According to a case study published by OpenAI on October 16, 2026, Benefit Street Partners (BSP), a credit-focused investment manager, is using OpenAI's Codex to build internal analysis tools. Portfolio manager Anant Kumar says answering an investment committee's follow-up questions used to take days or weeks and can now often be done within a day, sometimes during the meeting itself. BSP says it is using the tools to add review capacity and rigor to underwriting rather than to do more deals. Kumar built a tracker that uses public SEC filings from business development companies (BDCs) to compare how firms value the same loan, using Codex for code review. The first version took a day, with total development time estimated at 10 to 12 hours. A second tool analyzes credit agreements, often 300 pages or longer, drawing on memos from law firms about clauses and risks. An uploaded agreement takes about 20 minutes to produce a first-pass analysis, followed by an hour or more of human review. BSP has rebuilt the tool using GPT-6 Astra, now live in a company-hosted portal. "The ultimate investment decision needs to be made by a human," Kumar says. The account comes from OpenAI, the tool's vendor, with Kumar as the primary source.
- The BDC mark tracker took about 10-12 hours total to build, using Codex for code review
- The credit-agreement analyzer produces a first-pass analysis in about 20 minutes, followed by 1+ hour of human review
- BSP says it uses the tools to add rigor to underwriting, not to do more deals; humans make the final investment decision
What it means for you
A specialist at a finance firm built useful internal tools himself, in hours rather than hiring developers, because AI could draft and check code for a non-programmer. The quieter lesson is the discipline around it: AI collates and summarizes, but a person still spends an hour or more reviewing every credit-agreement output, and no investment decision is handed to the tool. That's the pattern worth copying, not the specific software.
Try this
Pick one recurring task in your operation where someone manually pulls data from several sources into a report or comparison. Map how that work flows today and note where a human judgment call actually sits — before you imagine automating any of it.
Who should care
Owners and operations leads at small firms who have in-house experts doing manual analysis or reporting but no software team; anyone responsible for underwriting, diligence, or document-heavy review work.
Skip this if
You don't have a repeatable, document- or data-heavy analysis task, or your existing process isn't yet clear enough to describe on one page. A vendor case study isn't a reason to act — fix the process first.
Sources: OpenAI — read the original