DevRev says its Computer and Operator agents are designed to handle customer and employee requests by reasoning across live company data, taking approved actions and, when needed, preparing a person to act. The company says actions can be reversed and answers link back to their sources.
The agents use OpenAI models, including GPT-6 Astra and GPT-6.1 Sol, alongside DevRev software that connects company knowledge, manages permissions and controls what an agent can do. OpenAI models support reasoning, planning and tool selection, while DevRev says its systems are intended to balance reliability and response speed with the cost of serving users.
DevRev evaluates the models together with its software using Enterprise-Bench. The benchmark tests situations involving data stored across different systems, conflicting versions of information and access that varies according to a person’s role. DevRev CTO Ahmed Bashir says GPT-6.1 Sol performs well in these evaluations, where agents must understand requests and respond quickly at a sustainable cost.
In DevRev’s Agent Studio, users can describe how a tool should behave in a conversation with AI and set fixed rules for parts of its behavior. The company is also exploring ways for agents to handle decisions that are difficult to encode as rules, particularly when the information behind those decisions changes frequently.
DevRev says its engineers have adopted Codex across the company after trying several coding tools, and that it dropped its Cursor licenses. Its sales engineers also use GPT-6 Astra to build demonstrations for the company’s booths at international conferences.
The company’s next goal is to enable Operator to handle longer and more demanding requests while keeping actions traceable, permission-based and reversible. DevRev sees tasks such as finding a knowledge-base answer, upgrading a plan or arranging a refund as early examples of the self-service workflows it wants its agents to complete.
Sources: OpenAI News.