Capital Allocators · Sep 14, 2026
With Abby Barlow, CIO of Westwood Management · Laura Hill, CIO of Advocate Health · Brian Sugrue, CIO of Shannonbridge · Jenny Heller, President and CIO of Brandywine Group Advisors · John Lawrence, CIO of Rice University · Matt Bank, CIO of GEM · Kristin Kallergis Rowland, Global Head of Alternative Investments for JPMorgan Asset and Wealth Management · Jon Webster, SMD and COO of Technology and Operations at CPP Investments
Ted Seides asked eight chief investment officers the same question, how are you actually using AI, and got back eight completely different answers, running from a single family office CIO who has never hired an analyst because AI does that job, to the $580 billion Canadian pension giant trying to make AI "read everything, remember everything, and challenge everything" across an entire institution.
What ties them together is where each one draws the line. Every single CIO, regardless of scale, said the same thing in different words: AI clears the routine work so there's more room for judgment, and none of them will let it make the actual investment decision.
"I joke with my team here in the family that my analyst's name is Claude and he works all hours of the night and he's amazing."
Capital Allocators, hosted by Ted Seides, covers the institutional investment industry; this episode (EP.515) compiled short conversations with eight CIOs across single family offices, a healthcare system's investment office, a university endowment, an OCIO and a $580B pension fund.
The full episode is covered here so you can skip it.
Here are the 8 takeaways that matter.
Key Takeaways
Abby Barlow, sole investment professional at a single family office, has delayed hiring specifically because AI does the analyst's job — she vibe-coded her own benchmarking tool from a half-finished spreadsheet in about half a day
Advocate Health's $26B private markets team built General Gist, an internal AI that ingests capital calls and statements and elevates analysts "from data entry to actually reviewer status"
Brian Sugrue at Shannonbridge did in a couple of days what would have taken two analysts two or three weeks — a global cost-of-borrowing analysis across asset classes and currencies — but flagged that AI "produces sort of garbage" outside a reviewer's own expertise
Jenny Heller's team tried a mid-diligence AI update and it failed — the context was too big and the output too unfocused, so the team went back to reviewing the Slack channel directly instead
Rice's John Lawrence used an AI-cloned voice of himself to narrate a board podcast, and separately said the firm's real bottleneck is building a centralized data lakehouse, not the AI layer on top of it
Matt Bank's team at GEM flipped from 90% externally-bought tools to 90% internally built in a year, with three in-house developers now generating about 30,000 lines of code a month
JPMorgan built its own internal LLM suite so client data never leaves the firm, and Kristin Kallergis Rowland's team now runs AI red team agents to challenge its own investment theses before they go to committee
CPP Investments' Jon Webster reduced a multi-day portfolio-wide forensic screen to under an hour, and says the single test for any use of AI there is simple: does it make CPPIB a better investor
1. Abby's Analyst, Claude
Barlow is the sole investment professional at a single-family office and has not hired since joining three years ago — in her telling, largely because AI arrived at the same moment she did.
An analyst on the org chart
"I think I'm going to put him on our org chart actually." — Abby Barlow
She uses Claude across the entire process, from summarizing a fund manager's first deck to reviewing legal redlines: "There were 185 tracked changes in this document... Now I can just take that, put it in Claude, and in one minute get back a very good summary of what's relevant... it is very good. like better than what I could do frankly on my own." Her benchmarking tool, a vintage-year comparison across funds that used to sprawl across 60 Excel tabs, was vibe-coded from a half-finished spreadsheet: "I literally took the spreadsheet halfway done... and in 5 minutes it thought about that and it spit out essentially like an app." It took about half a day to finish, and she now presents screen grabs of it directly to her investment committee. She also described asking Claude to interview her for one to two hours to build a standing context document about her role, the family and the organization's history, reused on every new project so she doesn't re-explain herself each time. On limits: she still doesn't feed account numbers or dollar figures into any model, using family-member code names instead, and said her biggest surprise was a double-counted commitment in a pacing model that she now catches by asking Claude to re-review its own work for errors at the end of every task.
