The Rise of CaaS: Context-as-a-Service for Agentic AI
AI Engineer World's Fair 2026 · 22:20
Web data infrastructure
Bright Data provides infrastructure for collecting and using public web data. Its platform combines proxy networks, prebuilt scrapers, structured datasets and APIs for unblocking websites, browser automation, crawling and search. Developers, data teams and AI teams use these services to build data applications and access current web information without maintaining their own scraping infrastructure. Bright Data MCP connects that infrastructure to AI agents, enabling web search, page extraction, structured data collection and remote browser interactions.
Founded in 2014 by Ofer Vilenski and Derry Shribman as Luminati, the company expanded from residential proxy infrastructure into broader data collection services. It adopted the Bright Data name in 2021 and is led by CEO Or Lenchner. Its residential proxy approach lets businesses view websites from different geographic and consumer perspectives, supporting uses such as price comparison and ad verification. EMK Capital acquired the business in 2017; the majority acquisition agreement specified a $200 million enterprise value.
As of August 2026, the company reported 50,000+ customers and more than 450 team members. In 2025, Bright Data reported annualized revenue exceeding $300 million and 429 billion cached web pages.
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here for Levi’s Web MCP demonstrations and discussion of CAPTCHAs, public-data access boundaries, and MCP tool-selection overhead.
Rafael LeviAI Engineer Europe 2026
Follow the implementation path: learn how Bright Data MCP, Claude Code, and the brightdata/skills repository support HTML inspection, selector extraction, and scraper creation.
Rafael LeviAI Engineer Europe 2026
Use Primor’s explicitly informal comparison of search, context vendors, and an in-house pipeline to frame the choice between buying and building context infrastructure.
Omer PrimorAI Engineer World's Fair 2026
Levi explains how blocked, empty, or stale web requests can appear to agents as successful searches, leading to fabricated citations and unreliable answers.
Levi contrasts repeated LLM page parsing with token-efficient scripts. His pipeline demonstration connects website inspection and scraper generation with scheduled monitoring and automated maintenance.
Primor argues that repeated retrieval increases rental costs and can strengthen the economic case for owned context pipelines. He also examines why point-in-time search limits longitudinal analysis.
Affiliations reflect each recorded session, not necessarily current employment.