Full episode transcript: AI & PE: The Future of Value Creation
Guest: Adam Silverman, Senior Director, Accordion Intelligence Lab | Host: Kyle Roemer
Adam Silverman: In 2026, if you’re not connecting your data warehouse to AI, you are not competing against the rest of the pack.
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 kicks off a new series of short, sharp conversations — starting with Adam Silverman, one of the leaders of the Accordion Intelligence Lab. Adam has been building custom AI tools since before most people knew what an AI agent was, and more importantly, he operates at the rare intersection of deep, hands-on AI expertise and the real practical needs of private equity. In this episode, Adam looks ahead to 2026 and what PE leaders actually need to know about AI right now. Let’s get into it.
Adam, before we get into predictions — you spend a lot of time working with leaders who are trying to turn AI ambition into real outcomes. I think you’ve looked at over 400 different types of agents and technologies out there. Where are we right now?
Adam Silverman: I think we’re at a point now where anybody — doesn’t matter if you’re a business user, doesn’t matter what level of the organization you’re in — can actually start leveraging AI to automate a lot of the work that you do. Two or three years ago, when we started building agents, it was something almost exclusively for engineers, or types of products that were a lot more technical in nature. We’ve gotten to a point now where just about anybody can get started building really reliable agents and start automating a lot of the work they do every day.
Kyle Roemer: I love that. It’s so interesting — the use cases have proliferated, the technology’s gotten a lot better, LLMs are getting bigger.
Adam Silverman: I’ve seen how these platforms can actually impact businesses all the way from the ground up. One of my biggest recommendations — and it’s worth sticking through the entire episode for — is understanding how your company can start using AI agents, and figuring out ways, even if they’re small today, to start incorporating AI both in your personal life and in the workflows you do every day.
Kyle Roemer: If you think about what you’re forecasting for this year — what’s one thing where you’re like, “this is going to be a real shift,” whether in adoption, types of use cases, or types of activity?
Adam Silverman: Every analyst, every associate within private equity firms is going to be using agents. A couple years ago, when we started talking to these firms about the work they were doing, it was kind of hard to actually pull people away from their job to go build these types of products. Now, with tools like Claude Code and Claude Cowork, just about anybody can get started building incredible automations, and can almost have a virtual assistant or copilot with them twenty-four-seven that automates a lot of the busy work they’re doing every day.
Kyle Roemer: It’s super interesting — a lot of mundane, manual activities can now be handled by agents, so you end up with a bunch of agents just operating in your general workflow. Where do you think things go in terms of — there’s getting answers to stuff, right? We’re used to a chat interface, we’re used to dealing with agents — but agents actually taking action. Do you think there’s going to be a leap there this year, or are we there already?
Adam Silverman: I think in ’26 we’ll see agents shift from answering to acting. So instead of just saying, “Here’s a response, here’s a PDF of a report we did,” it might actually go take action on that report, and figure out the next step needed to complete a goal. Using Cowork now, since it released, it’s incredible to see how you can start taking action directly within Claude. Previously, if I wanted to build an automation, I’d have to go to a platform like Zapier, link together all these little nodes, and try to understand the type of automation I wanted to build. Now, in natural language, I just talk to my computer using Whisper Flow. I describe the type of automation I want, and it’s automatically built for me — it logs into n8n and comes up with a full plan for connecting everything.
It’s taking a lot of the thought process out of building these automations, which is exciting. I think now every level of the organization is going to feel this — it’s just a matter of teams wanting to apply this technology, leaning into how they can use it. That task list everyone has — “here are the things I want to get done this week” — the best thing I recommend is people just put that directly into Claude or one of these other LLMs and see how far they can get without writing a single line of code.
Kyle Roemer: I find it so interesting — if you haven’t been around Adam in person, the amount of time he spends talking to his computer is super interesting. I think that’s the paradigm we’re going to be living in: voice-activated engagement, versus what we’re used to, which is just typing.
Adam Silverman: For sure. I don’t type anymore. If you get an email from me, it’s likely dictated completely.
Kyle Roemer: Adam, what do you mean you don’t type anymore?
Adam Silverman: I use Whisper Flow — there are a bunch of different tools people can use — but now I act at the speed of my thoughts, which is kind of amazing to think through. With Whisper Flow, I can really just talk to my computer. When it comes to prompting, I talk to ChatGPT, I talk to Claude, I talk to Perplexity, and I get results right away. I think this is a big paradigm shift in ’26 — you don’t necessarily need to be typing anymore. Typing is an old paradigm, in the same way people went from handwriting to typing and that was a huge innovation. I think the big innovation in ’26 and beyond is that people are going to stop typing completely, or at least limit how much they’re doing it. I still use typing when I want to refine what I’ve dictated, but you can get very far just through dictation, using these LLMs to analyze what you’re saying, consolidate it, and structure it.
My favorite tool right now for that is Whisper Flow, but I recommend all firms think about how people are going to be interacting with their computers going into this year. One of the cool things about Whisper Flow is you can actually whisper to your computer — you don’t need to be screaming at it. If you’re in a busy, open-concept office, people can still be dictating to their computer without disrupting anyone. I think it’s beautiful, because now you can act at the same speed as your thoughts. Previously, thinking about your fingers as actuators trying to type, you can maybe do 50, 60, 70 words a minute if you’re fast. With dictation, you can do 150 to 200 words a minute, which really sets you apart from everybody else.
Kyle Roemer: I love that — it’s so interesting, and we’ll see how it evolves. What do you believe is going to matter most in this next wave of tooling and products?
Adam Silverman: I hope 2026 is the year no one has to make an Excel model or PowerPoint ever again. I’ll go bold with that — that’s how I truly feel.
Kyle Roemer: I love that.
Adam Silverman: I think a lot of the busy work around making PowerPoints and Excel files will get automated, and people are going to have to actually use their brain to come up with interesting ideas around the models and the presentations, because a lot of the busy work of building the inputs and outputs will get automated. I never want to see a PowerPoint again — I hate making them. I’m hoping ’26 is the year agents will be able to fully make PowerPoints. There are a lot of great companies doing this — Gamma, Canva is obviously making big strides here — and natively integrating into PowerPoint itself. Google’s been doing immense work on this on the Slides front, and now I think it’s time for PowerPoint to really level up and figure out all the formatting things.
Before a deck goes out to an MD to review — spell-check the whole thing, make sure the fonts are correct, the grammar’s correct, the spacing’s correct. These are the little things that take so much time away from your analysts and associates actually doing real work, and this is stuff that should be automated. There’s no reason humans should still be manually reviewing these types of things.
Kyle Roemer: I totally agree. You’d think the age of slideware — having to put these decks together — would be behind us. It’s not, at all, because a lot of the tools just aren’t quite there yet. But I’m hopeful ’26 is the year slideware is one of the final frontiers of adoption in the enterprise — which is a funny thing to say, but it’s the reality we’re in right now.
Adam Silverman: The funnier part is that I think there’s a lot from the previous wave that a lot of the companies we’re speaking to haven’t even fully adopted yet. Products like Whisper Flow and Superhuman — these are day-zero, every-company-needs-to-be-using-them tools. They make you more productive in the work you do. I think a lot of the integrations into legacy platforms are going to be the biggest winners in ’26.
As I think through Claude for Excel — it’s insane. I’ll take a screenshot of a financial model and it’ll go recreate it automatically. That’s something you would have had an analyst go do manually, and now it’s completely automated — when it works in front of you, it’s mind-blowing to watch. I think a lot of that busy work is going to get automated this year. I don’t think Excel’s going anywhere in 2026, but the way people use Excel is going to change completely with AI-native plug-ins.
Kyle Roemer: Where do you see AI ambition either succeeding or stalling this year with finance teams?
Adam Silverman: I think you’re going to have Luddites on one side who just refuse to adopt new technology, and then you’re going to have innovative firms that run laps around them — you won’t even be competing in the same field anymore. You’ll have super-lean, AI-native teams using Claude for Excel, building models faster, using custom agents they built to get outcomes a lot quicker. And then you’ll have the Luddite firms still, for lack of a better phrase, going back to accounting basics — using T-charts, adding up totals by hand.
I think you’re going to see breakout firms — maybe firms that had trouble raising funds — now running super-lean shops of ten-plus people that lap firms with a hundred people, just because everyone’s leaning into AI-native from day zero. Honestly, a lot of these firms are in a lucky position — if you don’t have legacy technology to rip out, you don’t have to retrain everybody or manage change on what people are currently doing. You’re actually in a very advantageous spot. I think it’s going to be really interesting to see how these vintages compare, when you have AI-native firms saying, from day zero, “here’s every AI tool possible, our AI software budget is the same as what we’re paying employees.” I think the clearest change going into ’26 is that there are going to be these breakout winners.
I’m excited to see how they emerge, what tech they’re using, and how they’re able to automate a lot of work that even a year or two ago people would have thought was un-automatable.
Kyle Roemer: It’s so interesting — speaking on the private equity side, newer funds are starting out asking, “I’m going to have a team of agents and a few really experienced investors, and that’s going to be our strategy,” versus the traditional model. That relearning behavior, the change management piece — that’s tough. But I’m a believer that this year, Excel-first work should be heavily augmented with AI at a minimum, and you’re going to find a bunch of firms building these things more automatically with agents.
Adam Silverman: ’25 made it very clear that research could be done with agents, and I think financial research is something that’s going to be completely automated. You’ll still have a person who takes ownership — similar to an engineer using Claude Code to develop a codebase, you still need an engineer who reviews it and says, “this is good enough for us to deploy into production.” You’re still going to have a research person who has to look through the research, make sure everything’s accurate — sourcing information, pulling relevant facts, doing overviews — but the agents are just so much better at the legwork. I kick off Claude Code or Claude Deep Research multiple times a day, and I see it accessing 400 to 1,000 websites sometimes.
You think about a human analyst spending their entire day sifting through all those websites — now, within 15 minutes, you can get a report that cites everything and tells you exactly where the data is from. The cost of knowledge is coming down so much that I’m excited to see how this pushes the frontier of what’s possible.
Kyle Roemer: I love that. In a lot of ways, we’re becoming quality control, almost like architects, versus the work we traditionally would have done — which is super cool. One of the things that’s existed forever, Adam, is companies building dashboards on top of their data, with the data living everywhere. What do you see now with the models and tools that sit on top of these platforms?
Adam Silverman: In 2026, if you’re not connecting your data warehouse to AI, you’re not competing against the rest of the pack. One way to think about it — using a computer analogy — is short-term memory, like RAM. When you try these models out, you’ll notice they sometimes run into limitations around the context window and how much they can pull in. I think in ’26 we’ll see a lot more progress here — I’ve watched this evolve over the past three years, working out how to bring almost a solid-state-drive’s or hard-drive’s worth of data into the memory of these agents. That’s really important for a lot of firms that built out data lakes and data warehouses over the past few years and have been storing all this data.
It’s great to store the data, great to have a Tableau dashboard showing what’s going on — but what’s the next evolution? To me, it’s really about chatting with your data, with a high degree of accuracy, and being able to search through terabytes of historical data you’ve been collecting. I think in ’26 we’ll see the breakout of, “I just gave it an entire data room, or multiple data rooms, and now I can actually get answers from it” — something that wasn’t possible before. When ChatGPT first came out, everyone was trying to do this, and now we’re getting to the point where you can give it a hundred or a thousand documents and, with the right custom solutions, get a high degree of accuracy pulling information out of it. I think we’re at the cusp of this — going into Q2, Q3, there’s going to be a lot of innovation here, and I’m excited to support as much of it as I can.
Kyle Roemer: I’m excited about that innovation too. We sometimes feel limited by context windows, but what we’re building now gives you access to — and accuracy over — tons of historical data, which is super interesting.
Adam Silverman: One of the most important things in 2026 is going to be making use of all the data you’ve been collecting over the past decade, and being able to query it and answer questions based on it. That’s going to be extremely important for firms trying to make data-backed decisions using their historical data. I’m very confident 2026 is the year every company starts using agents in one way or another — whether at the customer-service level, the financial-prediction level, or in legal teams. Across entire organizations, people are going to see the productivity and throughput benefits of using these tools. I really recommend that anybody listening just go give them a try.
The worst thing you can do right now is sit on the sidelines, scared of this technology and how it’s going to impact your work. One of the best things you can do is go download whatever tool your friend in tech told you to try a couple weeks ago — sign up, try as many free trials as possible, and see what can help your workflow. I highly recommend giving it a shot and seeing how much your day can improve by using this technology. If you’re not using AI, your competitor is, and they’re going to leapfrog you completely. Something I’ve talked about publicly before is that I think there’s an opportunity for a “tenth player” in a given space.
Say you take the mortgage industry — the tenth player in that space could leapfrog to number one or two just by going AI-native from day zero. Hopefully, the big players are making huge AI pushes, and I’ve seen some of that — but if the second-, third-, or fourth-largest players aren’t leaning in completely on AI, that’s genuinely something to worry about. It leaves a huge opening for competitors to leapfrog the leaders in the space.
Kyle Roemer: Huge thanks to Adam Silverman for breaking down what he’s seeing going into 2026. If you’re listening and thinking “I’m behind” — well, you’re not alone. The good news is the entry point is simple: try a few tools, focus on automating the busy work first, and make sure you’re doing it safely. And the big unlock we kept coming back to — your historical data only becomes an advantage when you can actually query it. If you found this useful, be sure to subscribe to the show and share this episode with a colleague. Adam will be back in future episodes to keep breaking down what’s changing in the AI landscape, and what that means for PE.