompt and context design carries risks of over-reliance on templates and frameworks without human oversight.
Poorly-designed context pipelines may propagate bias, hallucinations (information conjured up confidently), or unsafe outputs at scale. For example, if a retrieval system feeds outdated medical information into a healthcare chatbot, the errors could be amplified across thousands of patients.
With context engineering also being automated, what’s next?
Context engineering is increasingly being automated by AI agents that handle memory, retrieval, tools, and workflows. Instead of relying on humans to manually shape context, agents now summarise transcripts, prioritise key facts, and discard noise on their own, like Microsoft’s Copilot, which compiles the most relevant meeting notes into a project brief. As these systems evolve, they will unify text, images, audio, video, and structured data into richer context frameworks.
This shift opens the door for humans to act as AI systems architects who define objectives, devise policies, and ensure ethics and governing behaviour, while leaving the cognitive micromanagement to the machines.
