Full episode transcript: AI & PE: The Future of Value Creation

Guest: Adam Silverman, Senior Director, Accordion Intelligence Lab | Host: Kyle Roemer

Adam Silverman: If your engineering team is not using Claude Code, Codex, Cursor — any of these tools — they’re not going to be competitive.

Kyle Roemer: Welcome to AI & PE: The Future of Value Creation. I’m Kyle Roemer, Head of US Data and AI at Accordion. Today’s episode is part of our series with Adam Silverman. Adam has been building custom AI tools since before most people knew what an AI agent was. He’s tested over 500 leading agents, and more importantly, he sits at the rare intersection of deep, hands-on AI expertise and the real practical needs of private equity. In this episode, I speak with Adam about what PE leaders actually need to know about AI right now. Let’s get into it.

Development costs and timelines for building software and products are shrinking — so much that if you thought you were building it one way six months ago, that’s changed in the last three months with some of the new tooling, agents, and LLMs. Adam, what are you seeing in terms of AI use in software development?

Adam Silverman: If your engineering team isn’t using Claude Code, Codex, Cursor — any of these tools — they’re not going to be competitive. And if you’re working with consulting firms, and the teams behind the scenes aren’t using them, you’re overpaying — I think more and more clients are going to start to realize that. It’s really shrunk the timeline for how long it takes to execute projects. From a prototyping perspective, the rate at which you can prototype is unprecedented — you can build truly amazing prototypes very quickly. And then productionizing those engagements — what would’ve taken a year or two to build out a full-fledged platform can now be done in a matter of months. I think the need for very skilled engineers — the most veteran engineers, with twenty-plus years of software engineering — isn’t going anywhere.

These are the people who need to learn how to actually control these systems, because at the end of the day, you get so much code generated that you need someone who can parse through it and understand whether it’s truly logical, what it’s coming up with. I think this will keep permeating — a lot of low-level engineering tasks will be the first to get automated. But I really look at these tools as the biggest empowerment, both for internal use cases — if there’s an internal use case you’ve been banging your head against, just go try Claude Code and see how far you can get, even if you’re not an engineer — and for outsourcing work. I think within a quarter, you can have a huge impact within a company that otherwise would never have been possible.

Kyle Roemer: I love this quote, Adam, from one of our lead engineers within the AI Lab. She says, “Look, we’re not handwriting any code anymore — we’re effectively architecting and doing quality control for what agents are developing.” I just love the idea that the role of the senior engineer is evolving, and that experience is so important to getting these products right.

Adam Silverman: We’re two years into coding agents. If we’re talking February 2028, it’s going to be a totally different world — that’s really what I believe. I think there’s going to be more empowerment throughout the entire organization. What you used to think — “I have to go to a data analyst,” or “I have to go to an engineer on our team to build this out” — I think every business user is going to have this kind of functionality directly in their toolkit. For more complex applications, that’s obviously where you’re still going to need an engineering team to build out solutions, but their speed is going to be unprecedented.

I’ve said this publicly before, but I think people are going to be spending as much on LLM tokens every month as they spend on salaries — or maybe there’s a new type of architecture by then. But I truly believe that if you have an engineer getting paid, say, $20,000 a month, they might be spending $20,000 a month on these coding tools, but their productivity is going to be the equivalent of 10 or 20 engineers at once. It’s a really interesting paradigm, and a lot of companies are grappling with this right now. It’s an exciting time — if you’re thinking about this, you’re in the right place, because the companies that aren’t are the ones that are going to be left behind in the coming decade.

Kyle Roemer: I totally agree. Well, when you take a step back and look at these three themes or shifts — what do they tell you about AI and private equity right now, and where it’s headed?

Adam Silverman: Overall, the more data, the better. If you think about the companies that have gotten these crazy valuations — Scale AI, Mercor, a bunch of these other companies providing training data to the labs — what you realize is that the more data you’re capturing internally, the more valuable that is. I listened to an interesting podcast recently with Arvind from Glean, and he was talking about how everybody’s going to have access to the same baseline information. You look at the BlueMatrix partnership with Perplexity, for example — everybody’s going to have access to that kind of baseline information, which is amazing. It allows people to build entire businesses; it allows a solo person to start running a hedge fund. But the real value comes from the ten years of historical data these firms already have and want to access. So the more data you can capture throughout the entire organization — whether at the portfolio company level, the GP level, or the LP level — these are things you cannot take for granted. I don’t think that’s going anywhere. I think this is going to keep permeating over the coming years, and by capturing all this data, as the models continue to improve, you’re going to have the best possible chance of succeeding by leveraging this technology.

Kyle Roemer: Yeah, the data aspect of this is just so paramount, and people have said this a variety of ways — but whether you’re a horizontal SaaS player, a vertical player, or an average service business: start capturing more data, get it clean, get it integrated so you can actually leverage it. That’s ultimately going to be a moat for a lot of businesses going forward, more than the software products or services they offer. That’ll be a really interesting movement over the next several years.

Let’s move on to the second theme — going from agentic AI to ambient AI. Adam, can you define what that actually means?

Adam Silverman: We think of ambient agents as ones that kind of live in the background 24/7, doing work — while you sleep, when you wake up, throughout your workday. You don’t necessarily need to call on them and say, “Hey, go do this task” — they’re just happening behind the scenes. I use these types of agents every day, directly within Perplexity. I’ll say, “Who am I meeting with today? Can you create a consolidated list of who I’m meeting with, find their most recent tweet or LinkedIn post, find me research on the company I’m meeting with?” Every day, I have this almost personal assistant that lives in my inbox and tells me, “Here’s something you should be aware of about the people you’re meeting with.” That’s ambient — happening in the background.

I’m not going and kicking it off every day at 6 a.m. saying, “Go conduct research on who I’m meeting with.” I set it and forget it, and it happens automatically, daily. That’s how to think about ambient AI as you apply it to private equity — as deals come in, or there’s a big news release on a company, these are data points that can get pulled into your financial model or your decision-making. Maybe you looked at a deal three months ago, and there’s been a major breakthrough — say, on a drug that company’s developing — you should get pinged automatically: “Hey, you looked at this company, here’s the major breakthrough that just happened.” And that happens without you or an analyst going and doing the research.

It’s just happening in real time as you go about your workday. I’m really excited about this — I think it’s the promise agentic AI always held: you give it a task, and it’s able to do it, time and time again. But your brain only has so much context, and if you have these ambient agents out there conducting research for you, telling you in the morning who you’re meeting with — that’s a huge value-add, no matter the size of your company.

Kyle Roemer: It’s so interesting. You think about a deal process, and some of the work we’re doing with investment teams right now — you might have two to eight weeks to go through a deal, fully diligence it, build your models, and so on. You’ve got documents being added to the deal rooms or data rooms constantly. Agents should just be picking those things up and refining your outputs along the way — whether it’s your LBO model or your IC materials. I think this is the world we’re moving to, where you’ve got agents effectively listening and acting on these things. The same would be true for research — you’ve got market events happening in real time, intraday, that should be impacting how you think about a deal, your investment case, your base case, and ultimately your value creation plan.

Adam Silverman: Switching to hedge funds for a second — if you had an analyst who was incredibly good at knowing five particular equities, and you could ask him anything at any point about those five equities, he’d be able to tell you every data point on them. You can think of ambient agents the same way, except the goal is for them to do that across hundreds of equities. Any data point that comes in throughout the day goes back into context — something you can research, something that becomes a living, breathing resource, and you’re not relying on one analyst having to be up 24/7, 365. These systems are really exciting. I think we’re still in the early stages of a lot of this, but we’re starting to see the light shine through in terms of what’s actually possible.

Kyle Roemer: Adam, it’s February 2026 — what do we need to know right now about the biggest developments in AI?

Adam Silverman: One of the biggest things that happened over the past few months is Perplexity entering into a huge partnership with BlueMatrix — and I think this is going to be the first of many we’re going to see in this space. I could see Claude, OpenAI, and Gemini going to traditional research providers like Tegus or AlphaSense, or even going directly to Bloomberg, and saying, “How can we query this decade-plus of information you’ve been gathering?” This is really exciting, because finance professionals, maybe two years ago, were using LLMs and getting results that I’d call inadequate. Now, with these types of partnerships, it’s going to be exciting to see how the space moves forward, and how finance professionals can actually start trusting LLMs to provide accurate, reliable research.

Kyle Roemer: It’s an interesting partnership. Perplexity is a fantastic technology, and it’s permeating a bunch of different industries and workflows. One thing that’s interesting — we deal a lot with investment teams, and part of this is only as good as the context you provide it. Even with research, what other developments are you seeing in terms of firms focusing on the data and the history of data they actually have?

Adam Silverman: Going to an external provider like BlueMatrix is great, because you get an understanding of what’s happening in the market broadly. But a lot of firms I’ve seen have this massive trove of historical data — maybe it’s in Snowflake, maybe in legacy applications — and they’re trying to make sense of it, as great as it is to also have access to the BlueMatrix data. How do we actually make this internal data we’ve been collecting for a decade-plus useful for making investments? I think over the course of 2026, you’re going to see a lot more firms leaning in and asking, “How do we make use of this data? We’ve been spending all this money, time, and energy capturing it — let’s actually make it useful, let’s have it lead to better and faster decision-making.” Everybody says “data-backed decisions,” but you’ve got to start with the data first. That’s a piece I’m really excited to see play out over the next year.

Kyle Roemer: Look, the way we build a lot of these products, we put a lot of measures in place to make sure hallucinations don’t get in front of the user — but I’m curious what you think.

Adam Silverman: Hallucinations aren’t going anywhere. It’s not a solved problem by any means, and I don’t think either of us would be sitting here right now if we’d figured that one out completely. That said, the Perplexity–BlueMatrix partnership does highlight that if you can get access to great data and make it available to a new set of users — in this case, finance professionals — and let them query over it in a way that wasn’t previously possible, that opens up entirely new opportunities for how to leverage the data.

From a search perspective alone, it’s going to be a huge opportunity to say, “Query historical research on any type of company through BlueMatrix.” We’re seeing this very much in its infancy right now, but as you get multiple providers — and to get a bit nerdy here, the same way that with LLMs you can choose between different web search providers, you could choose Exa, you could choose a bunch of different web search providers — you’re going to have the same opportunity to say, “I want to research Bloomberg, I want to research BlueMatrix, I want to research any other external vendors,” and from a hallucination perspective, you’ll be able to say, “Here’s the variance that occurs between each one. BlueMatrix said EBITDA was $11 million, Bloomberg said $11.2 million — okay, let’s dig deeper and understand where that discrepancy actually comes from.” Being able to surface the most accurate information is one of the most important pieces here. I’m super excited about this partnership, because now we’re going to be able to surface the most accurate information possible, instead of relying on something like Yahoo Finance to pull data from.

Kyle Roemer: It almost feels like you’re at a competitive disadvantage if you’re not leveraging these tools today. It’s fascinating — there’s a lot of news right now on the compounding effects of leveraging these tools, and leveraging newer versions of them.

Adam Silverman: I’m really excited for a bunch of these tools, and to see how firms of all sizes — whether you’re a solo founder or solo person running a hedge fund, all the way to the biggest firms in the world — end up leveraging this technology.

Kyle Roemer: It feels like in most fields — hedge funds, VC, private equity, private markets generally — you can be punching way above your weight right now with some of these tools. A really small team can compete against firms with hundreds of people and hundreds of billions under management.

Adam Silverman: It’s crazy — I was at a hackathon a couple of weeks ago, and I met this guy who basically trades stocks every day using Claude Deep Research. He’s on the $200-a-month Max plan, kicking off dozens of research sessions a day, doing very in-depth financial research. He has this whole methodology for how he picks stocks and equities, and I look at him and think — if he’d gone to Wall Street in the nineties saying, “Hey, I’m this guy who has access to all this information,” he probably would’ve raised a multi-billion-dollar fund. Now he’s just some guy in Palo Alto in his basement. It’s a really interesting dynamic to watch play out.

Kyle Roemer: Yeah, super interesting — we’ll see how it evolves. Adam, it’s awesome having you on, as always. I think we covered three really interesting themes in the space right now: AI getting deeply embedded into the investment research process; the shift from agentic AI to ambient AI — AI actually running in the background for much of our lives, let alone our workflows day to day; and how it’s transforming software development. I think this is one of the most interesting areas, and we see it day to day with our own teams. Thanks so much for joining, Adam — as always, we’re excited for the next round of predictions from you.