Context engineering means selecting and organising everything a model receives when it answers: instructions, documents, history, tool results, memory. Where prompt engineering focuses on how an instruction is worded, context engineering manages all the information that flows while an agent works over several steps. The notion took hold in 2025 with AI agents: Anthropic describes it as finding the smallest set of useful information that gives the best chance of getting the right result.
In practice at Gensai
It is at the heart of Gensai's projects: choosing the documents, tools and instructions the agent receives, so that it answers from the client's data rather than from its general knowledge.
What are the main techniques?
RAG, which retrieves only the useful passages from documents, summarising the history when a conversation grows long, notes the agent keeps outside its context window, and sub-agents that handle a task separately and return only the essentials.
Common mistake
Putting everything into the context because the window is large. The more loaded the context, the more likely the model is to get lost and the more each answer costs: better a little information, but the right information.
Same theme
Definition written with AI assistance, checked by Sébastien Lecocq, co-founder of Gensai.
Going beyond the definition?
From concept to project: Gensai builds custom, sovereign, GDPR-compliant AI solutions.
Related services: Chatbots and answer engines, Voice AI