Summary
Omer Primor, head of product marketing at Bright Data, uses an AI Engineer talk to introduce what he frames as a new infrastructure category: Context-as-a-Service (CaaS). The argument is that as AI agents increasingly need to perform knowledge work — enriching data, researching companies, monitoring markets — the web becomes not just a data source but an ongoing context layer that must be refreshed continuously. Bright Data’s own analysis shows that social media data decays in hours, while even financial and retail data becomes stale within 30 days, making one-time or periodic extraction inadequate.
Primor traces the evolution of web access for AI from pure human search (Google-dominated for 20+ years) through LLM-integrated web search (ChatGPT, Claude) to a third wave of AI-native search companies like Exa, Perplexity, and Tavily built specifically for agent consumption rather than human users. CaaS sits alongside these AI search engines as an alternative path: instead of querying a search index, agents retrieve structured, pre-mapped context about specific entities directly.
The talk includes a live comparative test run 100 times against conference sponsors, measuring how well AI search versus CaaS providers can enrich a company profile across 25 fields using Claude Opus 4 as the orchestrating model. Coverage results converged across most providers, though two CaaS solutions underperformed in ways that initially surprised the team. Primor introduces the concept of “web context engineering” as a discipline for AI engineers — deciding which retrieval path to use for which type of task rather than defaulting uniformly to AI search.
📺 Source: AI Engineer · Published August 14, 2026
🏷️ Format: Deep Dive
