YouTubeFeed

The AI Securing Your Workforce Before a Mistake Ever Becomes a Breach | E2314

Summary

Note on speaker attribution: the interview-transcriber’s diarization was seeded with “Jason Calacanis, Lon Harris, Brandon Dixon, David Im,” but Lon Harris does not actually appear in this episode. The host introduces himself directly (“this is Alex”) in the opening line and is later thanked by name (“Thanks, Alex”) by both guests. Every segment the raw JSON labels “Lon Harris” has been corrected to Alex Wilhelm throughout this write-up and the transcript below.

Jason Calacanis and co-host Alex Wilhelm open the show with Brandon Dixon, co-founder and CTO of Ent, a cybersecurity startup that just emerged from stealth with a $100 million seed round. Dixon’s thesis is that the security industry has “largely given up prevention” and become purely reactive — waiting for breaches to happen and cleaning up afterward. Ent instead puts an AI agent directly on employee laptops and phones that uses embeddings and semantic reasoning to build a behavioral baseline for each user, department, and cohort, then intervenes in real time — sub-second — before a risky action (an accidental data leak, use of unsanctioned “shadow AI” tools, or a rogue agent) turns into an actual breach. Dixon frames Ent less as a traditional EDR competitor and more as an “organizational work model,” analogous to the world models that unlocked self-driving cars: once a company has a behavioral picture of how work actually happens, that same data can flag inefficiency, unsanctioned tool sprawl, and candidates for agentic automation, not just security violations. A recurring theme is data sovereignty — Ent can run fully on-device or self-hosted inside a customer’s own cloud boundary so that sensitive behavioral data and IP never leave the enterprise, a decision Dixon and co-founder David Im say they made after learning firsthand at Microsoft how sensitive large companies are about where their data lives.

The second half brings back David Im (previously on the show with his “AI girlfriend” project Clawra) to demo Sume, a new project from his company Sumai Labs. Sume is a multi-model orchestration layer — a “wrapper” around five or six frontier video, image, and audio models — designed to solve the “slot machine” unpredictability of AI video generation, where the same prompt can produce wildly different results and creators burn time and money re-rolling. By routing a single prompt through multiple models (Gemini Omni, Seedance, and Sume’s own avatar model) and stitching the best clips together with persistent, consistent audio, Sume can generate up to 60 seconds of a single AI avatar with a consistent face — well beyond the roughly 15-30 second ceiling of individual video models. Im demos the live API, explains the product targets brands and UGC marketers on TikTok and Instagram (already 20,000 users, ~90% of paid customers being brands), and gets ribbed by Jason and Brandon over how cheaply he scored the sume.com domain.

Throughout, Jason repeatedly pushes both guests past the security and video-generation framing toward the bigger picture: Ent as a general-purpose “programmable endpoint” that could eventually surface productivity and automation opportunities across a whole enterprise, and Sume as part of a broader shift where AI orchestration layers — not raw model quality — become the actual product. The episode is sponsored by Vanta, Superhuman, and YSecurity.

Highlights

”We’ve largely given up prevention inside of security”

Brandon Dixon on Ent's prevention thesis

“You know, when my co-founder and I formed Ent, like the thesis for this was that we’ve largely given up prevention inside of security, right? We’re very reactive. We wait for the bad thing to occur.” — Brandon Dixon, 2:11

Clip command
yt-dlp --download-sections "*2:11-3:00" "https://www.youtube.com/watch?v=HSpG63Ryk_U" --force-keyframes-at-cuts --merge-output-format mp4 -o "ent-prevention-thesis.mp4"

”Citizen developers” and the risk of AI-empowered non-technical employees

David Im on citizen developer risk

“I’ve heard the term and I kind of like it, citizen developers. So we run into those people, ones that don’t have a technical background who are being encouraged by their leadership to solve problems with AI… and in that case, you know, they run the risk of deleting artifacts off their system, pulling in context that might be sensitive and then accidentally leaking it outside of the corporation.” — David Im, 4:22

Clip command
yt-dlp --download-sections "*4:22-5:41" "https://www.youtube.com/watch?v=HSpG63Ryk_U" --force-keyframes-at-cuts --merge-output-format mp4 -o "citizen-developers-risk.mp4"

”It has to be that fast to get prevention”

Brandon Dixon on sub-second on-device decisioning

“No, it has to be that fast to get prevention. So the idea, like the way that we roll out inside of a company is generally speaking, if you ask like a CISO, who are your riskiest users and why, they can’t really tell you, right?” — Brandon Dixon, 12:42

Clip command
yt-dlp --download-sections "*12:42-13:40" "https://www.youtube.com/watch?v=HSpG63Ryk_U" --force-keyframes-at-cuts --merge-output-format mp4 -o "sub-second-prevention.mp4"

”I’m terrified that one of the AI labs is going to try to scoop you up”

David Im on the value of Ent's behavioral data

“And this is why I’m terrified that one of the, you know, AI Lab, JV, FDE, private equity working groups are going to try to scoop you up and then take all the data you have and then apply it because it’s going to be incredibly valuable.” — David Im, 18:32

Clip command
yt-dlp --download-sections "*18:32-19:00" "https://www.youtube.com/watch?v=HSpG63Ryk_U" --force-keyframes-at-cuts --merge-output-format mp4 -o "terrified-ai-lab-scoop-up.mp4"

”That’s a $100,000 domain name”

Jason Calacanis reacting to the sume.com domain price

“Oh my god, you got a good domain name. How much did that cost you? Four letter domain name?” — Jason Calacanis, 40:04

“Really? That’s a $100,000 domain name.” — Jason Calacanis, 40:16

Clip command
yt-dlp --download-sections "*40:04-40:22" "https://www.youtube.com/watch?v=HSpG63Ryk_U" --force-keyframes-at-cuts --merge-output-format mp4 -o "sume-domain-name.mp4"

Key Points

  • Ent’s Lord of the Rings-inspired branding (0:44) - Named for Tolkien’s tree-guardians (“Ent” = “security for the enterprise”), a deliberate move away from the hoodie-hacker security aesthetic.
  • Founding thesis: security has given up on prevention (2:11) - Brandon Dixon says the industry is purely reactive; most breaches come from well-intentioned employees making mistakes, not just malicious adversaries.
  • AI is giving non-experts dangerous new capability (3:36) - Jason relates this to his own use of PowerShell/CLI tools via Claude Code that he’d never have touched before.
  • “Citizen developers” create accidental leak risk (4:22) - Non-technical employees encouraged to “solve problems with AI” risk deleting artifacts or leaking sensitive context; developers running many parallel agents face similar exposure.
  • Corporate policy + embeddings unlock real-time detection (6:10) - Ent starts from a company’s sanctioned-software list and corporate policy, then uses semantic embeddings to represent “words with meaning” in a way a computer can act on.
  • Baselines form per user, department, and cohort (7:14) - Ent continuously learns what’s “normal” for each employee and group so it can flag anomalies and enforce policy at the endpoint.
  • Single architecture runs on-device or on the back end (9:11) - Built as one system rather than separate edge/cloud products, so customers with regulatory constraints can keep everything local.
  • Cost-driven model choice: open models for embeddings (10:55) - Ent uses open embedding models it can optimize and run on CPU (not GPU), reserving larger cloud-provider or customer-supplied models for heavier reasoning.
  • Sub-second on-device decisions enable true prevention (12:42) - A ~2-week “burn-in” baseline period surfaces which policy violations a company actually wants enforced before Ent intervenes live.
  • “Organizational work model” framing, like self-driving world models (14:43) - Dixon compares Ent’s behavioral map of a company to the world models that made self-driving cars predictive rather than reactive.
  • “Too narrow to just call this a cybersecurity company” (15:55) - Jason argues the accumulated behavioral data has broad applicability - efficiency insights, SOC ticket triage, DLP context, training needs.
  • David Im: “I’m terrified” a big AI lab will acquire the data (18:32) - Half-joking concern about how valuable Ent’s aggregated behavioral/work data could become to a larger player.
  • Employee monitoring norms are unsettled (21:00) - Alex raises Meta’s unpopular internal tracking backlash as a comparison point for how granular workplace observability should get.
  • Data stays inside the customer’s cloud boundary (21:29) - Ent is deployed within the customer’s own environment for Global 2000-class clients who explicitly don’t want data leaving their control; granular toggles let companies choose what’s collected per group.
  • Self-hosting decision traces back to Microsoft (26:21) - David Im says the one-click self-host option came from seeing firsthand how sensitive enterprises are about data sovereignty and regulatory exposure.
  • Why raise $100M for an endpoint security product (27:50) - Dixon: building and maintaining a multi-platform, multi-cloud on-device agent plus landing Fortune 500 accounts is expensive; Ent positions itself to augment rather than replace commodity EDR.
  • “It’s been like a floodgate” since coming out of stealth (31:16) - Dixon says Ent hired 75+ people without paying a single recruiting fee, on the strength of founder reputation and pent-up demand.
  • Sume solves the AI-video “slot machine” problem (35:26) - Sumai Labs wraps 5-6 frontier video/image/audio models behind one API to get production-ready results without endless re-prompting.
  • Live demo: same prompt, three models, three results (36:11) - Gemini Omni, Seedance, and Sume’s own Avatar model shown side by side generating the same UGC-style script.
  • Roadmap: a one-shot, 60-second marketing video agent (36:58) - By routing and stitching multiple ~15-30 second model outputs with persistent audio, Sume produces up to 60 seconds of a consistent AI avatar.
  • Sume: 20,000 users, ~90% of paid customers are brands (40:31) - Core audience is on Instagram and TikTok, not Twitter/X.

Mentions

Companies

  • Ent (0:00) - Brandon Dixon’s on-device AI security startup, just out of stealth with a $100M seed round.
  • Microsoft (26:21) - Former employer of both Brandon Dixon and David Im (“Lou and I”); cited as where they learned how sensitive large enterprises are about data sovereignty.
  • Vanta (9:23) - Sponsor; AI-powered compliance/SOC 2 automation platform, cited as saving Ramp and Writer 82% of audit time.
  • YSecurity (29:46) - Sponsor; on-demand security team of 40+ engineers who’ve worked at Apple, Uber, Microsoft, Robinhood, and Brex, billed hourly with 6 free hours.
  • Grammarly (Grammarly Go) (19:29) - Sponsor; in-browser AI assistant sponsor read.
  • Sumai Labs / Sume (33:52) - David Im’s company; new project is Sume, a multi-model AI video orchestration API at sume.com.
  • GoDaddy (40:19) - Where David Im bought the sume.com domain at a 70% discount.

Products & Technologies

  • Ent on-device security agent (1:42) - Runs on laptops/phones, uses embeddings and behavioral baselines to stop policy violations before they happen.
  • Sume Avatar / UGC API (36:11) - Sumai Labs’ avatar-generation product; users pick an avatar and write a script, the API handles the rest.
  • Gemini Omni (36:41) - One of the video models shown side-by-side in the live Sume demo.
  • Seedance (36:11) - Another video-generation model Sume routes to; referenced in the episode description as “Seedance 2.0.”
  • Groq (video generation) (34:15) - Jason’s example of generating a Yoda clip (“frontier model wars begun they have”) that he rated 8.5-9/10.

People

  • Brandon Dixon (0:00) - Co-founder and CTO of Ent; formerly at Microsoft; has led multiple prior exits/startups.
  • David Im (33:37) - Co-founder of Ent and separately of Sumai Labs (Sume); previously appeared on the show discussing Clawra, an “AI girlfriend” product.
  • Alex Wilhelm (0:00) - Co-host; conducts the Ent interview and closes the Sume segment. (Diarized as “Lon Harris” in the raw transcript — corrected here; see note at top of Summary.)
  • Jason Calacanis (3:36) - Co-host; leads the business-model and go-to-market questioning for both guests.

Surprising Quotes

“I’m terrified that one of the, you know, AI Lab, JV, FDE, private equity working groups are going to try to scoop you up and then take all the data you have and then apply it because it’s going to be incredibly valuable.” — David Im, 18:32

“I don’t want to be nanny software. I don’t want to police how much AI somebody used throughout the day. That’s not what I care about. What I care about is stopping mistakes from taking place, removing adversaries from environments, making sure that people can adopt AI safely.” — Brandon Dixon, 24:22

“I’m never going to listen to a developer again who tells me they can’t launch on iOS and Android at the same time because it’s too hard. I feel like you’re taking on a much more difficult challenge.” — Jason Calacanis, 30:48

“Really? That’s a $100,000 domain name.” — Jason Calacanis, 40:16

“Now I’m starting to get AI slop sharks… sharks don’t interact for 90 seconds with a human. This is getting a little ridiculous here.” — Alex Wilhelm, 40:43

Transcript

Alex Wilhelm: 0:00 Hello and welcome back to Twist, this is Alex and right now AI and cybersecurity are hot topics because the leading frontier models are increasingly capable of finding and exploiting software vulnerabilities. Precisely how to harden global software is an open question that we’re all working on resolving. But there are startups working in the AI and cybersecurity domains that are not trying to build the next hacking tool. One startup fresh out of stealth has a novel approach to securing human action inside of corporations using AI that could help prevent the breaches of tomorrow. So, please join me in welcoming to the show, it’s Brandon Dixon, the co-founder and CTO of Ent. Brandon, how are you doing?

Brandon Dixon: 0:40 I am doing phenomenal. Trying to stay cool in this hot weather in Virginia right now.

Alex Wilhelm: 0:44 It’s bad up here, I can’t imagine how bad it is down in the sticky south. But before we get into anything important, uh, I’m shocked that after all the mining we have done on Lord of the Rings Arcana, Ent was not taken yet. How is that possible given that we’ve already gone deep into like the Silmarillion to find startup names?

Brandon Dixon: 1:05 You know, I think it’s just a happy coincidence. Like when we were building out the company brand and the name, we were thinking of, you know, enterprise as like kind of our core customer that we’re going after. And so like Ent, like security for the enterprise, um, was kind of the approach there. And then of course, being more technically inclined people, there was a little bit of a nod to the, to the kind of Lord of the Rings trees and the protection and as we were building out that brand, it was important to me at least, like, I don’t like the hoodies and hacker depiction of like, you know, the, the bad guys. I like, I like being outside, I like doing those sorts of things and I wanted to try and carry that through in the brand as well.

Alex Wilhelm: 1:42 Well, it works out well, it’s very memorable. It was not hard to recall who I was talking to today. Now, let’s dig into what you’ve built because I don’t think people are going to be as familiar with this tool. But to give people a quick summary from what I understand, you have built an on-device agent for endpoints, so laptops, phones and so forth, that can essentially tell when someone’s going to do something they shouldn’t do or maybe risky and then stop them. Now, tell me what I got wrong and break it down for me why we need this. I’m really curious.

Brandon Dixon: 2:11 No, I mean, no, that’s effectively it. Like, um, you know, when my co-founder and I formed Ent, like the, the thesis for this was that we’ve largely given up prevention inside of security, right? We’re very reactive. We wait for the bad thing to occur. We have the requisite information to troubleshoot, but it felt like there was, we were just accepting that the adversary was going to compromise the, the, the infrastructure. And a lot of the breaches occur because people are well-inclined, trying to do their jobs, but they make a mistake. And so when we were looking at the current AI advances, um, with some of these reasoning models and the ability to scale up understanding words and, and representing that in, in dimensional ways, uh, we wanted to apply that directly to where people worked. So we wanted to meet them in that moment, look at the work that was taking place… place and then make an assessment as to whether or not they were going to violate corporate policy or do something they shouldn’t, and effectively stop that from happening. So why do we need it now? Well, unfortunately, like breaches still occur from humans, right? Like people click things. They want to do the job. If they were otherwise security experts, we wouldn’t be dealing with breaches, right? Everybody would do the right thing. But, on top of that, we do have AI in the mix, and it’s a new technology, and it just brings new risks. So we believe that having that level of reasoning and capability at the endpoint is an important advancement in the future.

Jason Calacanis: 3:36 I want to get to the endpoint point and talk about models and such in a second, but the idea that people are doing more thanks to AI really resonates with me because up until when I had, I mean, frankly, OpenAI, Claude, Codex and Claude Code, I was not messing around in PowerShell or with the CLI. And now I’m doing all sorts of insane things with my computers that I’m absolutely not qualified to do. But that’s the homebrew side of this. So take me inside of a corporation that’s rolling out AI tools that give, you know, job functions more capacity, capability than they would have had before and are now seeing these problems. I’m curious about like how it manifests and which jobs are really pushing the envelope, if you will, and getting into trouble.

David Im: 4:22 I think it really depends on the maturity of organizations. So a lot of the people that we’re working with are obviously adopting AI, they’re kind of all in, and they’re looking to retool processes. I’ve heard the term and I kind of like it, citizen developers. So we run into those people, ones that don’t have a technical background who are being encouraged by their leadership to, you know, solve problems with AI and like, you know, accelerate their workflows. And in that case, you know, they run the risk of, you know, deleting artifacts off their system, pulling in context that might be sensitive and then like accidentally leaking it outside of the corporation. On the flip side, you have your developers who are of course trying to take advantage of new technology and in that case they might have like 20 different agents running to go and perform various roles across the enterprise, and there the same risks take place as well. But it’s not even just innately tied to AI, right? Like there’s still just mistakes that people do by accidentally sending an email, you know, with financial information to the wrong person, right? Or sharing credentials across chat ecosystems that otherwise could lead to a compromise. So we see both sides of the house. We still look at the user behavior, and we look at how they’re working with agents, and then we look at the agent behavior as well.

Jason Calacanis: 5:41 That’s really interesting to me, because it’s a bit broader than I thought. How can you have enough context about a company, its individual job functions, what they are allowed and not allowed to do from the corporate perspective to determine in real-time if person X with job Y in group…

Alex Wilhelm: 6:00 The responsibility, you know, Q is doing something that is suspect because to me that implies a level of specificity that’s almost, you know, crazy. So I’m impressed that you managed to figure it out. How?

Brandon Dixon: 6:10 So for us, like the setup of the product itself is predicated on like getting a corporate policy or getting something as simple as like, what is the sanction software that you use across the enterprise? And you’d be surprised with that little bit of information and understanding the context that we’re collecting, I can guarantee that there are policy violations that are occurring across the company. So people come to us and they say, look, um, I’m, I’m not going to restrict AI usage. I’m going to let it happen. Uh, but now my concern is I can’t keep up with all the AI tools that are coming out and I’ve given people access to use these tools, but I don’t know how they’re using them. So right away, if I understand what like the allowed software is, I can immediately tell you people that are using other software, be it AI remote access or something else that is not sanctioned for them. So like that’s like hour one sort of uh output for us. And then over time, Ent basically understands the operating aspects of the business because we’re constantly forming baselines of what’s normal for that user, what’s normal for that department, what’s normal for their cohorts, and that allows us to essentially start to put the policy to use on the endpoint itself and stop bad things from happening. So we don’t really require a lot of information. I think the, the big advantage that, you know, or what makes this possible today and why it wasn’t possible, you know, a couple years ago is the advantage that we have in, in things like embeddings. The ability to take semantic words like things that we understand have meaning and translate them into something that a computer can understand and make decisions on top of.

Alex Wilhelm: 7:52 You’re talking about vector databases and tensors more broadly.

Brandon Dixon: 7:57 Sure. Yeah. Like it’s, yeah. I mean, in its essence, it’s basically, you know, words have meanings. We know how to interpret those, but it’s been difficult to represent a lot of dimensionality uh for computers to understand. But now because we have these large language models as part of like what makes them work really well is that extreme amount of dimensionality, right? That they understand like when these words are put together, they have a more specific meaning uh than maybe when they’re, you know, split apart from each other. And so that was an advantage that we were able to take uh within building the product and that serves us well.

Alex Wilhelm: 8:33 I mean, you’re literally talking about why people are moving away from vectors because they want to have a, a more multi-dimension way to represent data as numbers, than, I sorry, words as numbers. That all tracks with me, and that makes good sense. But I’m curious of the learning loop. Because you said you provide value right from the start with some corporate policies, but then you keep learning. On the other hand, the way that I understand the way how Ent works is that there’s a basically an on-device agent, which I think is doing local compute. So I presume there’s kind of an SLM involved there. And… Does it then—does it then federate information back to a centralized database to learn about Alex Incorporated and then send that back to the on-device agent? How—how does that process function?

David Im: 9:11 Yeah, so quite simply, when we were building out the architecture, we didn’t want to have a back-end process and an end-point process. We wanted to have one architecture that could work for either scenario. Because you might have some systems where they don’t want to run anything at the end-point itself because of, you know, regulatory compliance or whatever. So the way that we’ve designed the system is it can run completely on the back-end or it can run on the end-point itself, but it’s the same architecture. So you asked about—

Alex Wilhelm: 9:31 No, I’m just curious why. I mean, I feel like if you can centralize the compute, why not have the, I don’t know, a more powerful model? You have more flops to play with.

Jason Calacanis: 9:41 As a founder, you’ve taken the time to become an expert in your niche, and you fully understand the entire landscape. But with so many regulations and rules that you need to follow from federal agencies and offices here in the US to the European Union and beyond, it’s impossible for even the most diligent person to keep up with every exacting requirement. That’s why you need to simplify compliance with a trusted partner like Vanta. All my startups use Vanta, and they love Vanta’s AI-powered platform that automates your entire compliance process. Whether you’re preparing for a SOC 2, or you’re running an enterprise GRC program, or you’re doing an audit, Vanta’s going to worry about the security so your team can focus on building great products. That’s why some of our favorite companies like Ramp and Writer are spending 82% less time on their audits by working with Vanta. Whether you’re a fast-growing startup or a global enterprise, Vanta is here to help you automate. Vanta is here to help you automate your security and compliance and earn and prove trust. So get started today at vanta.com/twist. That’s V-A-N-T-A dot com slash twist.

Brandon Dixon: 10:55 So, cost, you know, not everything requires like the most, you know, the the greatest model or the frontier model. The way that we use models is we use open models that have been published out there. We do rely, like we’ll make use of like the cloud service provider models or if a customer wants to bring a model. When it comes to embeddings, we control those, and we’re using open models to dictate that. Those embedding models can run on the end-point, they can also run on the back-end. We’ve made changes to those models to optimize them to be very performant and to do particular use cases to solve like certain problems. And so like we control that architecture, but if a customer otherwise wants to do deep investigations on all the data that we’ve collected, they have the capacity to do so.

David Im: 11:31 Alright. Effectively, the way that the system works is we’re building out a pipeline where we put this context in. We start with the raw behavioral information and then we’re decorating it with, you know, who’s logged in, what is it that they appear to be doing, and aligning that to whether or not it’s a corporate violation or not. We have a single architecture—

Brandon Dixon: 12:00 that basically is built to run on the endpoint or can run on the back end. Because of the way that embeddings operate, we want to control that process and we use embedding models. Embedding models are great because they’re very performant. They don’t require like GPUs, they can operate on CPUs. And when it comes to decision making, they can make decisions in sub-second timeframes. Like, we don’t have to use a heavy amount of reasoning.

David Im: 12:23 The sub-second timeframes thing is what’s the most interesting thing because the way that I think about Ent in practice, and you know, it hasn’t been rolled out at the company that I work at, so I’m kind of theory crafting here, but as I go to do something I shouldn’t do, it goes, ‘Bling, Alex, don’t do that, that might be insecure.’ Is it that fast of a process or am I overestimating how quick technology is today?

Brandon Dixon: 12:42 No, it has to be that fast to get prevention. So the idea, like the way that we roll out inside of a company is generally speaking, if you ask like a CISO, who are your riskiest users and why, they can’t really tell you, right? They know the workflows that are taking place, they know roughly what people are doing throughout the day, but they don’t emphatically know this. And so typically the way that we roll out is we start in more of a baseline mode. We install Ent across like the ecosystem. It starts absorbing the information, again it has that corporate policy, we’re surfacing violations. The company is going to have more than enough thing to do based on what we’re surfacing in that moment. We don’t want to necessarily get in the way of anybody’s workflow. I don’t want to annoy people, I don’t want to run in like, you know, prompt them in some form unless I have information from the company. So there’s typically a burn-in period where you might have two weeks to establish baselines of what’s normal, what’s not, you know, where are the violations occurring that I actually care about? Because in your corporate policy, you might have something that stipulates nobody should use social media, but you may not actually care to enforce it. But you might have on the other hand that AI use and sensitive data that goes into these models has to be done in this particular way. And so for that moment, like, we’re going to drudge up that behavior. We’re going to say, here’s your unsanctioned AI usage. By the way, here’s what people are doing. This is the information they’re sending in. This is the work that’s taking place. And it allows us, we’ve built the product to basically isolate that workflow and then say, well, when I see this activity in the future, now I want to intervene. And that intervention can be customized.

David Im: 14:23 The way that I think about the product as this stop you before you do it thing is one portion of it, because it can also do that for humans and apparently also for agents, and it can provide a bird’s eye view into how a company is using software and in particular AI more broadly, so people can learn from their own usage?

Brandon Dixon: 14:43 Correct. So the idea here is like, you know, I’ve toyed around with like the lingo, but it’s like an organization work model. We talk about world models, and world models were like super advantageous to cars, right? Because prior to a world model existing that encapsulated the entire world—

David Im: 15:00 environment, we were trying to tell the car like when to stay in the lanes, when to apply the brake, when to like kind of, you know, change lanes, whatever. And the problem was is that you were overfitting to a particular set of environment variables. When you had the world models, it allowed the car to essentially become more predictive, right? Oh, I’m anticipating that somebody’s going to walk out in the crosswalk or there’s a stop sign and I need to apply the brakes. It allowed the self-driving car movement to be more accurate. So when I say that most systems aren’t aware of the behaviors that are taking place, step one is to make them aware from an observability perspective of you have some problems inside your environment that you probably didn’t realize. And step two is to figure out which ones do you actually care about. And then step three is to determine how do you want to modify that user behavior, if at all, or your corporate policies, and then enact those using End.

Jason Calacanis: 15:55 So really, I think that calling this a cyber security company almost feels too narrow. You are building kind of a work model in a sense. How far can you push that? Because once you have this information about how a company works, you could do all sorts of fun things, like tell them where they’re being inefficient or recommend different ways to go about things. It seems like if you can get wide adoption and a lot of information, this is a really potentially lucrative and useful tool that you’re building as a lot of future applications that go outside of just cyber safety.

Brandon Dixon: 16:23 Absolutely. I think the, you know, again, toying around with like kind of lingo here, I don’t know the best way to describe some of these things because they feel new, like we haven’t had it at our disposal. One of the ones that I was playing around with was like the semantic substrate for security. So all of a sudden I have this like, all of this information that’s describing the work that people are doing inside of the business. And that is a, that is a massive, you know, set of context that we have at our disposal that can help accelerate closing out true positive benign tickets in the SOC, right? It could give further context to DLP-related events. It could surface inside risk activity, be it the 1% bad guy or the 99% mistakes that take place. It could be used to isolate and identify people that need training. So like there’s a broad applicability in having context that describes what’s taking place across the organization. And you hit the nail on the head. There’s a productivity angle to this as well, in which if I understand what people are doing throughout the day, then it becomes ripe to figure out what things might AI agents benefit, right? What is the mundane, monotonous work that is occurring across different departments inside of the business or ones that are like particularly risky across my enterprise that might benefit by having an AI agent do it? And then once that AI agent is in place, how do you ensure and keep it on the rails, right? How do you know that it’s doing the right thing, that it’s aligned to the task? And so for us, we’re concretely focused in security, but we’ve, we go and target big enterprises, our… Our environments are Global 2000 and above, so think Fortune 500. And the thing that you articulated is what they’re after as well. They say, I can start with security, I can bring a level of visibility and stop problems from taking place. But then there’s this downstream applicability that becomes really attractive, like can we start mining out of that information ways to do process distillation, ways to identify who are like—who’s using AI the most and how are they using it in ways that we can help others learn from that?

David Im: 18:32 And this is why I’m terrified that one of the, you know, AI Lab, JV, FDE, private equity working groups are going to try to scoop you up and then take all the data you have and then apply it because it’s going to be incredibly valuable. But just listening to you on the lingo point, Work Ops, maybe? I mean, it does feel kind of like general work observability, lots of data, you can do different things with it.

Brandon Dixon: 18:56 We’ve—we’ve also used like, you know, I heard data lineage for a while, which is like looking at how data moves through the enterprise. And I think that that has merits and it’s proven in the market. We’ve been toying around with behavioral lineage as well. It’s like, what are the behaviors that people do and how does that like then intersect with data lineage? How does a behavior of an agent, you know, operate? Where we’ve settled on the marketing side is like the intent-aware. That’s why you see that, but you know, TBD on like how people respond to it so far, that’s been, you know, the way that we’ve been pitching it.

Jason Calacanis: 19:29 Most AI tools are adding friction, not making your life simpler. And it’s another tab to switch to and maybe you forget to even do it. It’s arduous. What you really want is one system that’s going to make you more efficient and save you time every single time you do work. That’s why I love Grammarly Go from the amazing team behind Grammarly, which I have insisted all my team members use since day one. Now, it’s an AI chat that lives on the side of your browser. It’s always there. Maybe you’re drafting an email mid-meeting, it goes and helps you finish it without switching apps. Maybe you’ve got a 40-email thread to get through before that call, it’s going to summarize it for you in seconds without losing your place. No new tabs, no starting from scratch, no context switching. Grammarly Go has the context of everything you’re working on. It works inside the tools and sites you already use. Your inbox, your docs, your browser, maybe you’re doing social media all day long like me, it’s part of my job. You can try many of Grammarly Go’s features for free. Find out more at grammarly.com/go. That’s grammarly.com/go.

Alex Wilhelm: 20:31 I would say, you know, lineage and substrate are a bit too technical, but then again, I have to say the words agentic orchestration at least three times a day, so what do I know about branding? Okay, let’s talk about a couple of other things. One is just the employee element of this. Now, this computer that I’m on right now, because Twist is owned by Launch and Launch is a financial company, has all sorts of tracking software on it, right? Just for the—you have to do that. I don’t love it, even though I don’t actually care because no one cares what I do. But it still feels a little bit weird to me that, you know, there is a record somewhere of, you know, every tweet that I click on, right? Now, in what you’re building, it is more granular. And so I’m trying to kind of sort out how much do employees care? Because on one hand, I think it’s becoming the norm to have your work observed to some degree. On the other hand, Meta just made a big push to really look at what their engineers were doing and that was very unpopular. So, where’s kind of like opinion and norms around this type of observation?

Brandon Dixon: 21:29 You know, I think it’s TBD to some extent. At least on the enterprises that we work with, a lot of them have corporate policies that stipulate the asset that you’re using is subject to monitoring for the purposes of like it’s their corporate asset, right? Like the intellectual property is there. So I think, you know, most of the security software that has been deployed even in a traditional EDR sense has been historically collecting all of this information about the actions people are taking on their system, right? And I, I do believe that we bring a new level of granularity to it and we have to determine, you know, what businesses are comfortable effectively deploying. And I think it’s just a matter of being straightforward with your employees. And so it’s a matter of if it’s in the corporate policy and you’re able to collect it, then, you know, okay, that’s fine. Our kind of general view is when it comes to the information that we collect, we don’t want it coming back to a central authority. We can be the people that host that environment, so we can deploy as a SaaS-based product and host on behalf, give you a dedicated tenant, but more importantly what we’ve heard from Global 2000 and above is they want to own their data, right? They don’t want it going to somebody else. So, Enz is deployed within the customer’s boundary. Like we don’t even get to see it. So the limiting factors there are that’s one way of kind of or retaining that, like it’s only staying within the corporate environment, it’s not being shared with other people. And then the secondary item is how much do you want to collect and centralize back to that, you know, that corporate backend. And so we’ve provided an ample amount of configurability that if you don’t want to send something to the backend, you don’t have to, you can keep it on the edge. Now obviously, there’s an operational gain in putting everything in a central storage.

Jason Calacanis: 23:24 It gets smarter, more data to learn on, yeah.

Brandon Dixon: 23:26 Yeah, but like a corporation may not want to do that. So for everything that we collect, there’s toggles that basically allow the business to turn it on and off. We support user groups, endpoint groups in terms of like, you know, you might want to collect more on your developers just because they have a higher risk pattern, whereas your legal you may not really want to collect anything at all. And then beyond that within the product, there’s role-based access control and attribute-based access control.

David Im: 23:50 Yep.

Brandon Dixon: 23:51 So like if we’re collecting something like screenshots for an investigation, I don’t want the SOC to necessarily… see that. So we allow the business to basically hyper-tune the information that’s exposed to any given party.

Jason Calacanis: 24:07 I feel like you’re nibbling around the edges of agentic identity and the issues and lack of maturity in that product world somewhat. Am I—am I wrong? If I want to know what the agents are doing, I’m going to want them to have a distinct permission set and a distinct identity and—

Brandon Dixon: 24:22 Well, it’s the same thing with people too, right? I mean, I—I think part of the problem with like a corporate policy is that it—it lacks teeth. It’s only as good as its ability to put it into a control point. And because corporate policy is written in natural language, there’s some level of like interpretation that takes place by the employee. And I think that’s sometimes what leads to, you know, mistakes being made is that the employee feels they’re working within the boundaries of the corporate policy when they may in fact not be. And the way that you get around this, right, is you have a Draconian, you know, way of—of stripping down the asset and removing the freedom and you’re just saying like ‘thou shalt work this way.’ And—and I don’t think that’s particularly uh fun to work in those businesses. I understand it, but like what I want to do is balance like being able to give a new control point layer to actually stop bad things from occurring while also not infringing on like privacy, right? And so it’s up to me as I design that product to put the controls in place to not build something that could otherwise be abused. And that’s something that’s very top of mind for us. I don’t want to be nanny software. I don’t want to, you know, police, you know, how many—how much AI somebody used throughout the day. That’s not what I care about. What I care about is stopping mistakes from taking place, removing adversaries from environments, making sure that people can adopt AI safely. And that requires some level of—of observability and understanding what’s happening.

Alex Wilhelm: 25:54 WorkOps. I’m telling you, it’s gonna be big.

Jason Calacanis: 25:57 All right, uh let’s talk really quickly about some business questions. So, you’ve mentioned how there’s a SaaS uh version of this and a self-hosted version of this. Now, when I usually see that breakdown, it tends to be open-source software. As far as I know, you guys are not pursuing the open-source approach. So, uh talk to me about the business uh decision there to allow for self-hosting and then also, how do you charge for this?

David Im: 26:21 Yeah, so self-hosting—the reason for it was when we were at Microsoft, we learned during this AI movement, people are really sensitive to their corporate data. And I think from a regulatory perspective, we’re seeing more emphasis on data sovereignty, we’re seeing like the intellectual property of a business wanting to be contained within its environment, they don’t want it to go to, you know, a third-party supply chain. And so it was a—it was a first-principle decision for us. In the same way that we said we’re not going to depend on any other security product to get our telemetry, because that impedes our ability to be preventative and make decisions quickly, we also said we’re going to make it out of the box, one-click deploy inside of whatever cloud you guys operate in. And we support the—

Brandon Dixon: 27:00 And so that was just an important decision for us. If we host it, then effectively we take on the hosting costs and we pass that on in the licensing. If you host it, that becomes like a COGS implication that you have to effectively manage that spend and we give you the predictability of what that looks like.

Jason Calacanis: 27:18 But I’m still paying you yearly, quarterly, whatever it is?

Brandon Dixon: 27:21 Yeah, there’s a licensing cost. Yes, there’s a licensing cost associated with the product. It just will change if you’re hosting it because you’re going to take on the actual hosting.

Jason Calacanis: 27:29 Okay, that makes good sense, but it’s still basically charged the same way, lower cost. Okay, that makes sense to me.

Brandon Dixon: 27:33 Yep. Yep.

Jason Calacanis: 27:34 Now, you guys came out of stealth and announced a $100 million round. These happen more often than they used to, and I’m always curious why you need that much money. It’s a lot. That’s an old seed fund from when I was younger. So, what are you going to do with $100 million?

Brandon Dixon: 27:50 So the thing with endpoint is it’s a well-established market, right? It’s something that there’s a lot of players and incumbents there. And to be able to like penetrate inside of Global 2000, Fortune 500 and above, you need to, like building that endpoint company takes a lot of effort, right? I have to build an agent that works across multiple different platforms. I have to, like, concern myself in the research and development of putting AI directly at the edge while also being able to run it on the back end. We talked about like, you know, hosting inside the customer’s environment, making sure that all of that infrastructure is supported via if you’re using AWS, Google, or Azure, right? Like, all of that takes a significant amount of engineering to get it right, to make sure that it’s tested, to not make those mistakes. And so there’s a lot of money raised to basically go and do that, just expensive to build a product. But it’s also expensive to then like get out in the market and land inside of these like big enterprise accounts. Like, people come to us and they say, hey, look, are you going and competing with the traditional EDR? And the short answer is no, right? I don’t want to go and compete with them on the same playing field. Why would I do that? From my perspective, EDR is a commodity at this point. Everybody’s got something in place. They might be reasonably satisfied with what they’re getting. My job is to augment where that EDR solution is not meeting the need, where an inside risk solution is not meeting the need, or a DLP solution is not meeting the need. So we talked about like this semantic substrate for security. My business is trying to build, you know, like the programmable endpoint. Can I solve a variety of different use cases using AI as my advantage across any platform in any cloud and give someone that level of visibility to understand what is happening inside their enterprise? It just takes money and capital to do it.

Alex Wilhelm: 29:46 The team is pumped because you’re about to close a massive deal, but then the client’s lawyers get involved. What happens if you get hacked? How are we going to protect your data? There’s no need to panic. This is why you brought in Vanta. Vanta is staffed with over 40… 40 experienced engineers who have actually worked security for world-class companies like Apple, Uber, Microsoft, Robinhood, Brex, and so many others. But the best part is, you don’t even need to hire Y Security. Your company can rent Y Security’s elite team by the hour. That means no massive salaries to pay, no costly consulting, just real experts embedded in your company, helping you out with your SOC 2, ISO 4200, or any security or compliance challenge you’re facing. You can even set a monthly cap so you know exactly how much you’re spending. And your first six hours are completely free. Head to ysecurity.io/twist and book your free six-hour strategy call. That’s ysecurity.io/twist.

Jason Calacanis: 30:48 Given how many different endpoints, clouds, and services you have to make work to have this actually function at the speed you need it to, I’m never going to listen to a developer again who tells me they can’t launch on iOS and Android at the same time because it’s too hard. I feel like you’re taking on a much more difficult challenge. All right, last question for me is pretty simple. When you guys came out of stealth, you announced that Ent was generally available, which means your go-to-market, you know, starting gun was shot off.

Brandon Dixon: 31:12 Correct.

Jason Calacanis: 31:13 How has reaction been? How has the market responded?

Brandon Dixon: 31:16 It’s been like a floodgate. Even before that, like we were like, the reason why we effectively, like, we’ve been in the market to some extent, not like out of stealth, but like kind of pseudo out of stealth for a couple of months. And the reason why we just kept consistently holding it back is like, I’ve retained a pretty healthy pipeline of, like, you know, folks that are interested in what we’re doing. So like, when Lou and I left Microsoft, they were like, ‘Man, you guys are going to do something crazy, and I want to know what that is.’ And so like, we’ve got a lot of goodwill in having multiple exits and startups that we just had a bench of people that were like, ‘Tell me what you guys are doing.’ And the second that we were ready, we were entertaining, you know, POVs and people that wanted to go and deploy the solution, test it out, because they’re like, ‘We’ve never seen anything like it. This is exactly what we were expecting.’ And so beyond that, like typically you come out of stealth because you want help in hiring people. And I’ve, I’ve hired well over, you know, 75 people now, and I’ve not paid a single recruiting fee. And so like, coming out of stealth was not lackluster. Like we’ve got a lot of, like for us, like it didn’t feel like any material difference. Like we’ve retained a healthy pipeline of people that are interested. I’ve got more people now asking me like, ‘Dude, I want to see this thing in action, this is what I was waiting for.’ And so for us, it’s just a healthy amount of demand beyond what we already had.

Jason Calacanis: 32:40 Do you have enough GTM infrastructure in place, sales teams, etcetera, to handle all the new inbound?

Brandon Dixon: 32:47 That’s exactly where we’re hiring right now. So we’ve got several sales reps, we’re hiring for sales engineering, you know, we’ve got folks on like the forward deployed engineering side that’s been staffed out. And so like, we’re building that out as fast as we can. We’re getting our pieces there. I feel pretty comfortable, but like, you know, now it’s just a matter of like going through the motion and execute, right? You know, having multiple deals in tandem, managing the POV process, making sure that we’re delivering success and outcomes. That’s the biggest focus that we have right now. So, early beginnings of it. We’re hiring, and then it’s a matter of just like continuing to turn the crank here.

Alex Wilhelm: 33:23 Well, we are looking forward to when you start announcing ARR milestones, but in the meantime, it’s int.ai. Brandon, congrats on the round. Congrats on coming out of fake pseudo stealth, and come back on in six months and tell us how it’s going.

Brandon Dixon: 33:35 Appreciate it. Thanks, Alex.

Alex Wilhelm: 33:37 All right everybody, next up on today’s twist. You remember David Im, he was previously on our show. He had Clara, the open claw AI girlfriend. Do you remember this, Jason, from February?

Jason Calacanis: 33:51 Yes, I do. I do.

Alex Wilhelm: 33:52 So he’s back. He’s got a new product from his company Sumai Labs. It’s an agent orchestration layer utilizing multiple video generation models. The goal is generating high-quality video outputs in just one attempt, Jason. The idea you could finally one-shot your AI videos instead of multiple go-rounds to get it exactly the way you wanted. David, thank you for coming back to the show.

David Im: 34:13 Yeah, it’s good to have you back.

Jason Calacanis: 34:15 I mean, when you make video, it is literally like a slot machine. You put in your prompt. I did it the other day. Pull up my Yoda one. I did it in Groq. It was relatively good. It got it right. I said I want Yoda from the Clone Wars style to say frontier model wars begun they have. Oh, I see. Yes, here. (Yoda voice) Frontier model wars begun they have. Which is a favorite line, the Clone Wars begun they have. There we go. And honestly, I gotta be honest, I give it like a nine out of ten, eight and a half out of ten. It knows what it’s doing.

Alex Wilhelm: 34:46 It’s pretty good. It’s pretty good.

Jason Calacanis: 34:47 The voice is off, but I mean, I’m not paying a royalty to Disney here, so I don’t want to get in trouble with Disney. But David, why don’t you show us what you built because I do think there is something here to taking… when you get rid of the slot machine nature of these LLMs, it becomes more predictable and your utilization goes up.

David Im: 35:05 Yeah, yeah. The slot machine is not fun when you’re making video and images and they take 30 seconds. It’s incredibly frustrating.

Jason Calacanis: 35:13 I know. You know, the bowling right now is like prompting the videos again and again to get the right results because it’s unpredictable because you know, video generation models are stochastic.

David Im: 35:26 So what we’re trying to do is make an API that is basically a wrapper of like let’s say five or six models to get the production-ready result. So our belief is that once we make these one-piece piece piece, let’s say production level, production ready like API primitives, then once we got the APIs, then we can make our agent one-shot. …about a marketing video very recently.

Jason Calacanis: 36:01 Okay. I’m guessing you have a killer demo to show us.

David Im: 36:03 Yeah. There it is.

Alex Wilhelm: 36:06 Can you see us? Yeah, we could see it looks like your X feed.

David Im: 36:11 Yeah. So, yeah, this is kind of like our UGC API, it’s our avatar API. And then you could do kind of like, so this is basically the same prompt. So it’s a basic prompt with Gemini Omni, Ceedance, and Sume Avatar. And basically what we did is we are basically a router of multimodal models, not only video but image, video, audio, and clipping and everything. We’re a wrapper around all these models. And this is what we got with the same prompt.

Jason Calacanis: 36:41 You’re seeing on the left Gemini Omni, and it looks like a young adult, maybe a 25-year-old or younger, in their apartment in a city, or maybe they’re in high school in the second one.

Brandon Dixon: 36:51 And it looks like a classic selfie marketing, like I’m making a TikTok video about this product, yeah.

Alex Wilhelm: 36:54 Same prompt, you see three different results. Then what happens? What’s next?

David Im: 36:58 So what we are trying to make at the end goal, let’s say like after 6 months, is a video agent that one-shots marketing videos. So our users are brands or marketers want to promote their product and make video ads for them. And right now our first model was for UGC. So as you know, like if you want to make UGC videos with AI right now, you have to combine a lot of different models, let’s say like Ceedance or like ChatGPT image or Groq and everything. And then after that you have to combine those videos to make a long one because you know video generation models only like support up to 15 seconds. It’s like 30 seconds for our like Ceedance’s like next model, but it’s only 30 seconds. So what we did is we made a router of video generation models and image models and audio to like get the like audio persistent to make this kind of videos and then and then finally you could generate up to 60 seconds of consistent avatar videos right now from the same API.

Jason Calacanis: 37:39 Okay. So you put in a script. Hey, I want to promote my new app, it’s called Uber, and you open your phone, it’s rain, you go outside, it’s raining, you can’t walk home, you want to get a ride, and you write the script for this, and then it goes and makes you a bunch of different scenes. Do you tell it what scenes to make in your?

David Im: 38:22 No, actually. It’s only, so for the user, it’s basically just picking an avatar and writing the script.

Alex Wilhelm: 38:34 Got it.

Jason Calacanis: 38:35 The user writes the script or the LLM writes the script?

David Im: 38:37 The user writes the script. The API receives that.

Jason Calacanis: 38:40 Got it. So after I write the script, then each scene gets done three times and then it’s up to me to stitch them together? So I might say, ‘Oh, I like this one where it shows the car, you know, in the pouring rain. I like this one when it shows the guy getting out with his umbrella or putting his jacket, one of them has him put his coat over his head to walk to his front door.’

Brandon Dixon: 39:00 And I just get to essentially vibe code my way around a longer form video, yeah?

David Im: 39:07 100 percent.

Brandon Dixon: 39:08 Okay, cool.

Jason Calacanis: 39:08 Did you have any outputs like that that we can see? I would love to see like where you got to with this.

David Im: 39:13 Okay so, could show the video.

Jason Calacanis: 39:15 Yeah.

David Im: 39:15 So, this is like 16 seconds so you could see that it’s up like more than 15 seconds, but yeah, you can make these kind of videos with 60 seconds.

Jason Calacanis: 39:27 So it’s using multiple ones and then it stitches it together with one clean audio file across it, so a little bit of a hack there.

David Im: 39:34 Yeah, yeah, yeah. Yes.

Alex Wilhelm: 39:35 So around now, yeah. It’s also consistently her face remains totally consistent throughout. A big problem when you’re generating these kinds of like video clips I find is that it’ll for the first like seven seconds of the video it’ll look like the person’s face and then it sometimes gets like distorted towards the end. Like that’s a big problem with… yeah, it’s like the keeping that facial consistency the whole time.

Jason Calacanis: 39:59 What is the website? Do you have a website for this that we can see?

David Im: 40:04 Yeah, you can search sume.com, S-U-M-E dot com.

Jason Calacanis: 40:07 Ah. Oh my god, you got a good domain name. How much did that cost you? Four letter domain name?

David Im: 40:14 Yeah, we got it cheap.

Jason Calacanis: 40:16 Really? That’s a $100,000 domain name.

David Im: 40:19 70% discount from the initial price on GoDaddy. Yeah.

Jason Calacanis: 40:22 Love it. Love it. All right, well this is like a great start. And who’s it for? Who’s the customer you think? Startups making marketing videos or just general D to C marketers? Who’s your customer base currently?

David Im: 40:31 Yeah, so right now we have 20k users and the main customers, especially about 90% of the paid customers, are brands. Our main audience is not on Twitter, but they’re on Instagram and TikTok.

Jason Calacanis: 40:41 Got it.

Alex Wilhelm: 40:43 Yeah, people are trying to flood those spaces. I can’t even tell what’s AI anymore unless like you’re paying attention. I got a lot of shark videos because one of my daughters loves sharks and I’ll do like shark videos with her. And now I’m starting to get AI slop sharks. And you know they start out and it’s like a person on a boat and they’re pulling in a fish and you’re like, ‘Oh my god, this is going to be a shark.’ And then like this ridiculous shark jumps up, eats the fish, the person falls in the water, the shark is jumping in the water, I’m like sharks don’t interact for 90 seconds with a human. This is getting a little ridiculous here.

Jason Calacanis: 41:17 All right, well great job with the startup. Keep grinding and we’ll see you when you have your next update.

David Im: 41:23 Awesome. Thanks Jason.