AI Agents Cut 3-Day Portfolio Builds to Minutes — Jack McDonald, Apollo & Michael Manapat, Rowspace
Jack McDonald of Apollo's Redding Ridge and Michael Manapat of Rowspace discuss where AI becomes mission-critical in structured credit and CLO management. They cover compressing portfolio construction from days to minutes, automating the weekly trade reconciliation, and why embedding engineers with front-office teams, rather than relying on generic LLMs, is what earns trust in financial data.
Overview
Michael Manapat (co-founder of Rowspace) and Jack McDonald (Principal at Apollo and COO of Redding Ridge Capital, Apollo’s structured credit and CLO manager with roughly $50 billion in assets) frame the conversation around the gap between AI being a nice-to-have and AI becoming mission-critical. Rowspace positions itself as an agent-driven data engineering layer. Its agents pull fragmented data from many systems and reconcile it so portfolio managers and traders can decide faster and more accurately.
Why CLO Data Is So Hard
Jack describes the scale of the problem. Redding Ridge receives thousands of data points daily across its investments, tracks roughly 1,000-name loan pools, and assesses credits across several businesses. CLO management is active and trades in the broadly syndicated loan market, so having information at your fingertips matters a great deal. Unlike public equities or public fixed income, this information is private and scattered across documents. The real challenge is identifying the golden source of truth for each calculation, then keeping AI systems aligned with it. Rowspace’s approach was to sit with the teams, learn the actual calculations, and continuously improve the models from there.
Use Case One: Portfolio Construction
CLOs are highly rules-based, with precise waterfalls governing where money flows, all defined in dense indenture documents and legal agreements. Before Rowspace, portfolio managers reviewed those documents, built Excel models, and assembled mock portfolios of 200 to 400 single-credit names. That took two to three days. Now the system pulls real-time data from document libraries, trade and compliance systems, and live pricing, and produces a portfolio in minutes. Jack’s framing is useful: the tool gets the PM to 80 to 90 percent completion, and the PM’s judgment refines the final 10 percent to make the portfolio their own.
Use Case Two: Weekly Trade Reconciliation
Redding Ridge’s business depends on bank financing, and it buys loans in the fixed income market at any point. Historically, a weekly trade reconciliation summarized the week’s purchases and sales. It was highly manual. Staff literally went into bank accounts to check statements and settlement status, then emailed a recap to management. The same report can now be recreated in about a minute, and the team’s role has shifted to QA, checking the output each day. A process that once consumed many hours each week for two or three people has become routine.
Why Not Just Use a General LLM?
Jack says Claude, ChatGPT, and other LLMs are valuable for research, building new ideas, and getting up to speed quickly on a product or investment. But they work from general information, not the firm’s own data. The firm’s AI build-out has centered on owning its data and a platform that sits on top of it, validated by people who know the business. Jack notes that this is what distinguishes Rowspace from other technology vendors: its team is on the ground. Michael sits with the Redding Ridge team almost daily, learning the information flow.
The core argument is that one data point may come from twenty different sources, and only one is correct. Rowspace’s team works with portfolio management and finance to determine which source is right, overrides the others where needed, and aggregates the result. Inputs include the trade order management system, the performance system, Bloomberg, and various pricing indices. Before this, the same information was fragmented across the company and had to be pulled together by hand. Trust comes from having people validate the calculations and then relying on the system.
Apollo’s AI Posture
Michael observes that before code assistants like Claude Code matured, and when Cursor was still in the tab-completion era, Apollo was among the most AI-forward firms Rowspace had spoken with. Jack attributes this to Apollo’s long-standing identity as a leader in alternative asset management, a culture that is flat and open, and management that prioritizes upskilling and nimbleness. In his view, the firm recognized early that AI is here and must be adopted rather than dismissed. The aspiration was to rebuild the firm in an AI-native way, with a platform rather than a collection of point solutions.
Where Finance Sits on the Adoption Curve
Jack believes finance is still very early. Large banks and institutions have substantial middle and back office functions with plenty of room for improvement, and the complexity of these businesses makes adoption non-trivial. Model creation and research aggregation offer some value, but he considers that lower-value at this point. He expects that over time, certain functions will need less junior support. For now, the focus at Redding Ridge is making routine tasks efficient, which frees capacity for the many other projects the firm has.
Commodity or Durable Advantage?
Michael poses the central strategic question: will AI become a commodity like Excel, or will it structurally sharpen competitive advantage? Jack’s answer is that if AI is adopted and used properly, it will make firms more competitive. In asset management and investing, competitiveness comes down to process and the information at your fingertips, including liquidity, pricing, and valuation. That already matters in actively traded markets, and it is evolving in private markets, where Apollo has been working to lead. Firms with accurate, trusted information can avoid misinformed decisions, and that advantage compounds, flowing through to the enterprise and to investors.
What Has to Be True
When asked for a five-year prediction, Jack says it is too hard to forecast given the pace of change, noting that he cannot even predict whether his team will adopt a particular tool next week. His near-term hopes are concrete. He wants full confidence that information going to the investor base, which includes over 400 global investors, is accurate, without hours of manual checking. He also wants a more tech-enabled mindset across the team. Technology has been treated as an afterthought, and he wants it to become a primary way of thinking. Once routine work is automated, people have capacity to ask why a process exists, how it works, and whether it should change.
Advice for Builders
Jack’s guidance to engineers building AI for finance is to expect complexity. Structures that look strange or novel are often there for a reason. The most important investment is time spent on the ground with the front-office professionals you are building for. He credits Rowspace’s progress, including being ahead of its roadmap, to Michael’s near-daily presence with the team and his efforts to understand workflows. The discussion closes with a note that both Rowspace and Apollo are hiring people who want to apply AI to finance.