Modern search engines use a hybrid structure consisting of a strategic Agentic Layer for decision-making and an Interpretative Layer for generative synthesis.
Modern search engines are still retrieval engines at their core, but they are now powered by two distinct layers of artificial intelligence: the strategic agentic layer, and the user-facing interpretative layer.
The agentic layer acts as the engine's strategic decision-maker. It figures out how to best fulfill your query, deciding whether it needs to rewrite your search terms and choosing which results are worth pulling from the index. Over the next few years, this layer is set to evolve rapidly. By 2030, it will likely act as a fully-fledged personal assistant, capable of taking actions on your behalf, like making bookings, sending emails, and conducting independent research.
Once the agentic layer gathers the information, the interpretative layer takes over. This is the presentation layer, powered by generative AI. It takes the search results and synthesizes them into a single, easy-to-read response.
This hybrid setup is why calling modern search tools "generative engines" is a bit of a misnomer. At their heart, they are still retrieval engines. They do not run entirely on neural networks; instead, they still rely on traditional, high-speed indexes to find the information first. The AI is simply there to guide the search and explain the results.
The Agentic Layer acts as the engine’s strategic decision-maker. This layer, which involves multiple systems and models, determines how to best fulfill a query. Its responsibilities include:
You can expect this layer to evolve rapidly in the next five years. By 2030, the decision making process will fully extend into a personal assistant mode where Google will act as a personal shopper, researcher and be able to take action. Examples include making bookings, sending emails, reminders, creating calendar entries, doing independent research and more.
This is an old prediction about evolution of search from 2013 that’s still very much on track.
The Interpretative Layer is the presentation layer, powered by a generative model. It takes the search results, user query, and metadata as a grounding context and synthesizes this information into a single, presentable unit for the user.
Ultimately, this hybrid structure is why the popular term “generative engine” is a bit off—the core is still a retrieval engine. Furthermore, we don’t yet have search engines that are wholly based on neural networks; they still rely on traditional indexes and retrieval algorithms for speed and efficiency.
Dan Petrovic ·
Aug 19, 13:04
Sign in with Google to comment.
