DevRev deploys GPT-6.1 Sol for autonomous customer-service agents
DevRev's Computer and Operator agents use OpenAI's latest models to handle customer and employee requests with reversible actions and source attribution.
Last verified:
DevRev’s Autonomous Service Agents Reach Production
DevRev has deployed Computer and Operator, two autonomous agent systems designed to resolve customer and employee requests without human intermediation. According to the OpenAI Blog, both agents leverage GPT-6 Astra for foundational reasoning and GPT-6.1 Sol for enterprise decision-making, integrated with DevRev’s proprietary harness for connecting company knowledge and enforcing access controls. The agents are built around a core principle: actions are reversible, and all answers trace back to their source data.
The Enterprise-Grade Reliability Problem
DevRev addresses a critical gap in autonomous AI deployment—the gap between lab performance and production reality. Most benchmarks assume clean, centralized data and uniform access patterns. DevRev’s Enterprise-Bench evaluation framework tests models against messier conditions: multi-version data, information scattered across legacy systems, and role-based access that varies per employee. According to DevRev CTO Ahmed Bashir, GPT-6.1 Sol performs well under these constraints, balancing reasoning depth with sub-second latency and sustainable cost-per-request economics.
This framing rejects the notion that “better reasoning” alone solves enterprise deployment. Instead, it foregrounds three constraints that often determine whether an autonomous system ships: trustworthiness (alignment with company policy), speed (sub-second response for interactive workflows), and cost (sustainable per-request inference).
Agent Studio: Hybrid Rule + Learning Configuration
DevRev’s Agent Studio allows teams to configure agent behavior through a mixture of natural-language instruction and explicit rules. Users describe desired tool behavior in conversation with AI, then define fixed constraints—guardrails that the agent cannot violate even if reasoning suggests it should. This hybrid approach avoids the brittleness of pure rule-based systems while retaining the predictability enterprises require for compliance-sensitive workflows.
The platform enforces role-based permissions, meaning an agent responding to a junior employee’s request sees only the data that employee is authorized to access. This design choice prevents agents from leaking sensitive information through careless reasoning.
Codex Adoption and the Developer Choice Signal
DevRev’s engineering team has consolidated entirely around Codex, OpenAI’s code generation model, after organically exploring competing tools like Cursor. DevRev CTO Bashir attributed this shift to developer agency: “Developers are really just looking for the best tool, and you can’t fool them.” The anecdote signals a maturing market for AI coding tools, where adoption flows to the tool perceived as highest-quality rather than to the best-marketed option.
Sales engineers additionally use GPT-6 Astra to generate conference-booth demonstrations, suggesting enterprise AI’s move from proof-of-concept to standardized go-to-market practice.
Why This Matters
Enterprise self-service automation has historically failed because models either reason too slowly to replace humans or make irreversible mistakes that corrupt data. DevRev’s emphasis on reversibility and role-based access control narrows this failure surface. Teams deploying agents in customer-service and employee-request workflows—finance, HR, IT support—now have a template for safe autonomous systems that learn from company-specific context rather than generic fine-tuning. If Enterprise-Bench scores hold up under independent audit, this model architecture could become a standard for regulated industries where agents must prove they respect data boundaries and undo their errors.
Frequently Asked Questions
What are Computer and Operator?
Computer and Operator are autonomous AI agents built by DevRev that handle customer and employee service requests. They reason across live company data, take reversible actions, and cite the sources behind their answers.
Why does DevRev's approach emphasize reversible actions?
Reversible actions allow the system to undo mistakes without manual intervention, reducing the risk of autonomous agents corrupting data or creating inconsistent states in customer and employee records.
How does DevRev control what actions agents can take?
DevRev's Agent Studio lets users define tool behavior through conversation with AI and fixed rules. The platform enforces role-based permissions and company-context integration to restrict agent actions to approved workflows.
What is Enterprise-Bench?
Enterprise-Bench is DevRev's internal evaluation framework that tests how well AI models perform when company data exists in multiple versions, is scattered across systems, and access depends on user role.