Anmol Goel, CEO and Managing Partner of GACS Family Office, joins The Agentic Allocator for a practical conversation about AI adoption inside a family office. GACS invests over $1 billion across venture, private equity and private credit from London and India, and Anmol runs it in public, speaking at conferences and building a personal profile in a historically discrete community.

Anmol walks through how GACS is using AI within the investment office: a tech due diligence tool trained on the firm’s own past investment memos that has taken analysis from two months down to 2-3 weeks, an internal database that cross compares deals and surfaces unique portfolio connections, and a personal knowledge base built from roughly 10,000 contacts, past conversations and notes that he queries whenever a founder needs an introduction.

He is equally direct about the challenges and what has gone wrong. Team members have put sensitive documents into public tools. Work has gone out without being checked. He is clear that they are still working through where the AI privacy line sits, and that the answer will be different for every family. For an industry where most AI conversations only cover the wins, it is an important reality check. 

What You'll Learn:

• How he got his family comfortable with a public profile in a historically private industry
• How GACS built a tech due diligence tool on its own past investment memos, and why analysis went from two months to 2-3 weeks
• What happened when a new pitch deck went into the system and it surfaced a link to an existing portfolio company nobody had spotted
• How a personal knowledge base of roughly 10,000 contacts answers questions like which fintech regulators do I know
• The two adoption challenges he has lived through
• Why he thinks an extra 30 to 60 seconds on a prompt removes most of the hallucination problem
• Why he worries the next generation will lose the knowledge of knowhow, and what first principles have to do with AI
• What AI adoption looks like across the 30 to 35 family offices he co invests with, and why the split is generational rather than by size
• His starting point for families: repetitive tasks, a cost test, and ignoring vendors who use big numbers to create urgency

About Anmol Goel:

Anmol Goel is CEO and Managing Partner of GACS Family Office, a family office operating across the UK and India with over $1 billion in assets under management, investing across venture, private equity and private credit. He founded his first company at 17 and exited his second at 20 before moving into investing. He holds a degree in mathematics and economics and began his career at JP Morgan and Marsh McLennan. He has backed more than 80 companies and funds, co-invests alongside a syndicate of over 30 family offices, and has beta tested AI models ahead of public release.

Episode Highlights:

[01:42] Building in Public
Family offices have historically operated quietly. Anmol’s view is that a public profile was the fastest route to a network, because if you know something worth sharing, people will take the call. His grandparents’ investment knowledge was land, real estate and gold, so what he was doing looked alien to his family at first. What settled it was a stack of proof. 

[05:35] AI Inside the Family Office

Most significantly, GACS built its own tech due diligence tool, trained on past investment memos and structured around what the firm looks for in a founder, a market and technology. It is not perfect and human intervention is still required, but analysis that used to take two months now takes two or three weeks.

[06:43] When the System Spots What the Team Missed

As the database grows, GACS can cross compare deals and identify smaller M&A opportunities across the portfolio. Anmol describes putting a pitch deck for a company the firm was about to back into the system, which immediately flagged how it linked to an existing portfolio company. The team had not made that connection themselves.

[07:14] Ten Thousand Contacts and One Searchable Base

With close to 10,000 contacts on his phone, Anmol is compiling everything into a single base: contacts, past videos, conversations, and notes. When a fintech founder comes to him, he can ask which fintech regulators, policymakers, operators, founders and investors he already knows, and get back a list with context on how each one could help. For a firm that invests hands on, that is the difference between remembering a useful contact and not.

[09:20] Where It Has Gone Wrong: Privacy and Unchecked Output

Two challenges stand out. The first is compliance and privacy. It can be hard to know with confidence what is being stored, and members of the team have put highly sensitive documents into public systems. The second is cutting corners. Output passed on without proper checking has been wrong highlighting the need for a human-in-the-loop. 

[10:32] Prompt Engineering and the Two Week to Two Hour Shift

A new hire in a team meeting that morning had been hesitant about using AI at all. Anmol’s response was that work which used to take two weeks now takes two hours, and the skill that closes the gap is prompting. Being precise about what you want, and spending an extra 30 to 60 seconds specifying it, removes a large share of the hallucination risk.

[11:22] First Principles and the Knowledge of Knowhow

Anmol has a background in mathematics, taught to derive an equation rather than just use it. His concern is that the models will keep improving while the underlying knowledge of how something works quietly disappears from the next generation. He considers himself lucky to have been on the cusp, having learned it just before it became optional.

[12:48] The Adoption Split Across Family Offices

Across the 30 to 35 family offices GACS co invests with, the divide is generational rather than by size. Families rooted in agriculture, manufacturing or real estate, with principals in their fifties and the next generation not yet in seat, are still some ways behind. Founders who exited in their early thirties and now run a few hundred million are often far ahead, to the point where Anmol asks them how they built it.

[14:15] Where to Start, and What to Ignore

Start with the most repetitive tasks and see whether AI can take them. Run a cost analysis and check that one plus one makes two, because families understand numbers. Be skeptical of vendors who use big terminology and big numbers on an audience they know has a knowledge gap. Start small, understand what is actually making it work, and do not buy on FOMO.

[15:23] People Analysis Over Market Analysis

Anmol argues qualitative analysis matters more now. His own work has shifted from technical, financial and market analysis towards people analysis: how hungry the founder is, how driven, what the why is. That is the part he does not believe AI will pick up, so he lets the models handle the rest and spends his time travelling to meet people.

[16:22] Why He Meets Everyone Before Investing

Of roughly 80 funds and startups backed over the past two to three years, Anmol cannot think of one he invested in without meeting them first, even where that meant waiting a month or two. He accepts that this costs him deals. The trade is that the ones he does write are...

Dr. Ashby Monk, Executive and Research Director of the Stanford Research Initiative on Long Term Investing, joins The Agentic Allocator to explain why AI has become an existential question for pension funds and sovereign funds, and what a credible answer to the Board looks like.

Ashby has spent over two decades studying and advising the world’s largest asset owners on governance, organisational design, technology, and investment strategy. He is also Managing Partner of KDX, a venture capital firm backing founders building technology for institutional investors, which gives him a unique vantage point across the ecosystem. 

His starting position is that the core LP problem predates AI entirely: portfolio complexity has far outstripped the organisational capability to manage it. Institutions designed around 60/40 are now past 50 percent in alternatives, without the technology, process, or governance to handle that opacity. AI is the first thing Ashby has seen motivate asset owners to confront this challenge.

The conversation covers his on ramp for allocators, why investment decisions belong at the end of that sequence rather than the beginning, why handing a new tool to the person whose job it does will fail 99 times out of 100, and how the implementation gear of the industry could shift over the next ten years. 

What You’ll Learn:

• Why the fundamental LP pain point predates AI: portfolio complexity has outrun the capability to manage it
• Why AI is an existential question for pension funds rather than a market crisis, and what boards are now demanding from CEOs and CIOs
• Why wave one AI adoption is individual productivity, and why the real value sits in mobilising the organisation’s collective knowledge
• Ashby’s four step on ramp for allocators, and why the investment decision is the expert level that comes last
• What generalist venture capitalists misunderstand about LPs, and why they see them as a source of funds rather than a user of technology
• Why an industry managing more than $140 trillion has almost no specialist technology investors backing it
• The gearbox model of an investment organisation, and which gears AI is most likely to transform first
• Why asset allocation may be optimised to actual cash outflows within ten years rather than to a return target
• The ten-year view on implementation: SMAs at your own custodian, managers selling signals and custom indices, and a model that looks closer to technology outsourcing
• Why sending a new tool to the person whose job it replaces produces a negative answer 99 times out of 100, and what to do instead
• Why governance should function as a steering wheel rather than a brake
• What Ashby tells younger professionals whose seniors are hesitating around AI adoption, and why the pitch should be about capability rather than tools

About Dr. Ashby Monk:

Dr. Ashby Monk is Executive and Research Director of the Stanford Research Initiative on Long Term Investing. He has more than twenty years of experience studying and advising the world’s largest pension funds and sovereign wealth funds on governance, organisational design, technology, and investment strategy, has authored seven books, and has published hundreds of research papers on institutional investing. His book The Technologized Investor won the 2021 Silver Medal from the Axiom Business Book Awards.

Outside academia, Ashby is the Managing Partner of KDX, a venture capital firm focused on investment technology, and has helped build a number of companies applying advanced analytics to capital allocation, including RCI Navigator, acquired by Addepar, and Long Game Savings, acquired by Truist. He is a member of the CFA Institute Future of Finance Advisory Council and was named by CIO Magazine as one of the most influential academics in institutional investing. He holds a PhD in economic geography from the University of Oxford.

Episode Highlights:

[01:55] The Pain Point That Predates AI

Pension funds and sovereign funds were built around 60/40. Many are now past 50 percent in alternatives, and the word ‘alternative’ has become a conventional form of investment. The complexity of the portfolio has far outstripped the capability to manage it. What is missing is an organisation that can handle that complexity and opacity through technology, process, and governance.

[04:05] Why AI Is an Existential Question, Not a Market Crisis

Climate motivated asset owners. So did 2008 and 2001. AI is different. It is not a crisis in the markets, it is a question about whether the institution remains the one overseeing the corpus or ultimately relies on new forms of intelligence to manage it. Boards are asking directly, and leaders without an answer on organisational readiness are in trouble.

[06:55] Wave One Is Productivity. The Power Is Collective Knowledge.

Most organisations still have no AI strategy. What they have is individuals synthesising documents and formatting meeting notes. That is useful, but those individuals are tapping knowledge from outside the organisation. The firms doing this well are using AI to mobilise their own history, capabilities, and belief systems for the benefit of whoever opens the tool.

[08:40] The On Ramp: Everything That Comes Before the Investment Decision

Start by using AI to understand what the organisation already thinks and who is working on what. Then underwrite past deals again, which takes thick skin, because AI is good at pointing out what was missed. Then run red team analysis on live deals. Then move to value creation inside the portfolio, where the stakes are lower. Investment decisions are the expert level and come fourth, not first.

[10:30] The Silo Problem Across Client, Manager Selection, and Operations Teams

Allocator organisations typically run a client investment team, a manager research team, and an operations team on three systems that have never spoken to each other. Sharing collective knowledge across those functions has been structurally difficult. It is one of the clearest near-term applications of the technology.

[11:35] What Generalist VCs Get Wrong About LPs

Generalist venture investors are strong technologists, but they see LPs as a source of funds rather than a user of technology. Their service model is oriented to founders. One well known GP limits LP contact to once a year. That framing is why so little capital has gone into the technology these institutions actually need.

[13:55] $140 Trillion and the Case for Invest Tech

Asset owners may be bureaucratic and sit inside governments and universities, but they manage north of $140 trillion, and the work itself is processing data into information, information into knowledge, and knowledge into applied intelligence. Every stage of that is affected by technology. Defence tech supports 50 to 100 specialist funds on comparable technology budgets. Investment technology supports almost none.

[16:30] The Gearbox: How to Model an Investment Organisation

Every investor runs a set of interlinking gears. The organisation is governance, capital base and its preferences, culture, technology stack, and people. That gear drives a target asset allocation. That drives an implementation model. The outer gear is the global market. Ashby expects allocation and implementation to be transformed first.

[17:50] Alloca...

Ryan Kulig, Finance and Operations Officer at SITFO, the Utah School and Institutional Trust Funds Office, joins The Agentic Allocator to share how a $4.7 billion-dollar permanent fund for Utah's public education programs became one of the earliest institutional allocators to systematically build AI into the way it operates.

For the past 16 months, Ryan has led a rigorous AI landscaping exercise at SITFO, evaluating vendors, building agentic workflows, and integrating AI into the operational fabric of the agency. As a member of several industry networks of leading endowments, foundations, and health systems, Ryan found that almost no one else in those communities was actively implementing AI, which only deepened his conviction that SITFO needed to lead.

Ryan walks through why SITFO moved past using AI to generate investment memos, which it found to be a commoditized solution, toward a cross functional platform that serves finance and operations, strategy and risk, and manager research. He explains the integration work required to connect AI to the CRM, benchmarking systems, performance vendors, email, and shared document storage, and why that unglamorous plumbing work is what makes the AI powerful. He also shares his view that institutional investors have been slow to adopt AI because they treat it as a productivity tool rather than a transformational technology, and why SITFO concluded that the cost of moving early outweighed the risk of waiting.

What You'll Learn:

• How SITFO grew from three people and an inherited portfolio of Vanguard mutual funds into a sophisticated allocator over its first ten years
• Why SITFO moved past using AI for investment memos, which it found to be a commoditized solution, toward a single cross functional platform serving finance and operations, strategy and risk, and manager research
• Why SITFO prioritized integrating AI with its existing CRM, benchmarking, and performance systems
• Why Ryan believes institutional investors have been slow to adopt AI because they treat it as a productivity tool rather than a transformational technology
• Why SITFO concluded that the cost of being early outweighed the risk of waiting, and what that meant in practice
• How SITFO built a pipeline tool that pulls documents from its CRM daily and screens managers against desirable metrics across asset classes
• Why low hanging fruit such as reviewing limited partnership agreements, populating subscription documents, and redlining NDAs frees the team for higher value work
• Why Ryan believes manager research is shifting back to a people business, with more time spent on reference checks and relationship building and less on memo writing
• Why SITFO's two most recent hires were chosen for technical skill rather than manager research background, and what that signals about where the team is headed
• Ryan's advice for peers earlier in their AI adoption journey: assess your resources and objectives first, then run a structured landscaping process
• How SITFO secured board and management support years in advance, including through an internal working group it calls the Skynet group
• The data integrity and integration challenges SITFO worked through, including cleaning its CRM and controlling permissions across email and shared drives

About Ryan Kulig:

Ryan Kulig is the Finance and Operations Officer at SITFO, the Utah School and Institutional Trust Funds Office, a $4.7 billion-dollar permanent fund to support Utah’s public education programs. Ryan joined SITFO in 2016 to manage office operations, portfolio administration, and investment analysis, and has spent the past 16 months leading the agency's AI landscaping and implementation effort. Before joining SITFO, he worked at Sax Angle Partners, specializing in fundamental and technical analysis of equity investments. Ryan holds a Bachelor of Business Administration in Global Business from the University of Portland and an MBA from the University of Southern California.

Episode Highlights:

[01:57] From Three People to a Sophisticated Allocator

Ryan traces SITFO's growth from three founding employees and an inherited portfolio of Vanguard mutual funds to a fully built out institutional allocator, and what it took to establish the foundational governing documents early on.

[03:12] Beyond Investment Memos: One Solution Across Three Verticals

SITFO's early attempt to use AI for investment memos quickly proved commoditized. Ryan explains how that pushed the team toward a broader search for a solution that could serve finance and operations, strategy and risk, and manager research, and why integration with existing systems was a key priority. 

[04:44] Why the Industry Is Holding Back

Ryan's view on why institutional investors, as risk conscious fiduciaries, have been slow to adopt AI, and why many are still treating it as a productivity tool rather than a transformational one.

[06:42] Why SITFO Chose to Move Early

Ryan explains SITFO's calculation that the cost of being early outweighed the risk of waiting, and how the team built a cross functional case for AI that could benefit every vertical in the agency rather than a single department.

[07:39] Low Hanging Fruit: LPAs, Subscription Documents, and NDAs

Document intensive work is the clearest early win. Ryan explains why automating the baseline redlines of an NDA does not replace the attorney, it frees the attorney to focus on a more thorough review.

[08:37] Building the Pipeline Engine: Top of Funnel to Bottom of Funnel

Ryan describes the tool SITFO built that pulls documents daily from its CRM and files them into active and prospective manager hubs, allowing the team to landscape its entire network and screen managers against asset class specific metrics.

[10:31] The Cultural Shift: Back to a People Business

Ryan explains how AI is moving the manager research role away from quantitative screening and memo writing and back toward reference checks, relationship building, and firsthand observation of how managers operate.

[11:25] SITFO in Three to Five Years: Hiring for Technical Skill

SITFO's two most recent hires were chosen for technical aptitude rather than manager research background. Ryan explains why he expects the team to spend significant time coding and developing prompts to build out the agency's AI framework.

[12:16] Why Manager Research Is Becoming a People Business Again

Ryan's view on why AI, by making content easier to produce, will push allocators back toward firsthand experience and direct relationships as the basis for conviction.

[13:46] Advice for Peers: Resources, Objectives, and a Structured Process

Ryan's framework for allocators evaluating an AI solution: assess your team's resources and skill set, define what you are trying to achieve, and run a structured landscaping process. 

[15:11] Governance: Board Buy In and the Skynet Group

Ryan describes how SITFO secured support from its board and CIO years in advance, including through an internal working group established roughly three years ago to explore AI implementation across the organization.

[16:05] Challenges: Data Integrity and Integration Permissions

Ryan walks through the unglamorous work behind the AI bu...

Professor Emmanuel Yimfor, Assistant Professor of Finance at Columbia Business School, joins The Agentic Allocator to share his research that should sit at the centre of every conversation about AI in private markets. His work documents the core friction driving racial and gender disparities in access to capital: not quality, not track record, but networks. Who you can reach, not how good you are.

That finding has direct and urgent implications for how AI gets deployed across the LP/GP ecosystem. Used thoughtfully, AI has the potential to widen the top of the funnel dramatically, reducing the cost of due diligence enough that LPs can evaluate managers far beyond their existing networks. Used carelessly, the same tools will automate and entrench the same exclusions, encoding past decisions into future ones in ways that are subtle, hard to detect, and difficult to reverse.

Professor Yimfor walks through the mechanics of embedding-based matching and why it is a black box that can pick up on signals of race, gender, and network affiliation even when no one intended it to. He explains what the research on accelerators and structured access programmes shows about what happens when the top of the funnel is genuinely open. He makes a clear, practical case for what LPs, GPs, and technology developers should each be doing differently right now.

What You'll Learn:

• Why the core friction driving racial and gender disparities in private markets is networks and what the research evidence shows
• Why Black and Hispanic founders raise around 40% less capital than peers with identical patent holdings, educational backgrounds, and track records
• Why the gap in funding disappears entirely when access is structured, as in accelerators and grant programmes, and what that tells us about where the problem lies
• How embedding-based matching works, why it is a black box, and how it can encode biases in allocation decisions even when no one intended it to
• Why asking AI how similar a new manager is to managers you have backed before is not objective, and what the alternative looks like
• How structured, standardised due diligence processes enabled by AI can reduce the role of network signals and subjective impression in manager evaluation
• What GPs should do differently when preparing pitch materials and identifying which LPs to approach in an AI-enabled world
• What LPs should ask any technology developers and providers about how their existing AI tools are sourcing and filtering the managers they evaluate
• Why the industry is at a fork in the road and what Professor Yimfor’s research will be tracking to understand which path it is taking

About Professor Yimfor

Professor Emmanuel Yimfor is an Assistant Professor of Finance at Columbia Business School. His research focuses on the core frictions driving disparities in access to capital in private markets, with a particular focus on race, gender, and the role of networks in determining which founders and fund managers receive funding. His work has direct implications for how AI systems are designed and deployed across the LP/GP ecosystem, and he is currently researching how AI adoption is reshaping the equilibrium dynamics of capital allocation across the industry. Before joining Columbia Business School, he was an Assistant Professor of Finance at the University of Michigan Ross School of Business.

Episode Highlights:

[00:30] The Core Friction: Networks, Not Quality

The racial gap in access to funding disappears in structured settings like accelerators and grant programmes where anyone can apply. It shows up most sharply in relationship-driven contexts. Black and Hispanic founders raise around 40% less than peers with identical credentials and track records. The mechanism is the same for gender. Who you can reach matters more than how good you are.

[03:55] How AI Can Fix or Entrench the Problem

Whether AI amplifies or reduces existing disparities depends entirely on how the system is trained and what input data it uses. Ask AI how similar a new manager is to managers you have backed before, and the model will automate the same exclusions that drove the original gap, because the past portfolio was built through the same narrow networks. Use AI to expand the set of pitch decks you evaluate and the dynamic flips.

[09:10] The Embedding-Based Matching Risk

Embedding-based matching converts pitch materials into numbers and compares them to past allocations. Behind the hood, even if no one has consciously made a decision based on race or gender, the model may be picking up on those signals. The past decisions pollute future decisions in ways that are subtle and hard to detect. Opening the black box and auditing what features the model is learning from is not optional, but essential. 

[14:20] What the Research on Structured Access Shows

Where the top of the funnel is as wide as possible, with an Apply Here button and a structured evaluation process, the gap in access to funding for underrepresented founders disappears. That finding is the clearest signal in the research about where AI holds the greatest promise: using time savings from processing more materials to have more in-person meetings with people outside your existing network, rather than fewer.

[14:20] Practical Advice for GPs

Use AI to identify which LPs are most likely to be a fit for your strategy based on publicly available mandate information, rather than relying entirely on network referrals. Resist the urge to generate pitch materials using the same AI systems that LPs are using to evaluate them. The GP that has great ideas but historically lacked the resources to present them well now has a opportunity to close that gap.

[16:20] Practical Advice for LPs

Ask how your existing pipeline of managers came to you. Are there opportunities to expand the top of the funnel using this technology? Are managers running similar strategies being evaluated with the same questions, regardless of their background? Those are the questions any AI implementation consultant should be helping you answer.

[21:15] The Fork in the Road: What the Research Will Track

AI adoption in private markets will resolve one of two ways. If LPs use it to widen their search, more traditionally underrepresented GPs enter the market and the research will show whether they deliver. If LPs use past data and past networks to train their systems, disparities in capital allocation will widen. Professor Yimfor is using big data to track exactly which path the industry is taking.

Episode Resources:

Professor Emmanuel Yimfor on LinkedIn

Columbia Business School Faculty Profile

Victoria Sienczewski on LinkedIn

AuumAI Website

Disclaimer: This podcast is for informational purposes only. The views expressed are those of the speakers as of the recording date and may change over time.

Shaun Ng, founder of AI for Allocators and former Managing Director at the Cleveland Clinic Investment Office, joins The Agentic Allocator to share what three decades of capital allocation experience and over 50 newsletters on AI adoption have taught him about where LP organisations are going wrong and what they need to do differently.

Shaun's diagnosis is clear: the single biggest mistake allocators are making is misidentifying the AI challenge as a technology problem. It is not. It is the most consequential strategic transformation of their careers: a leadership challenge, a change management challenge, a cultural challenge. Every downstream mistake, from delegating AI to IT project managers to setting fixed start and end dates for implementation, flows from that one misdiagnosis.

In this episode, Shaun walks through the pre-AI pressures that were already straining investment offices: stakeholder demands, data complexity, talent, and explains how AI implementation maps onto each one. He makes the case for why CIOs who are not personally using AI are making a critical error, why creating the right environment matters more than choosing the right tools, and what the AI flywheel looks like when it is spinning properly. He also offers a vivid picture of what a genuinely AI native investment office looks like in four to five years. The edge will belong to organisations that start building that environment now.

What You'll Learn:

• The three core pressures LP organisations were already facing before AI arrived: stakeholder demands, data complexity, and talent
• Why every common AI mistake allocators make flows from one foundational misdiagnosis and what that misdiagnosis is
• Why CIOs who encourage their teams to use AI without using it themselves are repeating a strategy that will not work this time
• Why starting with tools is the wrong first step, and what to focus on instead
• What 'communicating your AI stance' means in practice 
• How to build the AI flywheel: the combination of communication, guidelines, and incentives that sustains institutional adoption over time
• Why moving from individual AI use to institutional value requires the decision makers, not just the junior analysts, to lead the charge
• What an AI native investment office looks like in four to five years: agents digesting manager letters, flagging inconsistencies, and routing human judgment to where it matters most
• Why getting proficient in AI in one small area produces unexpected benefits across completely different parts of the investment process
• How one allocator's AI fluency helped him identify AI slop and AI washing in manager meetings. A use case nobody predicted

About Shaun Ng:

Shaun Ng is the founder of AI for Allocators, an independent newsletter with over 50 editions dedicated to helping LPs navigate the complexities of AI adoption. He brings a 30 year career at the heart of capital allocation, most recently as Managing Director at the Cleveland Clinic Investment Office and previously in a senior leadership role at the World Bank Pension and Endowment Group. His work sits at the intersection of institutional investing, and the strategic transformation challenge that AI represents for the allocator community.

Episode Highlights:

[02:20] The Three Pre-AI Pressures LP Organisations Are Already Facing

Before AI entered the conversation, investment offices were already under pressure. Stakeholder demands were rising, IC decks were getting thicker, and team sizes were not growing. Managing data complexity, across both quantitative performance data and unstructured qualitative material, was consuming enormous time and resources. And talent remained a constant challenge: recruiting the right people, developing them, onboarding them quickly, and ensuring they could contribute at their potential. AI arrived and immediately touched all three.

[04:55] The One Misdiagnosis That Explains Every Downstream Mistake

Shaun identifies a single root cause behind the most common LP AI mistakes: treating AI as a technology problem rather than a historic strategic transformation. CIOs have been tasked with navigating their investment offices from a pre-AI to a post-AI world. The analogy is electricity. Factories had to be fundamentally redesigned to take full advantage of it. Delegating that task to IT, setting a project timeline, or skipping personal engagement with the tools: all of these are symptoms of the same misdiagnosis.

[08:30] Why CIOs Who Do Not Use AI Are Making a Critical Error

One of the most common missteps Shaun sees: senior leaders who encourage AI adoption without personally using the tools. In previous technology cycles, it was possible for a CIO to run an effective portfolio without knowing how to use Aladdin. That model will not work for AI. This is not a risk system. It is an infrastructure level transformation, and leaders who do not understand it from the inside cannot guide their organisations through it.

[10:15] Start With the Environment, Not the Tools

When allocators ask Shaun what AI tools to use, his answer consistently surprises them: do not start with tools. The temptation to build vendor shortlists and compare peer approaches feels like progress but stops organisations from building the long term capability they need. The real question for any CIO is how to create an environment in which the team can adopt AI effectively, in the areas that matter most. Not on low value tasks that do not move the needle.

[12:40] Communicate Your AI Stance, Build the Flywheel

Shaun outlines three elements that turn a one-off initiative into a sustained institutional capability. First: communicate your AI stance. Even a simple acknowledgement that AI is here to stay and the team needs to figure it out together removes the fear that stops people from experimenting. Second: give the team high level guidelines so they know they will not get into trouble exploring new tools. Third: build the AI flywheel using incentives: formal OKRs, informal celebrations of shared breakthroughs, so that adoption accelerates over time rather than fading after the first month.

[17:00] From Individual Use to Institutional Value

The gap between a junior analyst using AI to write investment memos and an organisation extracting genuine institutional value is significant. Shaun draws on research from McKinsey, PwC, and Stanford to explain what it takes to close it: the people leading AI adoption must be domain experts who understand the business, not IT professionals learning the workflows as they go. Decision makers, not just junior staff, need to be driving the change. And the mechanism for sharing breakthroughs: brown bag sessions, AI workflow days needs to be built deliberately.

[20:30] What an AI Native Investment Office Looks Like in Four to Five Years

In the near term, agents will handle the recurring, documentable tasks: reviewing emails, drafting responses, digesting manager letters, flagging inconsistencies against known mandates. Human judgment gets directed to the genuinely hard questions. Further out, GP agents and LP agents will begin communicating directly, and the frontier research techniques being developed by AI labs: auto researcher capabilities, autonomous investment thematic work, may reshape how allocators think about manager selection and portfolio construction entirely.

[23:45] The Unexpected Cross Pollination of AI Proficiency

Paul Fleming, Founding Partner and CEO of Fleming and Partners, joins The Agentic Allocator to share how his firm is delivering what global investment consulting giants have never been able to offer: truly independent, unconflicted, institutional-grade strategic advice for the world's most sophisticated family offices and asset owners, with AI at its core.

Paul brings over 18 years of experience, including as Head of Endowments, Foundations and Single Family Offices at Mercer, the world's largest investment consultancy, and as CIO of a prominent family office. His background gives him an authoritative view of where the traditional model falls short and shapes everything about how Fleming and Partners operates differently.

At the heart of that difference is what Paul calls the brain: an AI-powered knowledge nucleus that develops, holds, and curates valuable client data alongside market intelligence, so that every piece of institutional knowledge is retained, accessible, and compounding over time rather than walking out the door when a team member leaves. 

Fleming and Partners is not retrofitting AI onto a legacy model. It was designed from day one with AI as a core part of how it delivers for clients. That is a structural advantage that the world's largest consultants, with decades of inherited infrastructure and competing commercial interests, cannot easily replicate.

What You'll Learn:

• Why Paul founded Fleming and Partners on the conviction that traditional consulting models are not fit for purpose for sophisticated family offices
• What it means to be in build mode rather than dismantle and rebuild mode, and why that is a structural advantage for a boutique firm
• The AI Brain: what it is, why it matters, and why the absence of one is a central institutional knowledge problem facing advisory firms today
• Why privacy and discretion are central to building an AI-powered knowledge system for family office and endowment clients
• Why Paul believes the investment consulting industry will have big winners and big losers over the next ten years, and what separates them
• How Fleming and Partners is building a culture of curiosity: what that means in practice and why it becomes more important as technology improves
• Why the $6 trillion family office market is being served by tools built for pension funds, and why that gap is the opportunity
• How AI allows a boutique firm to compete at the scale of global consultants with $16 trillion AUA
• Paul's practical advice for family offices early in their AI adoption journey: baby steps, lean on practitioners, and do not try to reinvent the wheel

About Paul Fleming: 

Paul Fleming is Founding Partner and CEO of Fleming and Partners, a London and UAE-based investment advisory firm serving sophisticated family offices, endowments, foundations, and institutional asset owners. Paul has over 18 years of experience in investment consulting and advisory, including as Head of Endowments, Foundations and Single Family Offices at Mercer, the world's largest investment consultancy, and as CIO of a prominent family office.

Episode Highlights:

[02:00] Why Fleming & Partners Was Founded

The investment consulting industry is serving family offices with tools and frameworks built for the world's largest pension funds. Paul's conviction is that sophisticated family offices deserve something different: independent, rifle-shot strategic advice from people who sit on the same side of the table and have no products to sell.

[05:00] Build Mode, Not Dismantle and Rebuild

Being a young, lean organisation is a structural advantage. There is nothing to dismantle. Fleming & Partners can go directly to the best tools in the market without the institutional inertia that makes change costly for large established investment consultancies.

[06:40] The Brain: Fleming and Partners' AI Knowledge Nucleus
The central strategic build at Fleming & Partners is what Paul calls the Brain: an AI-powered system that develops, holds, and curates valuable client data alongside market intelligence. The goal is that every piece of institutional knowledge, rather than sitting in siloed departments or walking out the door with a departing employee, is retained, accessible, and compounding.

[10:00] ROI: Augmenting Human Judgment, Not Replacing It

Fleming & Partners are strategic advisors, and the power of human judgment remains central to that. What AI does is augment and scale that judgment: covering larger data sets faster, with less human error, across governance, data analysis, reporting, monitoring and other use cases. There is value add at every element.

[12:30] Culture: Curiosity as a Competitive Advantage

Paul's ask of every team member, regardless of seniority, is to remain curious about how AI can develop their thinking and add value. The culture at Fleming & Partners is global mindset delivered by partner minds: professionals who think like owners, test boundaries for clients, and bring genuine intellectual curiosity to everything they do.

[16:00] The $6 Trillion Opportunity

The global family office market is the same size as the global hedge fund market and is on its way to $10 trillion by 2030. It is currently being served by tools built for the pensions market. Paul's view is that many of those tools are not fit for purpose, and that the real opportunity is to deliver democratised, institutional-grade advice in a truly boutique and unconflicted way.

[19:57] Advice for Family Offices Starting Their AI Journey

Do not think too big. Do not try to build something at a scale that is unrealistic in the short term. Start with what is available, embed the tools into your existing processes, get comfortable using them, and build from there. Baby steps. Lean on the true practitioners and ask questions. Curiosity is the starting point.

Episode Resources:
Paul Fleming on LinkedIn

Fleming & Partners Website

Victoria Sienczewski on LinkedIn

AuumAI Website

Alex Harstrick, Managing Partner and Co-Founder of J2 Ventures, joins The Agentic Allocator to share how one of the most focused early-stage funds in the US is building AI into every aspect of how it operates. J2 invests exclusively in deep technology for the US government, and Alex brings a sharp view of what AI adoption requires from GPs and LPs alike.

Alex's position is unambiguous: you are leaving money on the table by not implementing AI right now. Cybersecurity questions, accuracy concerns and open questions of all kinds remain unanswered. But waiting for all the answers, he argues, means waiting for the industry to pass you by before you have them.

In this episode, Alex walks through how J2 is using AI to improve its internal operations, why the entire VC SaaS stack is ripe for disruption, and how AI is reshaping the sourcing game in ways that go far beyond automation. He also shares a pointed view on how LPs are thinking about AI adoption, why the largest and most sophisticated institutions are often the most afraid, and how AI can surface the kind of manager evaluation information that social relationships and slick pitch decks were designed to obscure.

What You'll Learn:

• Why Alex calls AI 'advanced computing' and what that signals about how J2 thinks about technology investment
• Why the entire VC SaaS stack is overpriced, under-delivering, and vulnerable to being replaced by tools you can build yourself over a weekend
• The time cost of manual CRM notes: five minutes per meeting, five to ten meetings a day, 120 hours lost per year, and what that means for deal flow
• How AI changes the sourcing game not just by finding founders faster but by helping VCs actually connect with them as human beings
• Why the largest and most sophisticated LPs are often the most afraid of AI, and why Alex thinks they are asking the wrong question
• How AI can evaluate managers against 56,000 alternatives and surface performance information that social relationships and polished track records were designed to hide
• Why AI could perpetuate backward-looking biases in manager selection and what LPs need to understand about that risk 
• The retrospective analysis opportunity: tracking the nine managers you passed on and using their subsequent performance to improve your own decision-making
• What the LP/GP ecosystem looks like in five to ten years: automated DDQs, blind ranking mechanisms, algorithmic portfolio company reporting, and the parts that will always need a human in the loop
• Why if you do not leverage AI, someone else will, and they will out-compete you

About Alex Harstrick: 

Alex Harstrick is Managing Partner and Co-Founder of J2 Ventures. J2 invests exclusively in deep technology for the US government. Before founding J2, Alex had a career that spans healthcare venture capital, service as an Army intelligence officer with special operations deployments to Afghanistan and Iraq, and a role at the Defense Innovation Unit where he helped deploy nearly $2 billion into defense technology startups. That combination of operational experience, government service, and investment expertise is what shapes J2's uniquely focused approach to backing founders building at the intersection of national security and advanced computing.

Episode Highlights:

[01:51] Why J2 Calls It Advanced Computing, Not AI

Alex believes the word AI has lost a lot of meaning. Alex explains why J2 uses the term advanced computing instead, and why an early-stage investor who only looks for AI in the broad definition will miss a lot of what is interesting

[03:12] Why the Entire VC SaaS Stack Is Ripe for Disruption

VC is a niche industry served by expensive software that ratchets up pricing once it has you, without meaningfully improving. The best CRM most VCs have ever used is a shared Google Sheet. You can now build something better, bespoke for your workflows, over a weekend.

[06:02] The Real Cost of Manual Note-Taking: 120 Hours a Year

Five minutes per meeting, five to ten meetings a day, 120 hours lost per year. Once you frame it that way, the urgency of fixing it becomes impossible to ignore.

[09:19] LPs and the AI Question: Why the Biggest Institutions Are the Most Afraid

The largest multi-asset managers are mostly asking about cybersecurity. Alex thinks that is the wrong question unless you are simultaneously asking how you are implementing AI across your workflows.

[15:04] How AI Changes Sourcing: Finding Founders and Actually Connecting With Them

Everyone is looking for expectational founders. What is differentiated is getting there first and walking into the room knowing the things about that person that create a real connection. AI helps with both.

[21:11] How AI Evaluates Managers Against 56,000 Alternatives

LPs often rely on relationship signals: impressive annual meetings, famous co-investors, high-profile connections. AI can evaluate a manager against the full universe of alternatives and surface what the polished materials were designed to obscure.

[22:44] The Retrospective Analysis Opportunity

When an LP meets ten managers and invests in one, the nine they passed on are almost never revisited. AI makes it possible to track those managers, compare their subsequent performance, and use that data to improve your own decision-making.

[23:19] The Bias Problem: Why AI Can Perpetuate Historical Heuristics

Train AI on historical manager data and it will tell you the best managers look like the ones who succeeded historically. LPs need to understand that risk explicitly and understand what to do about it. 

[24:40] The Five to Ten Year View

DDQs processed algorithmically, managers ranked against blind benchmarks, portfolio company reporting automated. The part that will not change: people want to meet the person they are giving up a meaningful part of their lives to work with.

Episode Resources:

Alex Harstrick on LinkedIn

J2 Ventures Website

Victoria Sienczewski on LinkedIn

AuumAI Website

Brad Calder, Managing Director, Head of Equities at TIFF Investment Management, joins The Agentic Allocator to walk through what two years of AI adoption looks like inside the OCIO. Brad leads TIFF’s public equity investments, including select arbitrage and crypto strategies, and plays a central role in manager selection and portfolio construction, and has spearheaded the firm's AI implementation. TIFF is an OCIO serving endowments, foundations, and other mission-driven organizations. In this episode, Brad walks through what AI adoption looks like on the ground: the use cases that are delivering tangible ROI, the ones that are still works in progress, the infrastructure decisions that turned out to matter, and the cultural conditions that made it all possible.

Brad describes one of the most ambitious projects on TIFF's roadmap: a tool for the investment committee that connects every memo the firm has ever written to structured returns and exposure data, automatically surfacing the three most comparable historical investments whenever a new decision is being discussed. For a 35-year-old organization where no current IC member has been there since inception, this tool would give the IC access to the full weight of every decision TIFF has ever made.

He also makes a sharp point about what AI adoption requires that most allocators underestimate: the willingness to keep re-experimenting, because the underlying models improve fast enough that something that failed six months ago may now be entirely viable. Think of it like high school chemistry, he says: if the experiment doesn't work, you alter it and try again.

What You'll Learn:

• How TIFF's decade of backing machine learning oriented hedge funds gave them a head start on understanding and trusting AI before the rest of the industry caught up
• Why TIFF's legacy research management system became the catalyst for their AI build-out and what the 'SaaSpocalypse' means for allocators still locked into incumbent technology providers
• How TIFF is using AI across unstructured data pitch books, quarterly letters, capital call notices 
• The collective intelligence tool TIFF is building for its investment committee: surfacing comparable historical investments, tied to returns data, going back to the firm's founding
• Why the ROI of AI for allocators is not just efficiency and where Brad believes the real value will ultimately be felt
• How TIFF combines offshore BPO with AI tools, and why the combination is complementary rather than substitutive
• What an AI-enabled allocator looks like in five to ten years: instant probability estimates on prospective managers, AI-shaped due diligence agendas, and humans focused on the relationships and control rights that resist disintermediation
• Why subscale endowments and foundations face a real risk of being left behind and why that is part of TIFF's mission to solve
• The high school chemistry framework for AI experimentation: alter the experiment, keep trying, and never conclude something can't work based on a model that's already six months out of date

About Brad Calder:

Brad Calder is Managing Director, Head of Equities at TIFF Investment Management, where he leads TIFF’s public equity investments and has spearheaded its AI adoption and implementation. TIFF is an OCIO serving of endowments, foundations, and other mission-driven organizations. Brad has been at TIFF for 11 years. Before joining TIFF, he built expertise in systematic and machine learning-oriented investment strategies that has since informed the firm's approach to integrating AI across its investment and operational process.

Episode Highlights:

[02:12] How TIFF Got a Head Start on AI

Brad traces TIFF's AI journey back a decade, to when the firm began backing machine learning oriented systematic hedge funds. That early exposure, learning how the technology worked from the inside, meant TIFF was ready to move fast when ChatGPT changed the landscape. The culture of continuous improvement that governs how TIFF evaluates external managers, Brad explains, is the same standard they've applied to their own organization.

[04:39] The SaaSpocalypse and the Case for AI-Native Vendors

TIFF's AI build-out was triggered, in part, by frustration with a legacy research management system that lacked the API capabilities needed to run LLM models over their database. Brad frames this as a broader market dynamic the 'SaaSpocalypse', in which allocators are waking up to the fact that incumbent SaaS providers that aren't investing in AI are creating white space for AI-native competitors to step in.

[06:14] What AI Is Actually Doing for TIFF’s Investment Teams

Brad walks through specific use cases: rapidly summarising incoming manager pitches against a structured template, comparing quarterly letters across managers over multiple periods to surface trends and outliers, and helping write investment memos. The shift from an analyst manually reading and cross-referencing multiple letters to an AI that can instantly identify what one manager is doing differently from the rest is, Brad argues, qualitatively different, not just faster.

[07:39] Building TIFF’s Collective Intelligence Layer

The most ambitious project on TIFF's roadmap is a tool for the investment committee that connects every memo the firm has ever written to structured returns and exposure data and automatically identifies the three most comparable historical investments.

[11:09] Where the Real ROI Lives and Where It Doesn’t

Brad is clear that AI adoption at TIFF is not primarily a headcount reduction exercise. The offshore BPO team in India has full access to TIFF's AI systems, and the combination is complementary. The true ROI, he argues, will be felt most in decision quality: the tools being built for the investment committee are designed specifically to help the team make better decisions and avoid the kind of mistakes that result in large losses. 

[14:08] The Challenge of Building Confidence in AI Output

Brad is candid about the adoption friction inside TIFF: even when validation processes are in place, some team members still feel the instinct to double-check, follow up, or verify. Building the confidence to trust a validated AI output takes time and deliberate cultural work. His advice: keep re-experimenting, because the models are improving fast enough that a use case that failed six months ago may be entirely viable today.

[15:56] The Five-to-Ten Year Vision: Humans Focused on What AI Cannot Do

Brad's picture of the AI-enabled allocator is specific: systems that can take textual data and make forecasts from it, instant probability estimates on prospective managers drawn from decades of comparable data, and AI-shaped due diligence agendas that surface questions humans wouldn't have thought to ask. Where he believes humans remain essential for longer is in the relationship and control dimension: negotiating GP terms, engaging company management, exercising the kind of judgment that equity ownership demands.

[20:08] The Scale Risk for Smaller Allocators

Brad closes with a note of concern for subscale endowments and foundations that lack the budget to make these investments. AI, he argues, is not a rising tide that lifts all ships. It only lifts the ones that invest in making it work. That gap is precisely why TIFF exists: to help smaller mission-driven organisations access the sam...

Al Hemmingsen, CIO of a prominent US-based single family office, joins The Agentic Allocator to share what hands-on AI adoption looks like from inside an institutional family office. Al is the only CIO, we know of, in the LP world actively publishing open-source code on GitHub for use by fellow allocators.

In this episode, Al walks through how vibe coding with Claude Code became a serious tool for unlocking legacy data and automating workflows, how he is building a culture of accountability across his team, and why the LP that gets to the best GPs first has a competitive edge that only widens over time.

He also makes one of the sharpest points we have heard on the competitive stakes: some aspects of this work should be shared freely with peers, but the ability to process information faster than other LPs, build higher-quality manager meeting lists, and get to the best GPs before they fill their calendars at conferences, is a genuine edge. If you do not build it, someone else will.

What You'll Learn:

• How a background in SQL and Visual Basic at business school laid the foundation for a career of hobbyist coding and eventually full vibe coding with Claude Code
• Why the single best measure of AI ROI at a family office is not efficiency gains but tasks that were previously too expensive or simply not feasible
• How $59.32 in API credits unlocked 50-plus years of legacy transaction data and saved $10,000 a year in perpetuity, in one afternoon
• How Al thinks about the boundary between enterprise systems of record and the custom scripts that live in between them
• The psychological dimension of AI adoption: how to take existential fear off the table so your team can actually engage with the technology
• The autopilot accountability principle: why the person who creates an AI tool is always responsible for what it does, regardless of what the machine decided
• How a standing Monday meeting agenda item and a shared Microsoft environment for corporate-owned apps is building a firm-wide culture of AI experimentation
• Why the investment case for AI comes down to the widest top-of-funnel possible: seeing more managers, filtering more efficiently, and getting to the best GPs faster than peers who are not augmented
• Why sharing freely with peers and competing are not contradictory, and how to think about the line between the two

About Al Hemmingsen:

Al Hemmingsen is the CIO of a prominent US-based single family office with a history going back to the mid-20th century. He has been at the helm for the past decade, managing a multi-generational pool of capital with a quasi-perpetual mandate. Al is one of the most hands-on AI adopters in the institutional family office world and is the only CIO, that we’re aware, of actively publishing open-source code on GitHub for use by fellow allocators.

Episode Highlights:

[05:56] The $59.32 Moment

$59.32 in API credits unlocked 50-plus years of legacy transaction data in a single afternoon. That moment set off a systematic search for every legacy problem in the organisation AI could solve.

[09:18] Building in the Seams

What is worth building are the tools that live between enterprise systems: workflows too specific for any software company to build a product for.

[11:23] Taking Fear Off the Table

The goal is to eliminate the 30% of every job that is tedious and repetitive, not the people doing it. Remove that fear first and the rest follows.

[19:09] The Investment Case

There are 56,000 alternative managers globally. The LP that processes information faster and gets to the best GPs have a genuine edge.

[20:55] Collaboration and Competitive Advantage

Al's view is that sharing tools and ideas freely with peers in the family office world makes everyone better, which is why he publishes on GitHub. But that openness has a natural boundary: the capability to see more managers, filter faster, and reach the best GPs before their schedules fill is a genuine investment edge worth building and protecting.

Episode Resources:

• Al Hemmingsen on LinkedIn
• Al Hemmingsen on GitHub
• Victoria Sienczewski on LinkedIn
• AuumAI Website

Disclaimer: The views and opinions expressed by Al Hemmingsen in this interview are his own and do not necessarily reflect the views, positions, or policies of any organization with which he is or has been affiliated.

Professor Ludovic Phalippou of Oxford Saïd Business School joins The Agentic Allocator to deliver a rigorous, unsparing look at what AI can and cannot do for Limited Partners in private markets. Drawing on his research paper, 'Limited Partner Versus Unlimited Technologies,' and hands-on experiments applying machine learning to real LP datasets, Professor Phalippou maps both the transformative potential and the structural risks of AI adoption in private markets. 

Private markets are not data-poor - they are data-overwhelmed. A single fund investment can generate thousands of pages of PDFs and countless spreadsheets. A large LP like CalPERS monitors 200 active private equity funds and receives quarterly reports on 3,000 underlying portfolio companies. No human can absorb that. The promise of AI is that it can. But Phalippou warns that the same tools capable of unlocking hidden signals in qualitative data can just as easily be gamed, misused, or deployed in ways that amplify the industry's existing distortions.

Professor Phalippou shares research showing how AI sentiment analysis of GP quarterly reports can predict future portfolio company returns with real predictive power, far beyond what the reported marks alone convey. He also lays out in precise detail how GPs could exploit AI-reliant LPs, from burying critical fee terms ever deeper in footnotes to inserting invisible prompt injections into documents. He outlines the governance framework LPs need to build before they trust any AI output with a real decision.

What You'll Learn:

• Why private markets have too much information, not too little, and why that is precisely where AI's power lies
• How AI sentiment analysis of GP quarterly reports can predict portfolio company returns beyond what reported marks reveal
• Why extracting a headline EBITDA figure with AI is a recipe for catastrophe, and what to ask for instead
• How GPs could exploit AI-reliant LPs: from burying critical terms in footnotes to inserting invisible prompt injections into fund documents
• The legal frontier: why AI provisions will soon appear in LP/GP confidentiality agreements
• A governance blueprint for LP organisations adopting AI: who is responsible when an AI-extracted number turns out to be wrong?
• Why the right question to ask AI is never 'give me the EBITDA,’ and what high-quality AI prompting for private markets actually looks like
• Why LPs must avoid both naive enthusiasm and reflexive dismissal, and what a mature, risk-aware approach to AI adoption actually looks like

About Professor Ludovic Phalippou:

Ludovic Phalippou is a Professor of Financial Economics at Oxford’s Saïd Business School and specializes in Asset Management, with a special focus on investments in private equity funds. He is one of the world's leading academic authorities on private equity performance, fees, and transparency. His research on private markets has been widely cited by LPs, regulators, and policymakers. He sits on the investment committee of Oxford’s Queen’s College, has collaborated with institutional LPs on applied machine learning research, and authored the paper 'Limited Partner Versus Unlimited Technologies,' which examines how generative AI and machine learning are poised to reshape LP decision-making, and the risks that come with it.

Episode Highlights:

[00:03:21] The Real Data Problem in Private Markets

Phalippou dismantles the assumption that private markets suffer from a lack of data. The problem, he argues, is the opposite: thousands of pages per fund, quarterly reports on thousands of portfolio companies, and none of it structured or comparable. AI's promise is to swallow that volume and surface what matters.

[00:04:31] Sentiment Analysis as a Return Predictor

Drawing on original research, Phalippou explains how running AI sentiment analysis on GP quarterly report text produces a powerful predictor of future portfolio company returns - one that adds significant information beyond the reported mark. Two companies both marked at 1.2x can have dramatically different outcomes depending on how positively or neutrally the accompanying text is written.

[00:09:19] Why EBITDA Extraction Is the Dangerous Use Case

The most common instinct for LPs, using AI to automatically extract and aggregate financial metrics across hundreds of portfolio companies, is also the most dangerous. EBITDA figures in private markets are loaded with definitional choices, add backs, and adjustments buried in footnotes. Asking AI for the headline number without the context around it can produce false confidence.

[00:11:40] How GPs Can Game AI-Reliant LPs

Once GPs know their documents are being processed by AI rather than read by humans, the incentive structure shifts. Critical fee terms and performance qualifications will be buried even deeper in footnotes. More alarmingly, Phalippou explains how invisible prompt injections, text hidden in white-on-white font, can manipulate AI summaries to generate favourable assessments. Academics have already been caught doing this with peer review systems.

[00:13:29] What Good AI Prompting Looks Like for LPs

Rather than asking AI for a single extracted number, Phalippou recommends asking it to surface all relevant passages in a document on a given topic, including definitions, footnotes, and any conflicting disclosures. The absence of a clear definition is itself an information signal. Over time, analysts can train AI on their own judgment: flagging specific types of add backs, identifying patterns that have historically indicated problems.

[00:19:29] The Legal Frontier: AI Provisions in LP/GP Agreements

Some GPs are already attempting to prohibit LPs from processing their documents with AI. Phalippou argues the logic does not hold: when an LP hired Cambridge Associates, they were paying for judgments informed by thousands of comparable documents. Giving a document to an AI does the same thing. The conversation predicts LP/GP confidentiality agreements will soon include explicit AI provisions, a structural shift in how the industry governs information.

[00:21:00] The Machine-to-Machine Future

Phalippou reflects on an emerging world in which GPs produce documents with AI assistance, LPs process them with AI, and the analysis in between is conducted by AI. He uses the analogy of education: when exams are written, answered, and graded by machine, what is the human layer actually doing? His answer: setting the values, defining the questions worth asking, and retaining genuine accountability for the answers.

[00:23:29] A Governance Blueprint for LP Organisations

Phalippou's closing framework centres on responsibility. When an AI-extracted number turns out to be wrong and a real investment decision follows from it, who is accountable? Organisations must resolve this before deploying AI, not after. He also stresses the importance of avoiding both naive adoption ('this will simplify everything') and reflexive dismissal ('it just generates noise'). The mature position is to use these tools seriously, with clear governance and a clear-eyed understanding of the risks.

Episode Resources:

Professor Ludovic Phalippou on LinkedIn

Oxford Saïd B...