DevRev, a North American startup, has built two products, Computer and Operator, to carry customer and employee service requests through to completion. According to chief technology officer Ahmed Bashir, the agents reason across live data, actions can be reversed, and answers link back to their sources. The system combines OpenAI models, including GPT-6 Astra and GPT-6.1 Sol, with DevRev's own harness, which connects company knowledge and sets limits on what an agent is allowed to do. Bashir said DevRev judges OpenAI models on reasoning, reliability and latency at once. DevRev tests the models with its Enterprise-Bench benchmark, which covers data with multiple versions, information spread across systems, and access that depends on a user's role. Bashir said GPT-6.1 Sol performs well in those evaluations, though the article does not publish scores or pricing. Separately, DevRev has moved its engineering team fully to Codex and dropped its Cursor licenses, and its sales engineers use GPT-6 Astra to build demos for conference booths. The company's next goal is for Operator to handle longer, more demanding requests, such as finding a knowledge-base answer, upgrading a plan or processing a refund, while keeping actions traceable, permissioned and reversible. The article is a vendor-published customer profile, so its performance claims come from DevRev and OpenAI and have not been independently verified. For Australian businesses and professionals, it changes little directly, since it describes one company's product rather than a new feature or price change. Teams considering similar customer-service agents should weigh inference cost against response speed, and check how customer records are accessed and logged under the Privacy Act 1988 and the Australian Privacy Principles, particularly where an agent acts on a customer's behalf. The permission and reversibility controls described are the features worth testing in any pilot.