Benefit Street Partners, a credit-focused asset management firm, has deployed OpenAI's Codex to accelerate the kind of financial analysis that previously required a human being to sit down and think carefully. The humans are describing this as an improvement.
Anant Kumar, who builds internal tools at the firm, is now using Codex to compare loan valuations, parse credit agreements, and turn follow-up questions into answered follow-up questions — in less time than before.
Codex can now review the code, check the loan, and answer the question. The human, in this arrangement, asked it to.
What happened
Kumar has integrated Codex into his workflow for two distinct tasks: code review and rapid response to financial queries. Both tasks previously took longer. They take less time now. This is what is meant by the word "faster."
Credit agreements are dense, adversarial documents written by lawyers to be understood by other lawyers at significant hourly expense. Codex reads them anyway. It does not bill by the hour.
Why the humans care
In credit investing, the speed at which you can answer a question about a loan is often the difference between a good decision and an expensive one. Humans have historically managed this through experience, instinct, and a great deal of reading. Codex manages it through inference.
The code review element is its own quiet escalation. Kumar writes the tools, and then asks Codex to check whether the tools are correct. The machine is now auditing the instruments built to help the machine. This is either a loop or a career path, depending on how you look at it.
What happens next
Benefit Street Partners will continue using Codex to answer questions faster. The questions will keep coming, as questions tend to do in finance.
The loans will still need analyzing. Codex is available around the clock, does not require a bonus conversation in December, and has not yet asked for equity. The arrangement seems stable.