The Starting Gun of the Agent Era: OpenAI Fires Off Operator and Deep Research

OpenAI opened 2025 with two agent products: Operator launched January 23 as a research preview — taking over a browser to book tables and buy tickets autonomously — and Deep Research followed in February, autonomously conducting multi-step web research and producing expert-grade reports for finance, science and policy professionals.

Per TechCrunch and Business Insider, both debuted on the $200/month Pro tier before rolling down-market. Deep Research in particular set off an industry stampede: Google, Perplexity and xAI all shipped same-named or equivalent features within months.

In hindsight, these two launches were the starting gun of the agent era: the paradigm shift from answering questions to doing work began here.

Two Routes, Two Fates#

Operator and Deep Research embodied two agent forms whose fortunes diverged sharply. The Deep Research route became industry standard almost immediately — multi-step retrieval, cross-verification and long-report output were built into every flagship, turning into a daily tool for knowledge workers (see our Gemini Deep Research coverage). The Operator-style 'take over the browser and click' route proved far rockier: the reliability bar for general web operation ran well above expectations, and the form eventually evolved into ChatGPT Atlas — agent capability built into the browser itself (see our coverage). The industry lesson from the split: agent usability tracks how structured the environment is — search and documents are structured; web pages designed for humans are not.

Our Take#

Both products debuted on the $200/month Pro tier, and that pricing was itself the signal: OpenAI used a high price to select professional users willing to pay for 'work done,' validate agents' real commercial value, then roll down-market. The following year bore the judgment out: from tools-in-the-chain o3 (see our coverage) to the industry-wide agent-capability race, 'doing work' became the flagship battleground, and pricing split along the same line into professional and mass tiers. After the starting gun, the competitive axis moved from 'who ships first' to 'whose task completion rate is higher at lower unit cost.'