AI Engineering Trends Show Growth in Agent Demand
November 7, 2026 · 2 min read
As of October 28, 2026, there are 1,458 job listings on hh.ru that mention the term ‘AI agent’. The term LLM appears in 1,261 job postings, RAG in 520, and MCP in 245. In contrast, only 13 listings refer to ‘prompt caching’ and 19 to tokens.
The data reveals a growing demand for AI agents, yet there seems to be a lack of inquiries regarding the costs and operational mechanics behind these agents. The article emphasizes the need to understand key components such as tokens and budget management associated with agent tasks. It notes that using caching can make a task 6.6 times cheaper, highlighting the importance of structuring requests to optimize costs.
In the realm of AI engineering, professionals are shifting from traditional machine learning roles to those focused on managing pre-trained models like Claude or GPT. An AI engineer’s role involves framing questions and supplying the necessary data rather than training the model from scratch. This shift is supported by a Stack Overflow survey from 2026, which indicates that 73% of respondents are coding daily with the assistance of agents.
However, challenges remain, as reflected in a prior survey where 66% of participants reported that AI responses are often nearly correct but not fully accurate. Understanding the nuances of agent interactions, including the tokens processed per request and the rules governing interactions, is critical for effective deployment.
The complexities of AI engineering are further illustrated through discussions on elements like context windows and system prompts, which play a significant role in how models interpret and respond to requests. The article also compares the costs of tokens, noting that Russian text incurs higher costs than English text, and provides insights into optimizing prompt construction for cost efficiency.
The piece concludes by categorizing the various components that constitute an AI agent's operation, including the agent cycle and the harness that encompasses the model. It emphasizes the importance of understanding these facets to maximize effectiveness and manage expenses in AI tasks.