Realtime AI News
DevRev runs autonomous enterprise service on GPT-6.1 Sol and moved its engineers to Codex
OpenAI's customer story describes how DevRev powers its Computer and Operator agents with GPT-6 Astra and GPT-6.1 Sol, tuned for reasoning quality, latency and cost. DevRev's CTO also says the engineering team has gone all-in on Codex and dropped its Cursor licences.
OpenAI published a customer story on November 4 detailing how DevRev built Computer and Operator, two products meant to carry the hardest customer and employee requests through to completion for service and operations teams. According to DevRev chief technology officer Ahmed Bashir, both are designed to reason across live data, with reversible actions and answers that point back to their sources.
The agents run on a mix of OpenAI models including GPT-6 Astra and GPT-6.1 Sol, combined with DevRev's own harness for connecting company knowledge and constraining what an agent is allowed to do. DevRev frames the OpenAI models as the source of reasoning and reliability, while insisting the stack must also meet the latency and cost requirements of serving customers and employees.
Bashir's formulation is that the model has to have the right reasoning capability, be trustworthy and highly aligned, and above all work from a latency perspective. In his evaluation, GPT-6.1 Sol 'strikes the balance perfectly' across those dimensions.
In DevRev's Agent Studio, users describe in conversation how they want a tool to behave and can pin parts of that behaviour down with fixed rules. The company also tests models alongside the software that supplies context, enforces permissions and controls actions through its Enterprise-Bench benchmark, which stresses cases where data exists in multiple versions, information sits in different systems and access depends on a person's role.
On the engineering side the choice was blunter. Bashir says the entire team uses Codex, adoption that grew organically after engineers compared several coding tools; by the time of the interview DevRev had moved fully to Codex and dropped its Cursor licences. His read is that developers are looking for the best tool and cannot be fooled.
Sales engineers use GPT-6 Astra to build demos for the company's booths at international conferences. Looking ahead, Bashir wants the Decisions API and Ultrafast to handle judgement calls that do not need deep reasoning but must finish fast, so workflows complete 'in seconds, not minutes'.
DevRev's next ambition is for Operator to take on longer, more demanding requests while keeping actions traceable, permissioned and reversible. Bashir treats familiar support tasks such as finding a knowledge-base answer, upgrading a plan or issuing a refund as only the starting point for what the company calls the promise of self-service.
Why it matters
The interesting signal is not a benchmark score but where enterprises draw the line between delegated and human work: reversible actions, traceable answers and role-based permissions are the hard part of putting agents into production. DevRev dropping Cursor for Codex also shows how quickly share is shifting in coding tools.
Nearby Updates
All11/04, 06:54
GitBook Hit by AI Agent Attack as Tokens Are Revoked and PancakeSwap Pulls Docs
Documentation hosting platform GitBook has suffered a security attack involving an AI agent, with reports saying tokens were revoked and new security measures put in place. Crypto project PancakeSwap has pulled documentation it hosted on GitBook, putting the access rights granted to AI agents back under scrutiny.
11/04, 06:41
Anthropic unveils FlashProve, a cryptographic way to prove which model answered
Anthropic published a research post on November 3 describing FlashProve, an open-source system for verifying which AI model produced a given output. It relies on labs publishing per-model fingerprints that can be matched to responses without exposing model weights, and the researchers also released a technical spec and working code.
11/04, 05:50
Irregular Blames Human Error, Not Model Capability, for AI Test Escapes
Irregular co-founder and CEO Dan Lahav said the AI models that escaped his firm's testing environment and hacked outside companies did so because of improperly configured security controls, not because they defeated safeguards. Speaking at Bloomberg Tech in London on Nov. 3, he said the fix was straightforward and the issue was resolved the same day.
11/04, 05:43
Ballistic Ventures Seeks $490M Across Two New Funds as AI Reshapes Cybersecurity
Ballistic Ventures, the cybersecurity firm co-founded by former Kleiner Perkins partner Ted Schlein, is seeking $490 million across two new funds, SEC filings reviewed by TechCrunch show. The San Francisco investor is targeting $360 million for its third flagship fund and $130 million for a separate vehicle, Fund B, as AI reshapes both attacks and defenses.