DevRev pushes self-service with GPT-6.1 Sol

Answering a support ticket usually means hunting across versions, systems and permissions before anyone can act.

DevRev says it built two agents to carry those requests through to completion, Computer to prepare the right human and Operator to act directly, in a Nov. 4, 2026 profile published by OpenAI News.

Speed has to hold.

DevRev combines OpenAI models including GPT-6.1 Sol and GPT-6 Astra with its own harness that connects company knowledge, enforces permissions and controls what an agent can do, the company said. The harness supplies context, the models handle reasoning, planning and tool selection, and every answer is meant to point back to its source. DevRev's chief technology officer Ahmed Bashir, who leads the technical approach to the agents, described the requirement as trust and latency together, saying “It’s a combination of being able to have the right reasoning capabilities, being able to be trusted, highly aligned, and most importantly, it has to work from a latency perspective.” He added that GPT-6.1 Sol “strikes the balance perfectly” between reasoning, speed and cost when a customer or employee is waiting.

The mechanism starts where a nontechnical owner meets the system. In Agent Studio, DevRev's builder interface, a user describes how a tool should behave in conversation with the AI and defines parts of that behavior with fixed rules. When a request arrives, Operator first tries to understand the issue, take an approved action or route to Computer if a person needs to act. Underneath, DevRev tests models together with the software that injects company context, using an internal evaluation called Enterprise-Bench that grades how agents perform when data has multiple versions, information lives in different systems and access depends on role. That is the hard enterprise condition most demos avoid.

Building and demonstrating the system runs on a different OpenAI surface. Ahmed said his engineers evaluated several coding tools and then standardized: “Our entire team uses Codex,” he said, adding the team dropped Cursor licenses after finding “Codex is far and away the best experience.” The company also uses GPT-6 Astra to generate sales demos for conference booths. For live service, Ahmed pointed to the Decisions API and Ultrafast, OpenAI primitives for handling quick decisions without full reasoning at every step, as a way to keep workflows to seconds, not minutes. He said some decisions are still difficult and expensive to encode as fixed rules when the underlying information keeps changing, so builders need to describe more logic conversationally and pick the right level of intelligence per task.

The bigger promise is still in progress. DevRev wants Operator to handle longer, more demanding requests while keeping actions traceable, permissioned and reversible, but frames familiar tasks like finding a knowledge base answer, upgrading a plan or issuing a refund as only the beginning. It has not disclosed customer counts, time saved or error rates, and the profile gives no independent benchmark scores. The space is crowded: ServiceNow already positions autonomous AI agents as central to transforming service management and empowering self-service, according to its site at https://www.servicenow.com/. Watch whether DevRev can prove sub second, auditable automation on messy, permissioned data without pushing cost or risk back to humans.