“In 2025, AI agent products like OpenAI's Deep Research have shifted to a multi-step "chain of thought" process, where the model can access the web, reconsider its approach, and call tools multiple times in parallel during its reasoning process.”