# Stack to Street — Episode 5 Title: AI That Cites the SEC Filing Guest: Dwight Gunning (Creator of EdgarTools) Project: EdgarTools Episode page: https://www.stacktostreet.com/episodes/edgartools/ GitHub: https://github.com/dgunning/edgartools Guest site: https://www.edgartools.io/ Show: https://www.stacktostreet.com/ This is a lightly cleaned transcript of the published cut. Speaker labels and timestamps are preserved. Wording stays close to what was said. --- Dwight Gunning (00:00) If you have an alarming article that's talking about Oracle's borrowing going through the roof, it's mostly right, but a lot of it is overblown, they have been borrowing a lot, but their borrowing has slowed a bit. they're moving towards customer funded commitments up front where they receive payment from customers for capacity that they're gonna deliver in the future. and that surprised me. because it was a common story that they're borrowing a lot. Michael McLaughlin (00:34) Hello and welcome to the Stack to Street Podcast. I'm Michael McLaughlin. Stack to Street brings finance and technology into the same room to understand the problems being solved and what they mean for the people working across finance and banking. Earlier this year, a lot of reports said Oracle's borrowing was climbing fast. My guest analyzed Oracle's own SEC filings and the real story was more nuanced. His name is Dwight Gunning, a 20 year veteran of financial technology and the creator of EdgarTools. A free open source Python library for SEC filings with more than 9 million downloads. We get into how he uses EdgarTools to check a big AI headline against the actual filings, and how you can also trace the hidden money trail between companies like the circular deals driving the AI boom. Here's my conversation with Dwight Gunning. Michael McLaughlin (01:23) Welcome Dwight. It's really good to see you. hoping you're having a nice day today. Dwight Gunning (01:26) Well, I'm having a great day actually. Michael McLaughlin (01:27) So I want to talk about EdgarTools, and before we start, I thought I would ask you about something that caught my attention on your website you say that corporate reporting is fascinating but the tools that work with it were not fascinating. what's so fascinating about corporate reporting to you? Dwight Gunning (01:41) I read an essay that spoke about two trillion dollars worth of commitments that were held by the cloud providers. Now that's an interesting story, but I guess the question is how do you even get that from corporate reporting? How do you even get that story together? So the whole point of EdgarTools is to be able for you to pull that story together from Python. The alternative would be months of digging. Michael McLaughlin (02:04) Take me back to what started EdgarTools. What was the beginning of it for you? Dwight Gunning (02:08) I worked at FINRA which is if you know FINRA it's a sister company to the SEC. So I'm sort of working on SEC related projects at FINRA and my curiosity was really about can I write a tool that in real time, as I'm asking the question, it will go and grab it from the SEC, find the right filing, and then pull the information for me. So I really wanted an API that just spoke for what I wanted. I ask and I get. the core of EdgarTools was something like you ask for a company by ticker, it finds the company, you ask for the latest filings, it goes and finds those filings, and then you could ask for little things out of those filings. it was a query ask paradigm that was interesting to me. And there was no other tool that did that little thing that I wanted. Michael McLaughlin (02:55) Was there a specific moment that led you to thinking that this was worth pursuing Dwight Gunning (03:00) In the first few months, you have an idea and you put it out there as open source and as the usage grows you can see what people are actually using it for. when I started in 2022, ChatGPT had just come out, and so we hadn't gotten used to the idea of you open a chat and you just ask a question and the AI will then just tell you what the answer is. And that was the design or pattern that I had with EdgarTools. But that pattern of I'm curious about something, tell me the story, what's there. That's what I found interesting. Michael McLaughlin (03:29) Can you give me an example of what an analyst ought to be able to do easily with the SEC database, but currently can't. Dwight Gunning (03:35) I think you're familiar with big BDCs or Business Development Corporations. Michael McLaughlin (03:38) Sure. Dwight Gunning (03:40) I think you've worked in the space a little bit. So BDCs report in their 10-Q's and so there's the tables in the 10-Q's. and they file at different times of the year. So you might want to ask questions across all BDCs, right? Or you want to find a specific BDC. Without EdgarTools or without a specific tool, you have to actually open a 10-Q you read it And that takes you half an afternoon. but EdgarTools, it aggregates all of that for you. So it's a library that's you can ask, tell me all the BDCs that exist, and tell me that specific BDC and what is the schedule of investments. So that's one use case and an analyst would take. Maybe days to figure that out. But with EdgarTools if you actually have a development team that is comfortable with Python, you can easily get it in a few lines. Michael McLaughlin (04:27) We've had XBRL for years now. and since you were at FINRA I presume you were using the SEC tools before they had XBRL. Has that really improved things? Dwight Gunning (04:36) XBRL has massively improved things. I think as of 2026, most or almost all companies have migrated onto XBR reporting. and it's really solved this the issue of putting it in a standard format. but if you're a contributor on EdgarTools, that's just a starting point. Let's say there are 7,000 companies. They use a standard of XBRL, but each company reports it slightly differently, right? So there is the huge, huge problem of how do you standardize it across industries, Intel versus an accounting company. Michael McLaughlin (05:15) Tell me your thought process to develop EdgarTools as an open source library as opposed to a proprietary piece of software. Dwight Gunning (05:22) I made it open source because it was my idea that I wanted to contribute to the world to say, okay, can you have interactive way of pulling data? Can we write a really convenient API and then see if people wanted to use it and see if people wanted to contribute to it. So that's where it had to be open source. within Closed source technology, you don't have the evolution, you don't have the contribution, you don't have the knowledge that's shared. So what's happened over the years, over the last few years, is that there's a big rise in the number of people are contributing to EdgarTools. I started with like three in 2022, it went up to 14 and I think 30 something in last year, but this year it's over 50 contributors to it to EdgarTools. And each one is bringing a unique part of what their knowledge is. And important, they also bring their agents, they're also bringing AI to actually look into the XBRL space, for example, and scan for better ways to organize information. So we're getting a huge rise in the PRs. Michael McLaughlin (06:22) When you look across filings, what are some of the things that investors can easily miss when they're looking at tons and tons of data. Dwight Gunning (06:28) If you're looking at the entire 10-Q, right? Or a 10-K, you are looking at a large volume of either text or what the companies wanting to tell you about the company. but what you can do is corroborate it by looking at the XBRL or looking at data, the financial data. you can't really interrogate, you can't really ask how does this compare with either the peers, how does it compare with what the company did before? If you're looking at one specific filing, where's the other filing? How do I compare it? Is there growth, right? What's happened before EdgarTools or maybe the tools like this is point to point. Like but what you can miss is how does it all come together and how you can query across everything to find trends. And that's where the library works. So the other thing it does is sort of cuts down on the amount of information that you actually have to look at. There's a lot of stuff in the 10-K or 10-Q that a typical person or a CIO might not know how to read or what's important, what is this form? So instead of having to know what a DEF 14A is, and a lot of people do, but some people don't, you can ask the question is who's on the board or what were the votes for? it's really an intelligence or knowledge layer that you don't have to know what a specific filing is. You don't have to know what a previous filing was, you don't have to know what the peers are. Michael McLaughlin (07:47) If you could go back to the two point three trillion dollar claim where the cloud providers have massive amount of backlog. I think what I like about that example is that you took a headline and decided to validate it using data. so tell us more about that. I think that's a really good story. Dwight Gunning (08:03) The genesis of that was actually an essay that I found about the current AI boom, and I just wanted to see can my tool actually corroborate the story. I had a Claude Code session, I gave it the article and say go corroborate it. use EdgarTools to find that information and pull it together. in the beginning it was human intelligence And right now I'm asking my agents to go look for and pull information from all the filings. And pull it together and then tell me if the story is right. Michael McLaughlin (08:30) Yeah, and you have the two point three six trillion contracted backlog spread across Oracle, Microsoft, Google, and Amazon. And I didn't count the lines of Python code, but it looks to be about twenty, so that's pretty powerful. Let me ask you, have you come across anything that genuinely surprised you, maybe on the other side, where you looked at something that was being held as a truth, but then you did the analysis and you couldn't corroborate it? Dwight Gunning (08:53) If you have an alarming article that's talking about Oracle's borrowing going through the roof, it's mostly right, but a lot of it is overblown, right? they have been borrowing a lot, but their borrowing has slowed a bit. they're moving towards customer funded commitments up front where they receive payment from customers for capacity that they're gonna deliver in the future. so they're not borrowing as much, but they're making use of that facility. and that surprised me. because it was a common story that they're borrowing a lot. Michael McLaughlin (09:18) You might be able to trace the circular economy that exists in the AI industry right now where the chip manufacturers are providing seller financing to the frontier companies. And then the frontier Dwight Gunning (09:29) Yeah. Michael McLaughlin (09:30) Companies are buying the chip companies products. Dwight Gunning (09:32) One thing I've thought about is can you find linkages in the filings, for example, a specific company, who are their suppliers, who's buying from them, how is the economy linked in that way? with AI now you can actually have like workflows that actually go find that for you. So if you know what questions to ask and you can spend an afternoon of compute. Yeah, you can get really interesting linkages between companies in a way that you've never had before because previously it was one individual filing. How does it link to another filing? No one knew, right? So with both the tooling has gotten better and the AI now you can actually use that to go find stuff. so that's where I'm going with EdgarTools. It's a newer world, I guess. Michael McLaughlin (10:15) So for someone running an investment bank or an investment management firm, it seems like the information is becoming much easier to collect and analyze. what changes do you think that those businesses are gonna have to go through in order to maximize the biggest opportunities? Dwight Gunning (10:29) Information becoming more I guess easier to analyze. you've pulled more information and you've had the AI summarize it, and put it in a report and gives you numbers. How do you know if it's correct, If there's an SEC filing, can I automatically verify? Can I use AI to make sure that it's accurate? So businesses are gonna be a lot more interested in verification and making sure the information is correct. If you're not using AI, you're not using the tooling, but your competitors are, then you're at a disadvantage as well, So that's something you have to think about. Michael McLaughlin (10:58) So you've been in financial technology for a long time, Dwight. If you look ahead three, four, five years, what do you think will be the most different about how investors use financial information right now? Dwight Gunning (11:09) Three years from now you will have an AI agent investor. So a user sets a goal and the AI agent does the work for them. So the AI agent now has to collect that information, reason over it and then do the investing. the reasoning capability of these models I think will surpass what humans are capable of. Michael McLaughlin (11:30) What are you most excited about working on next? Dwight Gunning (11:32) I'm trying to expand outside of the United States as well. the SEC universe is a really big thing, but I mean the world is also big. I'm based in Canada as well. So I'm looking at Canadian companies and so on. I was really interested in current tensions in the tariffs. So can I then ask a question about it and figure out where it's going and then can the information come out of the SEC filings for me. Michael McLaughlin (11:50) For listeners who want to learn more about your work, where's the best place for them to go? Dwight Gunning (11:54) If you go to GitHub, you can search for EdgarTools. it's on GitHub under dgunning, that's my name. So you can find it. It's quite a popular library. 2800 stars, that's the first thing. You can look at it, you can contribute to it. I encourage you to get your dev teams to start working with it. It's it will save you time. I also have edgar.tools, which is my commercial offering, which gives a little bit more than what you can get in the open source library. There's a lot of processing that you can't do in the open source library. And and of course if you don't have a development team, that's where you go. and then I also blog on edgartools.io. Michael McLaughlin (12:32) I've really enjoyed the conversation, Dwight. thanks very much for joining me on Stack to Street. Dwight Gunning (12:35) Thanks for inviting me, It was an interesting conversation. I was happy to be here.