Invest with AI

Jack Meehan on Automating Sell-Side Research with AI and the Bifurcation of Analyst Roles

1h 6m Listen to the original
AI AutomationSell-Side ResearchOperational RiskHedge FundsData ProductsCareer Advice

Summary

Jack Meehan, founder of Operon Research, discusses how he has automated approximately 70 days of work annually by integrating AI tools like Claude and Opus into his sell-side research workflow. He argues that while automation handles data processing, the future of sell-side analysis lies in 'relational moats' and primary research that AI cannot replicate. Meehan predicts a bifurcation in the industry, with top analysts moving to independent models while bulge bracket banks focus on corporate access. He also highlights the operational risks of AI adoption in hedge funds and the emergence of AI-native firms aiming to replicate large teams with small staffs.

Key takeaways

  1. 1
    Speaker_1 warns that hedge funds are 'fragile beings' where small drawdowns can lead to immediate redemption, creating an environment hostile to the experimentation required for AI adoption.

    Unidentified speaker (speaker_1) · 56:34

Episode sections

  1. 0:00 – 0:45

    Intro and Productivity Gains

    Jack Meehan states he has automated approximately 70 days of work per year. He believes the industry is at a threshold moment where AI adoption will shift dramatically, noting that some funds claiming full AI adoption should be questioned by their LPs.

  2. 0:45 – 3:10

    Operon Research and Career Background

    Brett introduces Jack Meehan, founder of Operon Research, which applies an AI-driven approach to sell-side research. Jack recounts his previous roles at Barclays and Nephron Research, covering life science tools and diagnostics.

  3. 3:10 – 12:34

    Sell-Side Analyst Workflow and Value Add

    Jack explains that sell-side analysts add value through deep domain expertise and primary research (surveys, KOL checks, management calls) rather than just model building. He emphasizes the importance of field research and conferences for gathering non-obvious insights.

  4. 12:34 – 16:54

    Automation and the NIH Dataset

    Jack describes automating the NIH Reporter dataset using Claude after a friend's suggestion. This automation transformed a once-a-year manual process into a weekly product called 'Operon Signal,' allowing for real-time data analysis and freeing up significant time.

  5. 16:54 – 22:57

    Data Verification and Model Accuracy

    To ensure data accuracy, Jack uses spot-checking and multi-source verification (e.g., comparing external providers with direct SEC filings). He notes that while AI tools like Sonnet, Opus, and Fables were used, rigorous validation loops are essential to prevent silent errors.

  6. 22:57 – 27:29

    Earnings Season Automation

    Jack details a new automated process for earnings season where an app parses documents and updates models within minutes. This allows him to handle multiple simultaneous earnings reports and produce first-look reports quickly, freeing time for deeper analysis.

  7. 27:29 – 29:12

    Insights from Automated Efficiency

    Jack provides an example where automation allowed him to notice a significant drop in 'cost per test' for a customer during earnings season. This insight led to a specific question during a management callback, revealing a platform conversion that would further reduce costs.

  8. 29:12 – 30:46

    Operon Knowledge Hive Mind

    Jack Meehan describes the 'Operon Knowledge Hive Mind,' a centralized data structure where insights from published reports, management transcripts, and extracted financial numbers are aggregated. This system allows for faster retrieval of company-specific data during previews or reviews, though he notes the process is still evolving.

  9. 30:46 – 30:47

    Preserving Analyst Judgment

    The discussion touches on ensuring AI supports rather than replaces analyst judgment. Jack argues that automation reduces drudgery, allowing analysts to focus on high-value tasks like identifying signals and engaging in relational alpha generation.

  10. 30:47 – 33:19

    Client AI Adoption and Operon's Infrastructure

    Meehan estimates that one-third of his clients use AI beyond basic chat features, with 5-10% being 'bleeding edge' users with dedicated hardware. Operon Research currently operates two dedicated AI computers named 'Batos' and 'Xyros' to handle continuous data processing and platform building.

  11. 33:19 – 37:25

    ROI of AI and Model Thresholds

    Brett notes that early adopters saw terrible ROI on time and token spend until recently. He identifies 'Opus 5.5' (released September 22) as a threshold moment where validation loops improved accuracy to 99%, making rigorous AI workflows viable. Meehan agrees, citing improvements from Claude to Opus to 'Fable'.

  12. 37:25 – 41:53

    Sell-Side Bifurcation and Relational Moats

    Meehan predicts the sell-side will bifurcate, with top analysts moving to cutting-edge independent models while bulge bracket banks focus on corporate access and capital markets. Brett argues that 'relational moats' built through human networks remain durable and hard to replicate by AI.

  13. 41:53 – 50:17

    Data Products and Expert Networks

    Operon is building a data product to launch commercially mid-year, aiming to provide quality control and benchmarking for multi-managers. The plan involves an MCP connector for Claude/OpenAI and an API ('Opera') to allow clients to ingest curated healthcare data directly into their workflows.

  14. 50:17 – 1:06:06

    Junior Analyst Roles and AI Adaptation

    Meehan advises junior analysts to focus on adding value to their team rather than performing grunt work, which is now automated. Brett suggests that junior talent will now earn their roles by being 'agent-pilled' and accelerating senior analysts' output, potentially allowing them to cover stocks faster than in the past.

Claims worth verifying

Jack Meehan has automated about 70 days of his year, meaning tasks that took 70 days last year are now fully automated.

Unidentified speaker (speaker_0) ·0:00

How to verify: Ask Jack Meehan or Operon Research for a breakdown of time saved; verify against his stated workload and product output frequency. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan believes that if a fund claimed last year they were doing everything with AI, one should counsel LPs to claw back their capital.

Unidentified speaker (UNKNOWN) ·0:24

How to verify: Check the podcast transcript for this specific advice; verify if any hedge funds made such claims in 2026/2026. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan started working with Claude in February to automate data feeds and research processes.

Unidentified speaker (UNKNOWN) ·2:51

How to verify: Verify via Operon Research product launch dates or press releases from early 2026 regarding their AI adoption timeline. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan covered life science tools and diagnostics at Barclays until 2020, then joined Nephron Research for six years.

Unidentified speaker (UNKNOWN) ·3:57

How to verify: Check LinkedIn profiles for Jack Meehan, Barclays press releases, and Nephron Research team pages. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Operon Research hosts its own conference called the Operon Summit in Cape Cod in May.

Unidentified speaker (speaker_0) ·12:11

How to verify: Search for 'Operon Summit Cape Cod' on event platforms or Operon Research's website for May 2026. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan automated the NIH Reporter dataset using Claude, turning a once-a-year update into a weekly product called Operon Signal.

Unidentified speaker (UNKNOWN) ·14:44

How to verify: Check NIH.gov for the NIH Reporter dataset update frequency; verify Operon Signal publication schedule on their website. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Operon uses two sources for financial data: an external provider and direct SEC filings, which they spot-check against each other.

Unidentified speaker (UNKNOWN) ·20:33

How to verify: Review Operon Research methodology documents or whitepapers on their 'Beacon' data portal. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Primary source

Meehan launched Operon Research on July 1, having left his prior firm on May 1 and come out of hiding on June 1.

Unidentified speaker (UNKNOWN) ·21:50

How to verify: Check SEC filings for any new entity registrations or press releases from Operon Research dated May-July 2026. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Primary source

During Q2 earnings season, Meehan published five client surveys immediately after earnings wrapped up, a feat he says was unheard of previously.

Unidentified speaker (speaker_0) ·24:58

How to verify: Check client communications or press releases from Operon Research post-Q2 2026 earnings season. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan attended the ASHG conference in Montreal and hosted interviews with Roche and Niall Lennon from the Broad Institute.

Unidentified speaker (UNKNOWN) ·7:38

How to verify: Check ASHG conference schedules for 2026 and Operon Research's social media or website for recorded interviews. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Jack Meehan estimates that about 70 days of work he used to do by hand are now handled by AI.

Unidentified speaker (speaker_0) ·29:12

How to verify: Podcast transcript description; verify via Meehan's public statements or Operon Research website. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Operon Research has two dedicated AI computers named Batos and Xyros.

Unidentified speaker (speaker_0) ·31:39

How to verify: Podcast transcript; verify via Operon Research website or press releases. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Opus 5.5 was released on September 22.

Unidentified speaker (UNKNOWN) ·36:44

How to verify: Podcast transcript; verify via OpenAI's official announcements or blog posts for that date. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan spent $10,000 on tokens personally for Fable over the summer.

Unidentified speaker (UNKNOWN) ·36:41

How to verify: Podcast transcript; verify via Meehan's public statements or financial disclosures if available. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

One-third of Meehan's clients use AI beyond the chat feature.

Unidentified speaker (UNKNOWN) ·30:47

How to verify: Podcast transcript; verify via Meehan's public statements or client surveys. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

5-10% of Meehan's clients are 'bleeding edge' users with dedicated AI computers.

Unidentified speaker (speaker_0) ·31:15

How to verify: Podcast transcript; verify via Meehan's public statements or client surveys. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan's CEO client increased productivity 40x using Grokbot.

Unidentified speaker (speaker_0) ·54:10

How to verify: Podcast transcript; verify via the CEO's public statements or company announcements. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Operon Research plans to launch its data product commercially mid-year.

Unidentified speaker (speaker_0) ·41:53

How to verify: Podcast transcript; verify via Operon Research website or press releases. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Meehan previously worked at Barclays and Nephron Research.

Unidentified speaker (speaker_0) ·29:12

How to verify: Podcast transcript; verify via LinkedIn profiles or SEC filings for his previous roles. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Primary source

Meehan's associate Jesse is supporting the research process and helping with the data product.

Unidentified speaker (speaker_0) ·51:04

How to verify: Podcast transcript; verify via Operon Research website or press releases. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Hedge funds are fragile; a drawdown of 3-5% at multi-managers leads to immediate redemption ('pass go, do not collect').

Unidentified speaker (speaker_1) ·56:34

How to verify: Verify via hedge fund industry literature on redemption clauses and multi-manager fund structures; check SEC filings of major multi-managers for redemption terms. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Primary source

Tiger Cub systems were powerful in the 2010s but occasionally went off the rails.

Unidentified speaker (speaker_1) ·57:14

How to verify: Search historical financial news archives for 'Tiger Cub' and 'quantum mental exercise' failures in the 2010s. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Speaker_1 ran a mock healthcare portfolio and encountered many issues with AI implementation.

Unidentified speaker (speaker_1) ·57:25

How to verify: This is a personal anecdote; verify by checking Jack Meehan's public writings or LinkedIn posts around late 2026 for mentions of mock portfolio testing. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Speaker_1 advises LPs to withdraw capital from funds claiming to do 'everything with AI' last year.

Unidentified speaker (speaker_1) ·58:42

How to verify: This is a subjective opinion/claim; verify by checking if any fund actually made such claims in 2026/2026 press releases or LP letters. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

New AI-native funds aim to replicate the output of 40 people with 4 people.

Unidentified speaker (speaker_1) ·59:25

How to verify: Verify by identifying new hedge fund launches in 2026 with small headcounts and AI-centric marketing materials. For recent events, verify 2026-08-20 through 2026-11-20. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify the release date and capabilities of 'Opus 5.5' and 'Fable' models to assess claims of 99% accuracy and decimal point precision.
  • Check SEC filings for Operon Research's DDQ responses regarding AI usage and data security protocols.
  • Review the NIH Reporter dataset structure to understand the scope of automation described for 'Operon Signal'.
  • Investigate the operational history of 'Tiger Cub' and other past AI/quantum systems in asset management to evaluate the 'cognitive debt' warning.
  • Assess the current AI adoption rates and infrastructure investments of major bulge bracket banks versus independent research firms.
  • Verify the existence and specific functionality of 'Grokbot' and its reported file deletion incidents in professional environments.
  • Look for recent hiring trends at Operon Research and similar AI-native firms to validate the claim of small teams replacing large analyst groups.
Full transcriptexpand

0:00Unidentified speaker (speaker_0): I think so far I've automated about 70 days of my year. Like things that took 70 days last year are now fully automated. My

0:07Unidentified speaker (speaker_1): strong belief now is we're at a threshold moment where the process will start to shift dramatically.

0:12Unidentified speaker (speaker_0): He told me in the last two weeks, he's increased his productivity 40x. Like

0:18Unidentified speaker (speaker_1): if you were a fund last year that said we're doing everything with AI, I would like counsel your LPs to

0:24Unidentified speaker (UNKNOWN): get

0:24Unidentified speaker (speaker_1): claw

0:24Unidentified speaker (UNKNOWN): all

0:24Unidentified speaker (speaker_1): their capital.

0:25Unidentified speaker (speaker_0): If you told me in January I would be doing what I'm doing today, I would have said you're crazy.

0:29Unidentified speaker (speaker_1): So the biggest transformation I've seen in my career. Of Artificial Intelligence and Fundamental Research. I'm really excited to have Jack Mehan here with me today. Jack is the founder,

0:54Unidentified speaker (speaker_2): co founder, and analyst at Opera on Research, which is taking a fresh AI pilled approach to sell side research. I first met Jack probably in 2010 or 2011, where he helped teach me about life science tools, stocks, and was always one of my first calls when I had a question about thermal. Thermo Fisher or Danaher. And so you know we've stayed in touch and been excited to see what he's built with Opera on. So so uh great, thanks for being with us today, Jack.

1:24Unidentified speaker (speaker_0): Yeah, thanks for having me, Brett. Uh really loved following you on Twitter and seeing everything you're up to because and glad that we could reconnect because everything that we're up to over here in terms of leveraging AI on the research process, it's really exciting to see you. know, like how a former client is putting this to work and kind of the tools that the buy side is using, because at the end of the day, that's who we're trying to sell into and kind of keep up to speed with them. So uh yeah, same to you.

1:51Unidentified speaker (speaker_2): Yeah, it's um it's a hard question to figure out how the the buy side will shift over the next 18 eighteen months. My strong belief now is we're at a threshold moment where process will start to shift dramatically and

2:05Unidentified speaker (UNKNOWN): yeah,

2:05Unidentified speaker (speaker_2): how does the cell side shift to

2:07Unidentified speaker (speaker_0): right

2:08Unidentified speaker (speaker_2): uh to sort of adopt adopt that uh adapt to that change? Uh

2:12Unidentified speaker (speaker_0): yeah, totally. Yeah. If you had told me back in January that my entire process would be turned upside down and I would be starting my own research firm built around AI at its backbone, I would have said you're crazy. And you know, because this job, there's so much drudgery to it. And I love it. You know, like when you're able to find a good nugget and, you know, come up with a trade idea for a client, that's what this is all about. Like you, you know, you're not getting paid for the actual calls on the sell side, but it's kind of for the glory. Um, and uh, but you know, it really started in February when we started working with Claude and just started automating everything. everything

2:51Unidentified speaker (UNKNOWN): in

2:52Unidentified speaker (speaker_0): terms of a lot of the data feeds that we look at. And then, you know, in terms of like the process, you know, I think a lot about just like how we build things, how we structure them, why we're building them the way that we do. Every part of my research process is totally changed since the beginning of the year. So it's really exciting.

3:10Unidentified speaker (speaker_2): Maybe let's start, maybe this does take it back um before we. get into the the AI agent pill part of the the Operone story,

3:20Unidentified speaker (UNKNOWN): you

3:20Unidentified speaker (speaker_2): know, because some readers will know you know intimately, some listeners will know intimately what a sell side research analyst does, but many don't. What what does a sell side research analyst do from a work stream perspective? What did that look like early in your career and how did that shift and evolve over over time?

3:39Unidentified speaker (speaker_0): Yeah, I think at the end of the day, you're just trying to add value to your clients. And so my background, I started at Barclays, covering the healthcare industry, initially covering the healthcare services industry. And I got really lucky and fortunate to get promoted to cover life science tools and diagnostics.

3:57Unidentified speaker (UNKNOWN): Uh

3:57Unidentified speaker (speaker_0): Stuck at Barclays through 2020 and then joined an independent research firm, Nefron Research. uh was pulled in by Josh Raskin, uh, who's you know one of the co-founders of that firm and a mentor of mine.

4:10Unidentified speaker (UNKNOWN): And

4:11Unidentified speaker (speaker_0): so

4:12Unidentified speaker (speaker_2): Josh Raskin, hopefully

4:13Unidentified speaker (UNKNOWN): you're

4:13Unidentified speaker (speaker_2): listening.

4:14Unidentified speaker (speaker_0): Exactly. He'll he'll probably be listening. Uh, worked with him uh in the team for six years and you know, so had good experience in an independent model. You know, at the end of the day, so you have like a defined coverage. Uh today I cover 20 companies in the life science tools and diagnostics industry. But I think, you know, a lot of what

4:35Unidentified speaker (UNKNOWN): you

4:35Unidentified speaker (speaker_0): know clients care about is kind of like domain expertise. So, you know, because clients on the buy side, you know, I'm formally in the weeds on 20 companies in this industry. They might be looking at hundreds of companies. So to get like that next layer deep.

4:50Unidentified speaker (UNKNOWN): Um

4:51Unidentified speaker (speaker_0): I think is value adding to stay on top of having, you know, I think one of our, you know, big value propositions is we have very deep models, um, company models. There's a lot of data that feeds into our models that you might not be able to get elsewhere. And some of that is because w you know, we just maniacally track all these huge government data sets. Uh, it's really a labor of love. And those are the things that kind of feed our forecasts. And, you know, at least for me personally, I start with the data. You know, we kind of follow these big data sets, then we build off of it with surveys and KOL checks. And then we end up having discussions with management.

5:36Unidentified speaker (UNKNOWN): And

5:36Unidentified speaker (speaker_0): I would say like once or twice a year, the there's like this beautiful moment where the stars align where we see it across all the data. You know, there's this, you know, either something that's really positive or something that's really negative. And it uh is not aligned with what, you know, the management talk track is or what the industry view is. And I think those are like some of the best recommendations that we've had over time. And you can't force it, they just like kind of come to you. Um but uh you know, clearly just there's our process being so data based, there's a lot of room for automation. And I think this is where we started to get really excited with AI earlier in the year.

6:19Unidentified speaker (speaker_2): Yeah, yeah, yeah. That makes sense. And what um what what is it the typical sell side analyst, analysts, what is their split of desktop research versus more relational research, either on the phone, talking to managements or primary research out in the field? And my my my guess is that the junior analyst is doing more of the desktop research, the data work, building the models, and the senior analyst is doing more of the relational components, both meeting with management team, but then meeting with clients. But what has been your I've never been a self-ci research

6:54Unidentified speaker (UNKNOWN): analyst.

6:54Unidentified speaker (speaker_2): So walk us through how that that split and how that evolves over time.

6:59Unidentified speaker (speaker_0): Yeah, it's something I think a lot about because I think most very good senior analysts either come up their way through the ranks as junior analysts to become a senior analyst, or they'll come from industry. And yeah, I think the junior analysts historically do a lot of the what you would describe desktop research. I'm personally someone who likes to roll up my sleeves and update the models. I feel like just like the way my brain works, like you kind of have to see it in the numbers and kind of understand how things are moving. But that said, like I I I'm traveling a lot and a lot of it is just getting out of the field. So a couple weeks ago we were up in Montreal for

7:38Unidentified speaker (UNKNOWN): uh

7:38Unidentified speaker (speaker_0): the ASHG conference, which is the American Society for Human Genetics. So that was a big check on the sequencing world for us. Uh next week, I'll be attending an analyst day for a company in Florida. Uh, then I'll be in Seattle for the Associ Association of Molecular Pathology meeting. Uh, then there's an analyst day in New York. Um, so I'm I'm I I a lot of the relational stuff kind of follows on me. And when we I think that's where some of the value add, and it's just it takes time to cultivate these relationships.

8:10Unidentified speaker (UNKNOWN): Um,

8:11Unidentified speaker (speaker_0): you kinda you're able to um get a good feel for what's going on in the industry that you might not see in the numbers, or it it'll add context to what you're seeing in the numbers.

8:22Unidentified speaker (speaker_2): Yeah. Yeah. Um, that makes sense. And what um

8:26Unidentified speaker (UNKNOWN): you know.

8:26Unidentified speaker (speaker_2): I'd say in the AI world now, there's this growing belief that AI can do anything a human analyst can do. And for desktop research, I'm like slowly by the day aligned with that that that viewpoint, which we'll come back to.

8:44Unidentified speaker (UNKNOWN): Yeah. What

8:44Unidentified speaker (speaker_2): is the what is the

8:46Unidentified speaker (speaker_0): Yeah? And before answering that question, I would just say like this is one of my core convictions is that as we start to automate a lot of the drudgery, the job, it's actually gonna put more emphasis on the humans to get out in the field and kind of come up with these insights. And so um I'm personally excited about, you know, the ability to attend more of these like off the bean track conferences and try and pick up on things. You know, at least for me, you know, we went up to uh Montreal, you know, there's the big trade floor, you know, so you uh I'll always spend a little bit of time walking around, um, seeing if there's any new companies that I hadn't heard about in the past, you know, stopping by companies, you know, meeting with, you know, and I think this is something that takes a long time and experience. You know, you you kind of meet,

9:29Unidentified speaker (UNKNOWN): you

9:30Unidentified speaker (speaker_0): know, executives and people in the field, you know them by name, so you can just walk up and have a conversation about what's going on in the industry.

9:37Unidentified speaker (UNKNOWN): Um,

9:38Unidentified speaker (speaker_0): while

9:38Unidentified speaker (UNKNOWN): I

9:38Unidentified speaker (speaker_0): was up there,

9:40Unidentified speaker (UNKNOWN): uh,

9:40Unidentified speaker (speaker_0): we hosted a couple of interviews uh with experts.

9:44Unidentified speaker (UNKNOWN): Uh

9:44Unidentified speaker (speaker_0): one was uh with Roche speaking about their new sequencing launch. We also hosted an interview with

9:50Unidentified speaker (UNKNOWN): uh

9:50Unidentified speaker (speaker_0): one of the top genomics centers, uh

9:53Unidentified speaker (UNKNOWN): a

9:53Unidentified speaker (speaker_0): gentleman, Niall Lennon, who runs the genomics core at the Broad Institute. Um,

9:58Unidentified speaker (UNKNOWN): but

9:58Unidentified speaker (speaker_0): there were are also some company-specific events going on, Ultima Genomics, uh who I personally think is making a lot of progress in the world of and winning a lot of these AI projects. They hosted a full-day user group meeting that I was able to attend and learn a

10:13Unidentified speaker (UNKNOWN): lot.

10:14Unidentified speaker (speaker_0): Um, and then some of the other companies in the space hosted workshops. So I think a lot of it, I think what's really exciting for me in this job, and one of the reasons I love staying on the sell side, even though I've had opportunities to go to the buy side over. time, you're just getting paid to learn. And it's so much fun. Um, we also had some new AI processes that we deployed while we were at ASHG, which was super cool. Um, so even making, you know, the desktop portion of a uh relational experience, you know, more automated, it was pretty cool.

10:48Unidentified speaker (speaker_2): Yeah.

10:49Unidentified speaker (UNKNOWN): Yeah.

10:49Unidentified speaker (speaker_2): It's what you know, one of the cool things about my career is I've been able to be in both like a super deep seat where at my first Tiger Cub we had three positions per investment uh professional. And so really the goal was to get us out on the field a lot. And I went to a lot of these conferences. And then in the multi-manager seat, where all of a sudden, as a PM, I'm covering 300 stocks. And so it's very hard for me. to go to every one of these conferences across sub industries. Um

11:20Unidentified speaker (UNKNOWN): so

11:20Unidentified speaker (speaker_2): I've sort of had that prior in my mind too of as more of the desktop research is automated, does that free investors to actually get back in the field more and to start to build this relational alpha, if you will, of building these relationships and networks? And and my my my guess is you've observed this as well. Like it's not just going to one of those conferences where you generate the insights, going to the same conference five years in a row and going

11:45Unidentified speaker (UNKNOWN): to other

11:46Unidentified speaker (speaker_1): conferences,

11:47Unidentified speaker (speaker_2): and you start to connect the dots in that, in that, in that way.

11:51Unidentified speaker (speaker_0): Yeah, totally. Yeah. I 100% agree with that. Um one other thing, like building off this, like clearly we're trying to do things a little different here at the new firm at Opera Operon on Research. Um, though one of the things uh we announced a month ago is we're actually hosting our own conference uh in Cape Cod in May uh called the Operon Summit. And um, you know, I one of my core views is that I think the traditional sell side is gonna lean more into corporate access and doing events like that. And so I had the question, it's like, oh, you're doing a conference now too? And I'm like, actually I this leans okay. we're kind of known for everything we're doing with data and AI. I'm like it this is an extension of my view because I think like hosting more events like this is kind of really where the world is gonna go. You know, when everything else is automated, we you know you gotta find new ways to add value.

12:46Unidentified speaker (speaker_2): Yeah. Um yeah totally agree. So so walk us through your your AI pill evolution, right? Like what was the first, if you remember, like what was the first time you fired up Chat GPT for anything career wise? Like what has the what has the journey been like for you?

13:06Unidentified speaker (speaker_0): Yeah, everybody remembers their first time. Um I think uh I think the hardest part about starting your AI journey and I don't know if you felt this way, but like the hardest part about gets getting started was getting started. Um I don't know if you feel that way, but

13:23Unidentified speaker (UNKNOWN): it

13:23Unidentified speaker (speaker_0): uh

13:24Unidentified speaker (speaker_2): first time I had a look at a stock and this this is terrible like this is terrible. And probably felt that way for the first six to twelve months, probably until the reasoning models and

13:34Unidentified speaker (UNKNOWN): you know,

13:34Unidentified speaker (speaker_2): deep research. But uh maybe notebook LM was the first time I

13:38Unidentified speaker (UNKNOWN): could sort of see

13:38Unidentified speaker (speaker_2): something. that was okay, maybe this is actually useful. But GPT three point five I found terrible.

13:44Unidentified speaker (speaker_0): Mm-hmm.

13:45Unidentified speaker (speaker_2): I just wasn't creative enough, I guess, to see

13:47Unidentified speaker (UNKNOWN): that

13:48Unidentified speaker (speaker_2): that it would improve.

13:49Unidentified speaker (speaker_0): Yeah. I just remember so for me it started like in January when um

13:56Unidentified speaker (UNKNOWN): you

13:56Unidentified speaker (speaker_0): know there was that week or two when all you had the SAS pocalypse where all the software stock stocks were getting destroyed because you know Claude had released co-work. And I'm the type of person that's like I can't just have somebody tell me like, oh, this is gonna change the world. Like I need to see it for myself. And uh one of the experts, you know, a good friend who've done a lot of work with uh mister Mr. Alex Dickinson, a former executive, you know, at Illumina and now in on the bunch of boards for companies. He was like, Jack, you like just download Claude Desktop. And he basically walked me through it. And

14:35Unidentified speaker (UNKNOWN): I

14:35Unidentified speaker (speaker_0): thought about like, what's like the one of like the gnarliest tasks that I have to do? And I started with a data set from the NIH. They have

14:44Unidentified speaker (UNKNOWN): uh

14:44Unidentified speaker (speaker_0): it's called the NIH reporter data set, where there's grants data that's updated on a weekly basis. And man, this data set is a bear. I can't tell you how many nights I'm I was like historically like sitting on the couch, downloading data, my computer would crash, I would want to throw it into the TV. Um because it's, you know, I have two little kids. They're five and three. And, you know, it's like

15:08Unidentified speaker (UNKNOWN): I

15:08Unidentified speaker (speaker_0): spend so much of my day on the phone with clients. So like by the time I like actually am doing work, it's like late at night. And

15:15Unidentified speaker (UNKNOWN): I

15:16Unidentified speaker (speaker_0): was like, I know there's a better way to do this. And so over I downloaded Claude on a Friday, and I I spent the weekend automating this to the point where it was like, click,

15:29Unidentified speaker (UNKNOWN): all

15:29Unidentified speaker (speaker_0): the data downloads, click, all of my charts are made, click, all of the slides are created. And

15:37Unidentified speaker (UNKNOWN): it

15:38Unidentified speaker (speaker_0): blew my mind. because not only the way I think about it is like, you know, I automated something that was like really laborious and like prone to error. And I was basically I was able to build it in a way that was consistent with the way I would curate the data. Um, but not only that, you know, in the past, because this data set was so huge, we would only update it once a year, twice a year, publish it for clients. It was like this whole big ordeal. And that was the start of now we have a weekly product that at Operon we call the Operon Signal. We publish it literally every Monday. There's almost 20 different data sets that you know we've fully automated. And so we felt like it created a whole new product for us, but

16:26Unidentified speaker (UNKNOWN): I

16:26Unidentified speaker (speaker_0): felt like my signal was getting better. I was seeing things in real time that I couldn't do in the past. And so that was really exciting because it was like not only was I getting some time back, but like I felt like my product was getting better. And by the way, I I

16:45Unidentified speaker (UNKNOWN): I

16:45Unidentified speaker (speaker_0): just re re-added added this all up. I think so far I've automated about 70 days of my year. Like things that took 70 days last year are now fully automated. I'm not working any less. Like, if anything, like I'm working more, some of it was starting the new business, but also like there's just more to do. Like you just want to keep digging.

17:07Unidentified speaker (speaker_2): Yeah. Yeah.

17:09Unidentified speaker (UNKNOWN): R

17:09Unidentified speaker (speaker_2): Really important question on that that data point. How have you built the confidence that that data process is is accurate? You go from doing, you know, inputting every number by hand to trusting this system. How have you built the internal conviction to trust this process? Because if you're sending this off to clients and there's errors in that data, that's

17:34Unidentified speaker (UNKNOWN): oh

17:35Unidentified speaker (speaker_0): yeah.

17:35Unidentified speaker (speaker_2): It doesn't matter that it's ten 10 times faster, you start to lose credibility. in that work.

17:40Unidentified speaker (speaker_0): A hundred percent. Yeah, this is like my paranoia is uh you want everything to be right. And there's a few ways. One is we we'll do spot checking. So there is like a summary table using this data set as an example, we can click a summary table and kind of be able to like look at it and make sure that things are aligned. And when there's changes in the data, you know, we'll certainly like take a second or third look. You know, so there's one data set that we've looked at recently, some outlays, data from the NIH that was really weak to end the fiscal year. And I'm I'm like

18:16Unidentified speaker (UNKNOWN): to

18:16Unidentified speaker (speaker_0): my associate, we gotta take a closer look at this. Um, the other one was the clinical trials data, which you know, we publish on a weekly basis for client. It's been so strong. It's actually that it's actually like one of the big calls that we made when we launched at Operon was the most bullish data set that I see is this clinical trials data. And I'm just paranoid that it's like this data cannot possibly be as strong as it looks but it is you know and we we do the checks. Maybe one interesting tidbit for you is a lot of the initial automation work that we were doing,

18:47Unidentified speaker (UNKNOWN): we

18:47Unidentified speaker (speaker_0): did it with Sonnet and then uh we re automated it with Opus, and then a lot of it we re automated with Fables. So this idea of like um how you know quality controls. I do think each time that we've automated and re automated it, we've tried to add in more processes to like have these checks in place to make sure that you're not you know. don't have things that are like silently breaking on you, but I'm still paranoid about it nonetheless.

19:21Unidentified speaker (speaker_2): I had the same experience where for for I I've done it more less with data work and more in financial modeling and there was these mod you know AI XL benchmarks last year that would show accuracy in this mid mid-60s. sixties. And I'm

19:36Unidentified speaker (UNKNOWN): like

19:36Unidentified speaker (speaker_0): for

19:37Unidentified speaker (speaker_2): if you if you if you have a junior analyst who's accurate 65% of the time, they're fired like six times by Tuesday afternoon. Um and so that had been a bottleneck for me in adopting processes, but the ability to run validation loops and sort of multi-agent validation now, like for financial modeling, what else they all I'll sort of ping numbers off of three MCPs. And if you know my system says gross margin is this, I'm actually pulling in three different MCPs to get you know triple verification. Um, and if they're if even if two of the MCPs differ, you're actually going into source document and you can design the system now. Like, holy crap! Like lately, I built a thousand models when I went and I spent about six or seven hours

20:24Unidentified speaker (UNKNOWN): to

20:24Unidentified speaker (speaker_2): do the same thing, like spot checking after finding. no errors. I'm like, all right, I think this is like uh I think we have something here.

20:31Unidentified speaker (speaker_0): Yeah, 100%. I think uh

20:33Unidentified speaker (UNKNOWN): the

20:33Unidentified speaker (speaker_0): way you describe it is the way we're developing our beacon data set, which is our data portal. And uh we have for financial data, we have two different sources of information. We have one coming from an external provider we're working with, but we're also pulling data directly from the SEC. And so we're doing a similar spot check. It's like, do these numbers align? Okay, good to go. Um definitely something that we think a lot about.

20:59Unidentified speaker (speaker_1): It's uh really interesting. So tell me more about this. I guess you know, the in the AI world they call it Jevons Paradox, where you had this stack of processes that you would do in your old job as a cell sell side analyst. It sounds like phase one is automating many of those desktop

21:18Unidentified speaker (speaker_2): steps.

21:19Unidentified speaker (speaker_1): How have you thought about what these new systems have unlocked to go to go deeper?

21:25Unidentified speaker (UNKNOWN): These

21:25Unidentified speaker (speaker_2): sort of

21:26Unidentified speaker (speaker_0): seven

21:26Unidentified speaker (speaker_2): days you talked about, like how are you redeploying those 70

21:29Unidentified speaker (speaker_1): days?

21:30Unidentified speaker (speaker_0): Yeah, so it might be helpful to talk about like our earnings process and how this has evolved. Um, so we launched so I uh left my prior firm on May 1st. We came out of hiding on June 1st, and we launched the new firm on July 1. Um and so

21:50Unidentified speaker (UNKNOWN): we

21:50Unidentified speaker (speaker_0): were obviously in commercial mode getting things kicked off, and we but at the same time, we were trying to automate as quickly as possible going into the second quarter earnings season. And in the past,

22:02Unidentified speaker (UNKNOWN): you know,

22:03Unidentified speaker (speaker_0): like if you think about earnings season through the seat of a sell-side publishing analyst, the report hits. And then in the past, you're trying to update the model as quickly as possible. And that would either be me or my associate Jesse.

22:16Unidentified speaker (UNKNOWN): And

22:16Unidentified speaker (speaker_0): then you're trying to get a first look report out as quickly as possible, with as much accuracy as possible. You know, then there's the company host a conference call, then there's a call. callback with the management team to like help align your model a bit. And then you're going to publish a follow-up report. And where it becomes truly chaotic is when you have multiple companies all reporting earnings at the exact same time. And so what I tell companies is in the past, when I have one company reporting at a time, you get 100% of my brain. When I have two companies, you get sub-50% of my brain because some portion of my brain is just trying to stay alive. And then when I have three, it's more like you're getting 25% of my brain. And then sometimes I'll have four companies and it's, you know, I'm like, you're getting 10% of my brain at this point because half of my existence is just trying to stay alive and keep things in line. So we have created a new process where the report hits along with any supplement or presentation materials, and we've developed an app that will parse the documents and pull out the numbers. You know, obviously, there's off-the-shelf vendors that do things like this. I think I think about for earnings season like speed, being able to and being able

23:32to own that process.

23:34Unidentified speaker (UNKNOWN): Um,

23:34Unidentified speaker (speaker_0): I felt like it was important that we do that. So now within a minute or two. our model is filled in for the current quarter. And, you know, I've had some people push back on me and be like, Jack, is that like really saving you much time? It only takes me 15 minutes to fill in the model. And I'm like, yes, but

23:54Unidentified speaker (UNKNOWN): what

23:54Unidentified speaker (speaker_0): happens when you have four companies reporting at once and you

23:57Unidentified speaker (UNKNOWN): have

23:57Unidentified speaker (speaker_0): four different presentations all over your screen and you're trying to like, I'm filling in one company, but I'm putting it in the wrong model, et cetera, et cetera.

24:05Unidentified speaker (UNKNOWN): It

24:05Unidentified speaker (speaker_0): like I think this process is working really well for us and it's very straightforward. But then what we do is we'll take our updated actuals versus estimates, and we have another app uh that my associate built where we'll fill that in with the prior quarter uh the prior quarter note, um, how the company reported all the documents and you know, we have a first cut of the report probably within 10 minutes. And then it'll have things that are highlighted that we need to take a look at. And so

24:38Unidentified speaker (UNKNOWN): in

24:38Unidentified speaker (speaker_0): the second quarter earnings season, you know, right out of the gate after launching the firm, we were already seeing major benefits, you know, from this automated process. And I wouldn't say earnings earning season was a pleasure, but it kind of gave us time to work on other things. We published five surveys for clients right after earning season wrapped up.

24:59Unidentified speaker (UNKNOWN): You

25:00Unidentified speaker (speaker_0): know, that's the type of stuff that, you know, like progressing non-earnings related goals during earning season was really unheard of. Or even last week, when I'm, you know, in Montreal at the ASHG conference, I had three companies. report earnings that I formally cover. I don't think I could have done that previously. Um so yeah, and this process is only getting kind of better and better because we're re-automating these things now. And I think about with what we're building with our data product, which we call Beacon, eventually we're gonna going merge to these processes. And I see a world where

25:36Unidentified speaker (UNKNOWN): a

25:37Unidentified speaker (speaker_0): company reports earnings and this workflow is automatically kicked off where my model is updated, we have a first look at the note, and then we're just like kind of quickly editing and getting the report out as quickly as possible. I can see in six months, you know, this process being really streamlined.

25:56Unidentified speaker (speaker_1): How do you how do you protect against this concern of cognitive deficit? Because if I am a hedge fund and I am a an Operan client, what I really want is Jack's opinion on the print. What I don't want is Claude's opinion,

26:16Unidentified speaker (speaker_0): yeah,

26:16Unidentified speaker (speaker_1): you know, sent to me as Jack's opinion because I can do that on my own.

26:21Unidentified speaker (speaker_0): Right.

26:21Unidentified speaker (speaker_1): Uh how do you think about the the the line and yes, automating what you're doing while making sure that you're adding your special sauce of ins insight and reconciliation?

26:35Unidentified speaker (speaker_0): Yeah. 100%. I personally think like as you automate a lot of the drudgery, it actually gives me more time, you know, to think about, you know, the things that we're putting on the cover or identifying the signal. Um, so maybe as a like a really good example, with second quarter earnings season, there was a night where there were four companies reported all at the same time. One of them was a very important vendor for three of the other companies who were customers in this ecosystem. And

27:09Unidentified speaker (UNKNOWN): in

27:09Unidentified speaker (speaker_0): the print for one of the customers, you could see the cog the cost per test. line item really came down like relatively dramatically quarter over quarter. And it was because they had done a conversion of the platform they were running their tests on during the quarter. And,

27:30Unidentified speaker (UNKNOWN): you

27:30Unidentified speaker (speaker_0): know, again, like in the old world, where I'm just trying to stay alive during earnings season, and it's like it's tough to compound earn learnings when you have four companies all reporting at the same time. But it was because we had this automated process, and I wasn't like me or my associate, you know, like spent burning a lot of calories just trying to fill the model. And it actually gave me time to like look at them all and be like, hey, what's going on here? And so when we do the callback with the management team, you know, you're smart enough to ask the question. It's like, hey, this important line came down over a hundred dollars per test quarter over quarter. What's going on here? And they're like, Well, we actually did this conversion and by the way, it's gonna come down even more next quarter. You know, so I felt like it was actually making me smarter and I was getting insights out of it that I wouldn't have done otherwise.

28:14Unidentified speaker (UNKNOWN): They were in

28:14Unidentified speaker (speaker_1): the past you might be still be updating the tax rate on another company just to make GPS foot so you're sort of automating that and giving you the ability to spend more of your brain cycles on matter. Yeah. Do

28:27Unidentified speaker (UNKNOWN): you

28:27Unidentified speaker (speaker_1): think that has led to a to the the caliber of your earnings notes going up?

28:33Unidentified speaker (speaker_0): I think so. Yeah. I I was speaking with uh someone recently who was like, whatever you're doing, keep it up because like I think your research has like never been better. And I that made me really proud. You know, like I think at the end of the day when you're on the sell cell side, your reputation is really all you have to go off. of. And um I was pretty happy about that.

28:54Unidentified speaker (UNKNOWN): Um Yeah,

28:55Unidentified speaker (speaker_1): sort of a forking, it's almost like this concept of agency. Like a the system could allow you to be very lazy and like put out mediocre research or or it could allow you to automate the drudgery and go deeper and and further drive you know quality and caliber of of research.

29:12Unidentified speaker (speaker_0): Yeah. Something maybe just to build off this point, something that I've been working on a lot recently is I'm calling it the operon knowledge hive mind. And I'm not convinced I'm doing this the most optimal way yet, but I do think it's a stepping stone to getting to what I think will be the optimal way. So basically, I've created a a structure on our drive where every time we publish a report, uh, the insights from that report are going into our hive mind. Anytime there's a transcript where management is talking talking, the key points are going to the hive mind, all of the numbers, whenever we extract numbers with our models app, they're going into a recurring, you know, kind of database of numbers. And so I think there's going gonna to be a way, you know, like when we're doing our previews or reviews. I can kind of point at this drive and be like, okay, like let's talk about, you know, what's the latest, greatest at this company.

30:10Unidentified speaker (UNKNOWN): I

30:10Unidentified speaker (speaker_0): think it's just gonna make me move a lot faster at the end of the day because I think you still need to come over the top as a human and know

30:18Unidentified speaker (UNKNOWN): kind

30:18Unidentified speaker (speaker_0): of like what's important, dig more into that, what to push back on because maybe the AI is walking you down the wrong path.

30:25Unidentified speaker (UNKNOWN): Um

30:26Unidentified speaker (speaker_0): there's still gonna be an important role, I think, for the humans and all this.

30:30Unidentified speaker (speaker_1): Yeah. What do what do you see in terms of um client adaptation? What is the how is the investment process changing? Are there, you know, what is sort of the early adopters look like? What

30:44Unidentified speaker (UNKNOWN): percent

30:44Unidentified speaker (speaker_1): of the your clients are in that bucket?

30:47Unidentified speaker (UNKNOWN): How

30:47Unidentified speaker (speaker_1): many people are holding on to the old old process, et cetera? Like what are your what are your sort of observations out talking to the bicep?

30:55Unidentified speaker (speaker_0): Yeah, it's a mix for sure. Um I would estimate a third of my clients are using AI beyond the chat feature. It could be more, you know, because some might be hiding the fact that at how heavily they're using AI, but I do think the ones that are really red-pilled, red pilled, you know, will we'll get on our calls. And I would say like in a 30-minute call, we're probably spending five to 10 minutes just talking about process with AI and like how to do things better. Um, I would guess that five to 10% are really bleeding edge. Um and I think of these people that have dedicated computers to that are,

31:37Unidentified speaker (UNKNOWN): you

31:37Unidentified speaker (speaker_0): know, on 247. churning out work for them, you know, in terms of being specific AI computers. Um they're that group is smaller, but they're some of my smartest clients. Um not to say like I would put myself in a that bucket, but we do have two dedicated AI computers uh at this point here for Operon Research. We we call them Batos Bothos and Xyros, but uh I think that's basically they're just chugging nonstop doing a lot of the data work and building our data platform for us. Um

32:14Unidentified speaker (UNKNOWN): I

32:14Unidentified speaker (speaker_0): do think there's some clients that haven't started their AI journey yet. And I talk with them about them like I mean these are some of my smartest clients. And I it I think it could be for a mix of reasons. I think some people are just their hands are tied, you know. by what their organization will let them do.

32:29Unidentified speaker (speaker_2): Yeah. And

32:30Unidentified speaker (speaker_0): I think there's some where you just spend so much time,

32:33Unidentified speaker (UNKNOWN): like

32:34Unidentified speaker (speaker_0): there's a cadence to what a junior analyst at one of these funds might do, where it's like you're in earnings season, then you go on the conference circuit, and then you blink and you're back in earnings season again. And so I think you really get as much out of it as you're willing to put in. And some people just

32:49Unidentified speaker (UNKNOWN): don't

32:50Unidentified speaker (speaker_0): have a lot of time. to put it in. And so it's not really a critique. It's just, you know, it's and I think about it for myself. It's one of the reasons I felt like it was important, like kind of needed to start my own firm to own this process and build it the way it needed to be built.

33:09Unidentified speaker (speaker_1): I would even push back a little bit on that. I'd say sort so of I I had a client ask me this in April is like, what's the R what's the ROI of AI for investment process?

33:19Unidentified speaker (UNKNOWN): And

33:19Unidentified speaker (speaker_1): my like honest observation is like it's been terrible, right? From like the early adopters who spent nine to eighteen months up until about probably August

33:30Unidentified speaker (speaker_0): this

33:30Unidentified speaker (speaker_1): year. I would say the return Turn on both token spend, but more importantly, time and attention

33:38Unidentified speaker (speaker_0): has

33:38Unidentified speaker (speaker_1): just not been there. Like firms and individuals spent a lot of time.

33:42Unidentified speaker (UNKNOWN): Yeah, you're making better

33:43Unidentified speaker (speaker_0): stock calls. Yeah,

33:44Unidentified speaker (speaker_1): dialing in these systems, even not just stock calls, but more just process. Like the process wasn't, you know, to a point where it could be trusted in so many dimensions. Um

33:57Unidentified speaker (UNKNOWN): do you

33:57Unidentified speaker (speaker_0): think it's gotten better as the models have gotten better?

34:00Unidentified speaker (speaker_1): For sure, 100%. I think to me, like open. Opus 5.5 is this huge threshold moment. Effectively, what people were doing in the in a harness where you had like a m you know orchestrating multiple tools and putting validation loops. To me, like valid the the ability of Opus 5.5 to effectively put a validation loop in a skill

34:21Unidentified speaker (UNKNOWN): is

34:21Unidentified speaker (speaker_1): something that's really huge because then you can do that in the right skill, and all of a sudden your accuracy gets to nothing. ninety-nine percent on a lot of

34:29Unidentified speaker (UNKNOWN): these

34:29Unidentified speaker (speaker_1): these workflows. But until then, even through the summer, I was finding so many errors in the basic workflows. And almost like, you know, investors who were saying I'm doing everything with AI, they're sort of they were sort of telling on themselves by in my by my assessment that the the rigor wasn't there in their investment process. So I like my view is only

34:51Unidentified speaker (UNKNOWN): in

34:51Unidentified speaker (speaker_1): the last two or three months can you actually run a process with the right level of rigor, right? In a way that's really speeding you up. Like we've had this, we've had this threshold moment. But so many of the early adopters who have tried things that they heard out about a hallucination or they know tried

35:09Unidentified speaker (UNKNOWN): something

35:09Unidentified speaker (speaker_1): with AI and it didn't work. They were like, hey, this people are sort of and there's so much hype and disinformation, you're turned off. Um, but that's like how you tr how do you try something

35:21Unidentified speaker (speaker_2): seventeen times knowing that the eighteenth time it will work, right?

35:25Unidentified speaker (UNKNOWN): Right.

35:25Unidentified speaker (speaker_2): It's hard when you're trying to keep your head above water with your your portfolio.

35:30Unidentified speaker (speaker_0): Yeah. And I while you're describing that, what goes through my head is like the things that we were originally doing in February, March and April and like I think about some of like the initial slide decks that we were making for clients where like I had to be relatively prescriptive with Claude. It's like put this chart in this location on the slide and this here and here and uh to like make it look the way that I would want it to look for a client. And then with Opus, it was just like you know, like it took

35:56Unidentified speaker (UNKNOWN): a

35:56Unidentified speaker (speaker_0): step up. But then with Fable, it really it was like another leveling up, in my opinion. On the model building. side, I I remember a client telling me in August that with Fable, it was the first time that they felt like they could really build a model and it be the exact way that they like it to look and feel, you know, and get to the number of decimal points on organic growth that they wanted it to be. Um,

36:22Unidentified speaker (UNKNOWN): you

36:22Unidentified speaker (speaker_0): know, and capture restatements the way that they would want to do it. So yeah, they're the tools are getting better. Probably still work to do.

36:31Unidentified speaker (speaker_1): Yeah. And Fable over the summer I think was eye opening, but then I quickly ran up like ten thousand dollars of tokens, like personally. Yeah my

36:41Unidentified speaker (UNKNOWN): credit card was on

36:42Unidentified speaker (speaker_1): fire. So I'm like, all right, you start to see something. But to

36:44Unidentified speaker (UNKNOWN): me

36:44Unidentified speaker (speaker_1): like Opus five 5.5, five September twenty 22nd, second, that was really like the I

36:49Unidentified speaker (UNKNOWN): pushed the

36:49Unidentified speaker (speaker_1): thing as hard as possible. And like some of it's subsidized, obviously, but the amount where you can really start to scale this rigor to me for the first time, like end of September. Like we're just a few weeks into these systems. And then you can start to like, you know, put a lot of these skills into a harness that's much more complicated. You can start to do the full end-to-end end to end research process. Like it's um so I sort of I my personal observation. is we're in this moment where the technology can lead to take off of what you can do with digital processes now.

37:22Unidentified speaker (speaker_0): Yeah totally.

37:23Unidentified speaker (speaker_1): What um

37:25Unidentified speaker (UNKNOWN): what

37:25Unidentified speaker (speaker_1): how do you think you know the how do you think classic cell side research responds uh responds? I wouldn't say classic cell Cellside side research has a has a uh best reputation for for being innovative

37:38Unidentified speaker (speaker_0): with

37:39Unidentified speaker (speaker_1): regard to two new technologies. Um what do you think? What do you sort of see um and what do you think classic sellside research will do over the next two or three years?

37:49Unidentified speaker (speaker_0): Yeah, we'll see. Um

37:52Unidentified speaker (UNKNOWN): yeah,

37:52Unidentified speaker (speaker_0): and I know we talked about this a few weeks ago about some of like the more progressive bulge bracket banks versus others. And I think one of the big questions, question and I whenever I run into my competitors at

38:04Unidentified speaker (UNKNOWN): you

38:04Unidentified speaker (speaker_0): know industry conferences or analyst days, I'm always asking them like, what are you allowed to do now? You know, because I think a lot of this is,

38:11Unidentified speaker (UNKNOWN): you

38:12Unidentified speaker (speaker_0): know, I think I hear a lot of folks that use Copilot now. Um and maybe they have some of the Claude models turned on within Copilot. pilot, But I I

38:23Unidentified speaker (UNKNOWN): I

38:23Unidentified speaker (speaker_0): don't think you're getting the full experience at this point. Um

38:28Unidentified speaker (UNKNOWN): one

38:28Unidentified speaker (speaker_0): of my worldviews is I do think the sell side like starts to bifurcate a little bit where it's the people who really, you know are doing real research are gonna go like more and more cutting edge. At least for me at Operon, the bar to starting your own research firm is dramatically lower than it would have been in the past. Um our tech infrastructure, you know, for what we use for publishing is a fraction of what,

38:59Unidentified speaker (UNKNOWN): you

38:59Unidentified speaker (speaker_0): know, I know was spent in the past. Like, and by fraction, I mean like pennies on the dollar. Um, I think setting up the processes to have like a world-class compliance organization and have all the legal infrastructure set up. Those things we did it all in a month. Um, I do think some of it was I probably had the benefit of you know working in an independent model so I knew what which buttons to push at the right time. But I guess the point I'm getting to is I wouldn't be surprised if you see more like top analysts hang their own shingle because I think the buy side is getting more accustomed to hang paying one off. for research that they really want. So I think you're going to see more of that. And then I think a lot of bulge bracket research, you know, focuses more. I think they lean into kind of what they do best, corporate access, I think capital markets.

39:58Unidentified speaker (UNKNOWN): Um, you

40:00Unidentified speaker (speaker_0): know, you clients m want to pay for that stuff. I think in the world where the humans are doing more uh, you know, trying to get out in the field, I think the value of corporate access probably goes up too. So um

40:11Unidentified speaker (UNKNOWN): I

40:11Unidentified speaker (speaker_0): kind of see it as like a kind of a barbell approach to research in the future.

40:15Unidentified speaker (speaker_1): Yeah. Yeah, there's this very common sort of

40:19Unidentified speaker (UNKNOWN): s

40:20Unidentified speaker (speaker_1): uh saying now in AI circles, like there's no moat. Like AI is AI the A the AI typhoon is um overcoming all moats. And I think in a lot of areas, that's right. I think the one thing that I sort of think is this concept of relational moat. Right. And to me, that's extremely enduring for just basic human reasons, right? It means that when you walk into one of these conferences, you've built these relationships with the right people over time, you've built trust of the investors over. time. And that's extremely hard to recreate. Um, it takes you years to to build that. And there really is relational alpha, right? It's it's just

41:01Unidentified speaker (UNKNOWN): the

41:02Unidentified speaker (speaker_1): it's the inbound of a network. Hey, Jack, this is changing in this and market. Have you heard this? What are you talking about? Oh no, I hadn't. That's interesting. Let me pull

41:09Unidentified speaker (UNKNOWN): on the

41:09Unidentified speaker (speaker_1): thread, right? Like that's something that that uh ultimately, at least for the time being, you still. we still have humans operating these businesses until we

41:18Unidentified speaker (UNKNOWN): have

41:18Unidentified speaker (speaker_1): agent to agent, you know, disintermediation. Um right, being in the field and having that network. And that's what great Southside analysts do, right? They're not sitting behind a a desk, they're out in the field, they know the man management teams, etc. et cetera.

41:34Unidentified speaker (speaker_0): Yeah. For the data product that we're building, I think a lot about what our motes will be for that business because, you know, I think by the time we've planned a flag for, you know, doing alpha testing before year end, doing beta testing in the spring, and then a mid-year commercial launch where we're having clients actually pay for it. And it was interesting just speaking with one of our clients at one of the big multi-managers who told, you know, obviously all the big multi-managers are making a lot of these investments themselves. And going back to something we were talking about earlier, this aspect of quality control, the client was like, you know, even though we're building a lot of this stuff ourselves, um, I think it's important, you know, I think I would still pay you a lot of money just to have you as a quality control that I can kind of lean on and benchmark a guess against um to know that the data that I'm looking at is what it is. And some of that, you know, again, it comes down to kind of like the relationships and the,

42:31Unidentified speaker (UNKNOWN): you know,

42:32Unidentified speaker (speaker_0): the reputation that you build over time. You need to be a trusted source. So

42:37Unidentified speaker (UNKNOWN): uh

42:37Unidentified speaker (speaker_1): I would even just you know give you a sort of a a a deeper idea of like you have these sort of data sets that have a qualitative signal. So you s see or sorry quantitative signals. So you see a quantitative sort of inflection. right the question from the investor seat is is that noise is that signal or is that noise right and so what an investor will do is see this inflection and then go try to get confirmatory sort of qualitative consideration. Let me go and do a dozen calls with people in the channel. Hey are you actually seeing something real shift here? And so part of what I think could be built now on your side of the wall is all of these expert calls and check-in, management check-ins, and actually turn that into unstructured data. Right.

43:20Unidentified speaker (UNKNOWN): So

43:20Unidentified speaker (speaker_1): you have sort of the confirmatory qualitative overlay, and where an investor can see both the quantitative signal spiking and management teams and experts in the field sort of confirming that based on, hey, this is why that signal, and this is not, it's not seasonality. or weather, it's something that actually fits to a longer term thesis. Like that's that that could be incredibly powerful across your vertical, but across 30 other cell systems.

43:50Unidentified speaker (UNKNOWN): Oh,

43:50Unidentified speaker (speaker_0): yeah, for sure. Yeah. I think about it, but the way that I think we're currently trying to structure that is when we launch, we're going, and I think probably by beta testing, we'll have Operon oper on MCP. You know, so for your Cloud or OpenAI, you'll have the connector into our hive mind. Um,

44:08Unidentified speaker (UNKNOWN): and

44:08Unidentified speaker (speaker_0): we'll also have Operon API, which we call Opera. So you can drink directly from our data lake. And so I think about it like for clients in the future, uh, because kind of at the end of the day, like where the world's moving, I think we're trying to sell to the computers, you know, but we got to keep the humans alive in this process. But I think about how a buy-side role might evolve in the future where you're working in Claude or uh ChatGPT or whatever you want to work in, you might have the Operon MCP connector in there to have the context of all this curated healthcare data. You might also have Alpha Sense

44:43Unidentified speaker (UNKNOWN): or

44:43Unidentified speaker (speaker_0): GuidePoint MCP or You're one of these other expert networks. So you have like the context of all the humans as well. So um yeah, we definitely kick the tires. Like, should we walk down that path on our own or partner just let the clients handle it? Um, but I think this actually shows up in our own survey work too, is like it I think in the world of AI, the value of the data really is goes up.

45:05Unidentified speaker (speaker_1): Yeah. We'll we'll take this we'll take this one offline, Jack, because I have some I have some good ideas

45:09Unidentified speaker (UNKNOWN): I love it

45:10Unidentified speaker (speaker_1): on uh uh on this. But I think, yeah, generally

45:13Unidentified speaker (UNKNOWN): I'd

45:13Unidentified speaker (speaker_1): say like my worldview is like all of these systems are really wonderful for the top third, the top third of sell side analysts, and the bottom two-thirds, where

45:23Unidentified speaker (UNKNOWN): the

45:23Unidentified speaker (speaker_1): bottom two-thirds that are do putting out generic initiation reports, generic earnings recaps, you know, generic topical deep dives. More and more AI is spinning up those reports incredibly effectively. And actually they're built on internal fund con hedge fund context. So built on alternative data and calls, all the expert network calls, all the alternative data.

45:49Unidentified speaker (UNKNOWN): So

45:49Unidentified speaker (speaker_1): actually you could build a a primer off of all of that data that has better data than the cell site would often.

45:55Unidentified speaker (speaker_0): Oh yeah. And

45:56Unidentified speaker (speaker_1): so how do you stay perpendicular on the cell sell side to that, to that sort of internal build? I think that's a

46:03Unidentified speaker (UNKNOWN): good thing.

46:03Unidentified speaker (speaker_0): Yeah, without giving up all of my secret sauce, um one of uh the cool things that I did at this recent conference is, you know, in the past, when you when you're either in the sell side or the buy side and you attend these companies are always doing like presentations and you're like, click, you know, you bring out your phone, you're taking a photo. In the past, I would be up late at night in my hotel room going through all the photos I have. on my phone, like squinting at the tiny screen, trying to read, you know, the data that I can't even make out what it was to begin with. Um, and then like come up with a summary for clients. And uh this time, you know, in ASHG, I'm like taking photos. It's going into a folder on my computer. I have a co-work session, I'm pointing it at this folder with all the photos, and I'm like, okay, Claude, like, let's talk about this. And like, here are the things that I thought were important that I learned today. Like, can you help me summarize this? Can you help me build charts? And uh look, I think we got I think about like this Ultima user group meeting that we attended. We got a report out later that day for clients. And I think in the past, you so you think about like the things where I'm adding value with AI that I might not this question you had. earlier about like a client that's like, I want to make sure I'm getting the Jack view and not like the AI view. It's like, well, I was able to get this report out a few days faster.

47:23So I think the speed was quicker. And I do think because like

47:27Unidentified speaker (UNKNOWN): I

47:27Unidentified speaker (speaker_0): was able to take in the context of everything that I learned and then like talk about it. And it was in the context of our hive mind Mine where all of our research is,

47:35Unidentified speaker (UNKNOWN): the

47:35Unidentified speaker (speaker_0): quality was pretty good too. Like I had management teams later. that week being like, yeah, like there were no headlines that you know Ultima had a sixty dollar genome announcement. You know, it was people that were like, Oh, I read it in your report. You know, so I think uh yeah, it's kind of seeing the payoff there too.

47:52Unidentified speaker (speaker_2): Once you get these systems in and dialed in,

47:55Unidentified speaker (UNKNOWN): uh

47:55Unidentified speaker (speaker_2): it's like the idea of going back and doing the old way feels like impossible.

48:01Unidentified speaker (speaker_0): Right. Yeah. Uh there's no going back. I've definitely been red pulled. And again, I'll go back to like what we said at the very beginning of this. It's like the hardest part was just getting started. You know,

48:10Unidentified speaker (UNKNOWN): because

48:10Unidentified speaker (speaker_0): it's a little overwhelming. There's like so much to do. It's like, where do you begin? You just got to start stacking brick by brick.

48:18Unidentified speaker (speaker_1): I'd say also it's like the a really hard part is like having the judgment to know what's a good process and not, because you see a lot of AI slop out there as well too. Uh and that's why I think we're like, you know, we're like old enough to to have eaten so much glass over two decades doing it the hard way. So we like have deep into our nervous system, we have the feeling of wanting to throw the laptop at the computer. But

48:44Unidentified speaker (UNKNOWN): I don't

48:44Unidentified speaker (speaker_1): think I don't think we're too old yet, hopefully I got a couple gray hairs in my beard now coming in. We're not too old that we can't, you know, learn some new and we're not too old as dogs to learn some new tricks. Um so I

48:57Unidentified speaker (speaker_2): do

48:57Unidentified speaker (UNKNOWN): wonder

48:58Unidentified speaker (speaker_2): like how much of this is just gen like we're in a nice generation to make this this switch.

49:03Unidentified speaker (speaker_0): For sure. Yeah, and I'm not gonna like go so far as to say I'm good at what I do, but I think like one of the reasons that like I'm I try to be right over fifty. 50% of the time. I might be 50.01%. I I make so many mistakes. Like I've made I've done things wrong so many times that like I think deploying AI, it's like I kind of like know where the guardrails all, or you need to know where the edge cases are to protect against, to reel it back in. Um so I think that is important. too.

49:40Unidentified speaker (speaker_1): What advice would you give w in the context of this broad conversation, what advice would you give twenty three year old Jack? Because the way you learned, I'm guessing, was

49:49Unidentified speaker (UNKNOWN): you

49:50Unidentified speaker (speaker_1): know, you're you're you're the guy on the desk that's a cost center. You're not adding much value. You're doing the annoying data work and model updates. And you know, by by doing that for your team, you learn through osmosis. now that bottom rung of the workflow has been automated in your system what does that what does that mean uh for your hiring over time what does that mean for the future of cell site juniors coming up the curve?

50:17Unidentified speaker (speaker_0): Yeah I I I think it's just all about adding value to your team. You know if you're a cell sell site side junior on

50:26Unidentified speaker (UNKNOWN): your

50:26Unidentified speaker (speaker_0): you're reporting reporting to your manager, it's like, how do you make their life easier at the end of the day? How do you make them, how do you make their profile better? How do you make the work product better? Um, those are the things that I would really recommend. I know what I would do. I I don't know if it's how most banks would do

50:43Unidentified speaker (UNKNOWN): it. Um,

50:44Unidentified speaker (speaker_0): I do think about it a lot for my associate, uh, Jesse, who's incredible. Um, and what I keep telling him is there's you're gonna have a huge runway here at Operon Research. I don't know exactly where it's gonna take you, but there's gonna be so many things that we're gonna be doing, you know, in the years ahead. Um, because when we launch our data product, it's gonna be healthcare wide. And so I think we're being very careful to think about like what jobs do we want humans to do in the future versus what are the things that we're gonna automate. There's gonna be roles for the humans as well. So I think finding right now, Jesse's doing an incredible job, you know, supporting the research process for insights. He's also helping with our data product.

51:28Unidentified speaker (UNKNOWN): Um

51:29Unidentified speaker (speaker_0): yeah, I do think like a lot,

51:31Unidentified speaker (UNKNOWN): I

51:32Unidentified speaker (speaker_0): do think a lot about like what the future of the junior seat's gonna look like. Yeah, I

51:37Unidentified speaker (UNKNOWN): I do

51:38Unidentified speaker (speaker_0): think there's gonna be a role for humans, but a lot of it's gonna get automated. So we'll see. Um

51:45Unidentified speaker (speaker_2): Yeah, I

51:47Unidentified speaker (UNKNOWN): the

51:47Unidentified speaker (speaker_1): way I've thought about it is, you know, listen, I I can do 85% of what I did my first year at a Tiger Cub with with Grokbot now. Uh-huh. And so the way I've thought about it is you earn that role on your team. You used to earn that role on the team by updating models, doing the sort of grunt work. I think you earn the role on your team now by being the most agent-pilled

52:13Unidentified speaker (speaker_0): person

52:14Unidentified speaker (speaker_1): you can, right? And sort of come in and help someone like you accelerate what they're doing. That's how you buy your year three years as sort of a cost center. Um and then ultimately you know, could could Jesse cover stocks two or three years faster than he could have in an alternative world because we're all so much more productive in that in that system. And then I I actually think like the note like the tails are much wider for for junior talent. If you don't adapt, you can be your skill set is really not helpful to where people like you were going. But if you can adapt, you can get to a point where you're driving revenue for your organization in a way that wasn't wasn't possible in in prior incarnations.

53:00Unidentified speaker (speaker_0): Yeah, totally. I think leaning into the things that the human only the humans can do. Um, but at the same time, in the short term, there is no going back. Like I think uh at least for me with Operon my view was like, you know, if AI is gonna take over everything, at least I'm going down swinging. Um, we'll

53:20Unidentified speaker (UNKNOWN): see where this

53:21Unidentified speaker (speaker_0): goes. Like I I think looking out into the future six to twelve months, like again, like I go back to January. If you had told me in January I would be doing what I'm doing today, I would have said you're crazy. So for me to look out six to twelve months and be like, I have my I have thoughts on where the world's gonna be, but we don't really know. Uh it's changing so quickly. So actually, maybe a good example is there's um an expert that we do a lot of work with. He's the CEO of a private company. Um

53:50Unidentified speaker (UNKNOWN): back

53:50Unidentified speaker (speaker_0): in February, I remember there's a big genetics conference in Florida called AGBT, and uh we we actually went I out and played golf together. He doesn't know how to golf, so I was teaching him how to golf while we weren't around. But he he was doing really bleeding edge stuff then that were, I think we might have finally caught up to where he was then today. But I met with him a couple weeks ago at this genetics conference in Montreal, and he told me in the last two weeks, he's increased his productivity 40x using Grockbot. Grokbot. And he was already incredibly productive. He has like an army of hundreds of Grokbot agents. He has one senior bot. And it,

54:33Unidentified speaker (UNKNOWN): you

54:33Unidentified speaker (speaker_0): know, I immediately messaged uh Kofi White, who's our CTO at Opera. And I'm like, we need to look into this. I don't know if this is the best approach for us here, but we just got to keep going. And what scares me a little bit is like things are snowballing.

54:47Unidentified speaker (UNKNOWN): And

54:47Unidentified speaker (speaker_0): it's not a linear snowball. The snowball is accelerating. And so you just gotta get on board, you know, because uh I

54:55Unidentified speaker (speaker_1): bet

54:55Unidentified speaker (UNKNOWN): the same

54:55Unidentified speaker (speaker_1): I would encourage you to sort of uh uh dig into Grokbot. To me, it's had a similar uh impact, almost unbelievable, like what

55:05Unidentified speaker (UNKNOWN): it

55:05Unidentified speaker (speaker_1): can do. Um pieces of what Claude could do, but in a way that's just fat you know, faster, better, more token efficient. It's really astounding, uh, to

55:17Unidentified speaker (UNKNOWN): be

55:18Unidentified speaker (speaker_1): honest. And if you start to take that into sort of the vertical, obviously it's a generic tool, it still has the sort of gen, you know, limits, limits of general model weights. Um but I think we are at that inflection point because if you start to take that sort of multi multi-agent agent concept deep into the specific vertical. verticals in a compliant,

55:40Unidentified speaker (UNKNOWN): um,

55:41Unidentified speaker (speaker_1): you know, domain, um, intelligent way,

55:45Unidentified speaker (UNKNOWN): um

55:46Unidentified speaker (speaker_1): I've seen it. Like it's it's no longer just this hand wavy, this hand wavy concept. You know, I think

55:52Unidentified speaker (UNKNOWN): um

55:52Unidentified speaker (speaker_1): these systems are coming and they're gonna be, I think, blow people away. way.

55:57Unidentified speaker (speaker_0): Yeah. I'm curious to hear from you, just because I know with fundamental edge, you interact with you know. a lot of the big research shops and also buy-side shops. What's your opinion on like where they are with the AI journey and where do you think uh traditional research is going to go?

56:14Unidentified speaker (speaker_1): It's not a great it's not a great sandbox for experimentation, unlike software coding, where if I want to test out if I want to test out, you know, a coding agent, I can write the code by hand. and then I can also test out a coding agent. It's very verifiable. You can go and inspect the code. I'd say I sort of start with the reality that an asset management organizations are fragile, fragile beings, particularly hedge funds. If you are down at most multi-managers, you know, three to five percent, do not PASCO, pass go, do not collect two $200. hundred dollars. So that does not create the most uh generous and environment for experimentation, trying new things. Having sort of lived through the the Tiger Cup quantum mental exercise of the 2010s, those systems were across firms were incredibly powerful for a period of time, but they occasionally would go off the rails. Um and so I think that's that will be the same dynamic with adopting AI and the investment process. There are elements that that are astounding. And

57:15Unidentified speaker (UNKNOWN): there are

57:15Unidentified speaker (speaker_1): elements where if you create cognitive debt and sort of rely on the systems for too much will will cause some issues. And so I'd say almost until summer, fall, like actually going I would counsel people to like not buy into the hype, not drink the Kool-Aid and go slow, right? Because in my own experimentations, I actually didn't see a lot. I've been trying to run a mock port mock healthcare portfolio. Um, and I've run into a lot of issues with that. Um, so I'd I'd say getting the timing right, not pushing too hard, right? Um is really is really important. But man, And as you know, it just takes a lot of hours just to dial these things in and

58:01Unidentified speaker (UNKNOWN): new

58:01Unidentified speaker (speaker_1): model drops and stuff. it's like, what does that mean to everything? Re-rebuilding your skills and the harness has been getting thinner and thinner, but allows you to sort of build this multi-agent harness, which is thicker and thicker. So there's just it's this constant evolution where I pro I spend, I don't know, I spend like in my my entire time experimenting with these things now. How do you do that when you're also covering 300 stocks and earnings? And so I think it's just been very hard. I think across my clients, um there there are people who like the people uh the median client is is not that far along at all in the

58:41Unidentified speaker (UNKNOWN): gym.

58:42Unidentified speaker (speaker_1): Um and I don't think like compared to other verticals, I think that's been right. Like if you were a fund last year that said we're doing everything with AI, I would like to counsel your LPs to get for all their capital.

58:53Unidentified speaker (speaker_0): Right. Yeah.

58:54Unidentified speaker (speaker_1): Um

58:55Unidentified speaker (UNKNOWN): but

58:55Unidentified speaker (speaker_1): I think what you're seeing now, and I think you have classic innovators' dilemma, like how do the big firms, because like doing this as a multiplayer scale dimension, I think that's really hard. Some will make

59:05Unidentified speaker (UNKNOWN): that

59:05Unidentified speaker (speaker_1): transition, others won't. Um but I think the interesting, the really interesting evolution I'm seeing is a handful of really AI-pilled people getting free. frustrated at the internal journeys of large funds and starting new funds, right? To create create what they used to do with 40 people, trying to do that with four people. And you're extracting some of the best stock pickers and and human investors from those firms and build the right pipe for those teams. That's where I think you're that journey. is starting now.

59:40Unidentified speaker (speaker_0): Yeah.

59:40Unidentified speaker (UNKNOWN): Um,

59:41Unidentified speaker (speaker_1): and I think that's where you'll see the early part of the cycle, these firms come out and just put up some really phenomenal three-year track records. Yeah. And that that might my guess will start to drive the acceleration of those sort of strategies across the more established established firms. But I think you're exactly right. Like more of the tier one firms are still more on the long only side than the multi-manager side are still stuck with cloud with with Microsoft Code

1:00:07Unidentified speaker (UNKNOWN): pilots.

1:00:07Unidentified speaker (speaker_0): Yeah.

1:00:08Unidentified speaker (speaker_1): Um

1:00:08Unidentified speaker (speaker_0): yeah.

1:00:09Unidentified speaker (speaker_1): So uh but um yeah as part of like what I try to do in my work too is just like show what's possible like show what you can do with Grockbot now. And um obviously compliance InfoSec is a huge issue. Like Grokbot Grockbus still tries to delete files occasionally like you know you're like, wait, why are you deleting four hundred and sixty-three files? Like uh I do that in a sandbox, you know, uh Google Drive where I have copies of everything and no

1:00:36Unidentified speaker (speaker_2): client.

1:00:36Unidentified speaker (speaker_1): So if it deletes my whole drive, like, all right, like I've lost a day of work. Um

1:00:41Unidentified speaker (UNKNOWN): but

1:00:41Unidentified speaker (speaker_1): if

1:00:42Unidentified speaker (speaker_2): you're

1:00:42Unidentified speaker (UNKNOWN): doing

1:00:42Unidentified speaker (speaker_2): that

1:00:42Unidentified speaker (UNKNOWN): where

1:00:42Unidentified speaker (speaker_1): you have fiduciary duty to institutional

1:00:45Unidentified speaker (speaker_2): institutional clients, like that's a whole different ball of ball of wax. But the same way that you you've seen workflows that are adding to your process, which I which like I hear what you're saying. I see hear what you're saying like awesome. And also like I think your mind will continue to get blown over the next nine months. It's not just you haven't found sort of cruise control like it will crock bot and all these other things will continue to it's the the horse isn't going back in the barn on this. I think things are fun fundamentally changing and the sort of the biggest transformation I've seen in my career.

1:01:20Unidentified speaker (speaker_0): Yeah, absolutely. Yes. I think about like some of my that like five to ten percent of clients that are like really at the bleeding edge. And I think like when I talk with them, I sometimes get the sense that they're like thinking to It themselves, in it's like, you know, am I in the right seat to really execute on this? I think they see how prolific this change has the potential to be. And I I wouldn't be surprised if they're kicking the tires themselves and like, should I just start my own fund and like and get backing to build this kind of the right way? Um, I know one uh one who had to like get special approval to get their own AI computers allowed to like run it the way he you wanted to do it, and um it's like a giant. pain in the neck. You know, so I think uh

1:02:06Unidentified speaker (UNKNOWN): I

1:02:06Unidentified speaker (speaker_0): so maybe just like as one other example here at Opera on research, like, you know, when we think about everything through like, you know, how do we how are we going to sell to these like massive funds? You have to be able to pass the compliance test on day one. And something as simple as like our DDQ, our due diligence questionnaire that we give to every client,

1:02:25Unidentified speaker (UNKNOWN): you know,

1:02:26Unidentified speaker (speaker_0): it says we use AI. You know, like and um in the past, you know, the DDQs at prior firms might be like, we don't use AI. And so, you know, it's we haven't gotten any pushback, you know, because I all these clients are using AI too. But I think like to like have it ingrained from day one as part of your process. Um, you know, I think about a lot of these firms, you know,

1:02:50Unidentified speaker (UNKNOWN): uh

1:02:50Unidentified speaker (speaker_0): who have a lot at stake, you know, sort of like it's like trying to turn the Titanic to kind of get

1:02:56Unidentified speaker (UNKNOWN): into

1:02:57Unidentified speaker (speaker_0): even though these tools are moving so fast, it's still like it's hard to implement.

1:03:01Unidentified speaker (speaker_2): And these concerns are real. I mean the hugging face and ha like the ability like to, you know, could could a firm like yours train on client data, right? Like that those those concerns are real. So I think listen, there's so much capital and so many smart people trying to solve these issues. Like I sort of operate under the assumption that they will get solved. Um but man, it's gonna be a fun it's gonna be a fun few years. Um it's gonna be a fun

1:03:29Unidentified speaker (speaker_0): weird. It's

1:03:30Unidentified speaker (UNKNOWN): gonna get weird.

1:03:31Unidentified speaker (speaker_2): It's gonna get weird. Yeah, it's gonna get really weird. Um but uh if you're trying to make sense of the noise. subscribe to invest with AI. We'll have Jack back on as a regular guest to sort of get a check on what's going on. And uh Jack, what's the best if people want to get in touch with you to learn about Operon, what's the best way to do that?

1:03:53Unidentified speaker (speaker_0): Yeah, the best way is to email me, jack at operonresearch.com, or you can find me on LinkedIn. Um love to work with you. Uh Brett, like the this was a lot of fun. I I really enjoyed following everything that you post on uh on X. Um, no, I mentioned this uh one of the times we spoke, but like I do feel like there's kind of a lot of parallels, like from what I see, like what you're doing with your clients to what we're trying to do, and even like it's more than just like the work, it's like lifestyle and personal, like

1:04:25Unidentified speaker (UNKNOWN): you

1:04:25Unidentified speaker (speaker_0): know, there's more to it. than just doing this job, even though I love my job a lot. I also love my family. And so I think, you know, a lot of what we're trying to do is,

1:04:34Unidentified speaker (UNKNOWN): you

1:04:34Unidentified speaker (speaker_0): know, keep things in balance, you

1:04:36Unidentified speaker (UNKNOWN): know, so

1:04:36Unidentified speaker (speaker_0): we can keep doing this forever. Um, but yeah, ha love doing this with you. I'm excited to keep

1:04:43Unidentified speaker (UNKNOWN): you

1:04:44Unidentified speaker (speaker_0): know, exchanging best practices and uh maybe we'll get you an as an alpha user when we launch uh Beacon later this year. We'd love to have your insights on that too.

1:04:52Unidentified speaker (speaker_1): I'd love to.

1:04:53Unidentified speaker (speaker_2): And I like I had this realization like

1:04:55Unidentified speaker (speaker_1): Grockbot is a godsend

1:04:57Unidentified speaker (speaker_2): for people like us

1:04:59Unidentified speaker (speaker_1): that

1:05:00Unidentified speaker (speaker_2): are always torn between wanting to be a great father and wanting to be great at our jobs. And I

1:05:05Unidentified speaker (UNKNOWN): think

1:05:06Unidentified speaker (speaker_2): you told the story of updating data late at night. And when your son comes in, is he's like, Daddy, let's go play pool, or Daddy, let's go play ping pong, you're like, Well shit, I have to, I have to get this data out for a client. Right. That's a very painful to have the ability to set up the system, to have your Brockbot or multi agent do that analysis. But, yeah, son, let's go play. And then

1:05:31Unidentified speaker (UNKNOWN): come back

1:05:31Unidentified speaker (speaker_2): and then validate it. It's, it's, it's this beautiful thing, right?

1:05:36Unidentified speaker (UNKNOWN): It really is

1:05:36Unidentified speaker (speaker_2): this beautiful thing that I'm having the realization. too it's like it's no longer this false choice of great at your job or great as a father like you can have you can have both which is

1:05:47Unidentified speaker (speaker_1): like such a blessing

1:05:49Unidentified speaker (speaker_0): totally

1:05:50Unidentified speaker (speaker_2): awesome good stuff Jack thank you so much for for for for for being with us really fun stuff

1:05:55Unidentified speaker (UNKNOWN): yeah

1:05:55Unidentified speaker (speaker_0): thanks again Brett this was fun

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights