Benefit Street Partners (BSP), a credit-focused investment manager, uses OpenAI’s Codex to build tools for loan valuation comparisons and credit agreement analysis. Portfolio manager Anant Kumar says analysts can now answer an investment committee’s follow-up questions in a day instead of days or weeks, and often during the meeting. BSP uses the added review capacity to strengthen underwriting within the same timeframe, rather than to pursue more deals.

Kumar built a tracker from public SEC filings by business development companies (BDCs) to compare how different firms value the same loan. He used Codex to draft and review code. The first version took about 1 day; later features and interface refinements brought total development time to about 10–12 hours. Previously, building a shareable dashboard required help from someone with software expertise.

Kumar also built an analyzer for credit agreements, which can run 300 pages or longer. The tool draws on law firm memos about clauses and risks that may shift value away from lenders. Processing an uploaded agreement takes about 20 minutes for a first-pass analysis, followed by 1+ hour of human review. BSP refactored the tool with GPT-6 Astra, and it is live in the company-hosted portal. Kumar says AI helps collect and synthesize information, while a human makes the final investment decision.

Underwriting involves multiple rounds of further diligence and revised memos responding to committee questions. Deal timelines range from a few days to a few weeks, so the number of rounds varies. Kumar says access to more information has increased the team’s confidence in its decisions. BSP is working to connect information across deals; continuous underwriting, including monitoring loans as news happens and examining portfolio effects, remains a longer-term aspiration. Kumar says Codex has freed time for higher-level work and reduced routine tasks.

Benefit Street Partners (BSP), a credit-focused investment manager, uses OpenAI’s Codex to build tools for loan valuation comparisons and credit agreement analysis. Portfolio manager Anant Kumar says analysts can now answer an investment committee’s follow-up questions in a day instead of days or weeks, and often during the meeting. BSP uses the added review capacity to strengthen underwriting within the same timeframe, rather than to pursue more deals.

Kumar built a tracker from public SEC filings by business development companies (BDCs) to compare how different firms value the same loan. He used Codex to draft and review code. The first version took about 1 day; later features and interface refinements brought total development time to about 10–12 hours. Previously, building a shareable dashboard required help from someone with software expertise.

Kumar also built an analyzer for credit agreements, which can run 300 pages or longer. The tool draws on law firm memos about clauses and risks that may shift value away from lenders. Processing an uploaded agreement takes about 20 minutes for a first-pass analysis, followed by 1+ hour of human review. BSP refactored the tool with GPT-6 Astra, and it is live in the company-hosted portal. Kumar says AI helps collect and synthesize information, while a human makes the final investment decision.

Underwriting involves multiple rounds of further diligence and revised memos responding to committee questions. Deal timelines range from a few days to a few weeks, so the number of rounds varies. Kumar says access to more information has increased the team’s confidence in its decisions. BSP is working to connect information across deals; continuous underwriting, including monitoring loans as news happens and examining portfolio effects, remains a longer-term aspiration. Kumar says Codex has freed time for higher-level work and reduced routine tasks.