Why Enterprise AI ROI Fails

Speaker 0:00

There's a lot of conversation, still is a lot of conversation around why you're not seeing AI ROI in the enterprise. Why is it not succeeding? And to me, it was the same problem that you had a great product in these chat interfaces for AI, but they were great product if you spent your time chatting with PhDs all day long, which is what coders do when their code doesn't work. They go to their CTO and say, hey, fix my code for me. The rest of us don't do our jobs that way. The challenge with all these AIs, you know, we've got a bunch of big AI companies. And then, you know, short time ago, you know, a single dude wrote this thing called OpenClaw and put it out in the world. And if an individual can give us something that's that powerful, we know there are going to be a lot of different AI models and applications that are launched over the next several years. And no company can afford to just say today, we're only going to use this one.

Rajiv Parikh 0:52

We really have a great show for you today with Oren Michaels, the founder of Barn Door AI, and who's also the founder of Mashery, and tremendous amount of insights as a multi-time entrepreneur and investor. I think what's interesting about him is that not only did he have a success early on, but he sees that same opportunity in putting his company, Barndoor, into a spot where it can make a difference in this agentic world. And it's not like, oh, I did this in the API world, let me move to the agentic world. It's he has a framework for how he thinks about things, this notion of a control plane. And really, it's this notion of AI needs governance. These agents need to know what to do. And in terms of knowing what to do, they need to be guided. But then the company or people can't be restricted on what tools they use. And I think that's really fundamental to this and a very interesting topic for us to explore. There's other great statements that he had during the conversation of what the product that he built that's for everyone before they get to the enterprise product. It goes along that notion of go to market where you need the ability to sample, try, get started before you move to an enterprise sale. There's him as a person and his favorite director and why he chose him. But I think it's going to be really interesting when you get to that portion of how he admires those who are the root of a tree and those who inspire others and being open and sharing with others. And I think there's this whole notion of advice for entrepreneurs in terms of if you have a great idea and you're the right person to pull it off, and it's so amazing that it changes the world, you shouldn't be afraid to share it. So this is gonna be a great show for you to listen to. Previously, Oren co-founded Mashery in 2006, serving as the CEO until the company was acquired by Intel in 2013, where he subsequently served as chief strategist for the Intel Services Division. A true multi-hyphenate with an electrical engineering degree from MIT, Oren is also a Tony-nominated Broadway and off-Broadway producer, whose credits include Romeo and Juliet and Good Night and Good Luck starring George Clooney. And if that wasn't enough, he even spent time early in his career as the CEO of Groundlings, the legendary Los Angeles comedy improv theater. So some of the takeaways you can expect from this episode: the critical shift from read-only demos to write access ROI in the AI world. Why managing AI agents is the new API economy. And finally, blending technical structure with creative spontaneity. Oren, welcome to the Spark of Ages. Thanks, Rajiv. It's great to be here. Great to have you here. Who knew that you and I both have your investor, Eric Chin, in common? He's a close friend of mine, longtime collaborator.

Speaker 4:02

He's a lovely guy, our Chinny, yes, indeed.

Rajiv Parikh 4:05

Were you involved with the whole thing with uh alpha networks as well?

Speaker 4:08

Yeah, I was a very early member of Alpha. That's how I met Eric. And back when I was first raising money for Mashery, I went to one of the very first Alpha dinners. And Alpha is a networking group that Eric and another VC, Mike Jung, put together. They started it very early in their careers, and it really has given them an unparalleled network both in Silicon Valley and throughout, you know, all the various places. And it this very early dinner, I was raising money from Mashery, and the guy sitting next to me wrote a check for the company at the dinner to invest. And so I was the very first alpha success story. And they dined out on that for a while. And I've been, you know, involved on their alpha group since then.

Building An AI Control Plane

Rajiv Parikh 4:46

Who knew? Yeah, we've been pretty involved ourselves. We actually built the alpha website. Oh, cool. And uh gone to a number of their events, and it's been really amazing. So let's start. What sparked you to start Barndoor?

Speaker 4:56

Well, you know, you mentioned Mashery, and Mashery was an interesting company. We were doing API management. We were the first API management as a service company. It sounds like a really technical thing, but it actually we started the company to solve a non-technical problem. APIs were this way that companies could connect together. And what I found is I was a business guy at a little startup, and I needed an API to be able to be the business development guy and get my little technology built into other companies. And I didn't have the technology I needed, that the technology was all stuck in the data center where some dude had a password to a router and maybe could turn it on if I asked nicely. And I couldn't run a business that way. And so that sparked Mashery. And similarly, a couple of years ago, there's a lot of conversation, still is a lot of conversation around why you're not seeing AI ROI in the enterprise. Why is it not succeeding? And to me, it was the same problem that you had a great product in these chat interfaces for AI, but they were great product if you spent your time chatting with PhDs all day long, which is what coders do when their code doesn't work. They go to their CTO and say, hey, fix my code for me. The rest of us don't do our jobs that way. We don't spend our lives chatting with PhDs. And so what I felt was necessary for AI to really succeed in the enterprise and do the work and help us do the work that we as humans do is what we have as humans, which are two things. We have connectivity, we can actually use these tools, and we have trust. People, you know, our bosses assume that if they let us use Salesforce or Gmail or whatever it is, that we're not going to do stupid stuff. And, you know, the AIs had neither the connectivity nor the trust, which to me was a significant challenge. And so that's what led me to go sort of have that conversation with Eric at Crosslink. And a week later, he and his firm gave us a term sheet to get started. And, you know, here we are. And we now have solved the connectivity and the trust problem. And we're seeing some really interesting adoption of AI.

Rajiv Parikh 6:56

Yeah, I could definitely see that. I mean, when I look at other companies that are talking about governance and security with AI, they're pushing you into their application structure, application to element structure. And it looks like from what you're doing, you're saying, hey, pick the tools that you need. We're going to enable all these components to work together in a way that's trusted and a way that you can govern it. Is that a good way of framing it? I mean, I think about like in the security world, I think about Wiz versus Palo Alto Networks. Like Palo Alto Networks has an amazing environment, but you have to buy all the pieces. Whereas like Wiz came in and said, wait, wait, we could just tell you all the security vulnerabilities in a really straightforward way. And I know you think it's already been done, but hey, this is a way to do it better. And they and they explode with growth.

Speaker 7:42

Right. And I think that, you know, the the challenge with all these AIs, you know, we've got a bunch of big AI companies. And then, you know, short time ago, you know, a single dude wrote this thing called OpenClaw and put it out in the world. And if an individual can give us something that's that powerful, we know there are going to be a lot of different AI models and applications that are launched over the next several years. And no company can afford to just say today, we're only going to use this one. So instead, what you're stuck with is having to deal with governance and visibility and management piecemeal on every single one of these things that you do. It's not something that people really want to deal with. Instead, what we offer is the ability to have that control in one place. And that way, when you bring in the soap, you want to bring in open claw, great. Point open claw at us, and all of those privileges are granted. And you can say, well, actually, OpenClaw is a little different. So I want to lock it down even further and only let it do like I have a thing with OpenClaw, it reads all my news emails every morning and tells me what I actually need to know. I don't want it sending emails. I have other AIs I'm willing to let send emails, but I don't trust OpenClaw to do that. So OpenClaw is very specifically limited to reading emails that have a certain flag saying that they're news and it's allowed to set a flag on them to archive. And that's all OpenClaw is allowed to do. And that way I'm not worried that it is gonna get some idea of what I might want and go off and start sending emails out there.

Rajiv Parikh 9:04

That makes sense. That way you're segmenting, you're controlling it much better. So now think about this. So you've mentioned there's a future where there's a hundred thousand agents that have right access to our most critical systems. So how do we introduce a programmatic equivalent of consequence or liability for agents without completely crippling their autonomy?

Speaker 9:23

Well, you know, that's the important thing is that if you want good things to happen, there is always some risk of bad things happening. And so the idea is if they're not allowed to write anything, they can't do anything terribly useful. So the idea is to break these tasks apart. So we're working with a friend of mine who's also using OpenClaw right now. And he has about 20 different things that OpenClaw is doing for him. And each one, we're setting separate guardrails so that that particular task goes through and identifies the task that it's doing. And for performing that task, a very narrow set of guardrails that really only allow him to do that task are available. But for a different task, it's a different set of guardrails. And I think that that's how you're going to see these agents deployed. If you really want them to be autonomous, they're going to be purpose-built workflows that come from a variety of different places where there's a ton of folks out there providing, you know, workflow capabilities and AI capabilities. And we all want the right tool for the job. There's no one company's going to win that. And so, you know, the idea is to manage these things. And when you get to a point that it's that large, you're doing it programmatically. Unsurprisingly, we have an API for our platform and you know, folks can do it that way as well.

Rajiv Parikh 10:31

Right. So you can choose any tool you want and still have the ability to govern it, to manage it. And you could have this 100,000 agent world and still make sure that there's a level of control to it. You've also pointed out that when companies don't trust their security controls, they water down their AI use cases until they're safe. They keep them stuck in read-only mode and they destroy their ROIs.

Speaker 10:51

It's you know, glorified enterprise search. Absolutely.

Rajiv Parikh 10:54

Yeah. So, I mean, what's an example of a high value right access use case that enterprises frequently shrink away from out of fear?

Speaker 11:01

And so let's talk about how we do our job. So, yesterday I wanted to go through over the course of last several months, Eric and my other investors have introduced me to a bunch of prospective customers. And I wanted to go through and make sure that I hadn't dropped the ball on any of the follow-ups I was supposed to be doing. And you know, in theory, Salesforce is supposed to track that, but we all know that it's not necessarily easy to use. So, what I said to my clubs, I said, look, go and get the Google sheet where I track all of the these introductions. So go grab that Google sheet and then go into my emails and find me every conversation I've had with anybody on that sheet, summarize those conversations, go into Salesforce and check and see if we have opportunities for them, and check in and tell me who the contacts are and what the activities that the rest of my team has done with those folks. So it got all that stuff, it put that into that same Google sheet where I track it. And then I was able to say, okay, with all of this information, you know, who needs to be followed up on? And given what you've seen here, suggest to me how I follow up on it. And all of that got pulled from different places, written into a Google sheet where I could see it all. And then I was able to go through and then I went through and I gave it little tweaks for each of the introductions. And I said, okay, now write me the emails. And it wrote me the emails. I went back and forth, and ultimately I had to create the emails and put them in drafts so I could send them easily. Right. Took me an hour and a half to manage follow-ups with almost a hundred different companies.

Rajiv Parikh 12:25

That could normally take forever and probably gave you the ability to reach people in a way that that you can become more scalable than ever before.

Speaker 12:32

Absolutely. I think I I sent 40 of them follow-ups yesterday, and I did that in my one hour and a half empty block in the day.

Rajiv Parikh 12:39

It was great. So that's a case where you're still interacting with it. Right, right. In the go-to-market world, we live with that all the time, right?

Speaker 12:45

That's called human in the loop, right? That's human in the loop.

Rajiv Parikh 12:48

Do you see this as multiple salespeople doing this at scale? Or do you see agents that a company would create that would be that salesperson that are sending them out? Non-human salesperson sending them out.

Speaker 12:58

There's lots of sales automation tools out there, right? Some have some about AI and some don't. But for me, it's more I want it to react the way I react, learn from how I learn. So, like my little open claw that does my news each morning, I have some rules for it. I say, okay, if it comes from these places, just go ahead and archive. I don't read that anymore. And it's too hard to unsubscribe. If it comes from these folks, I like about a third of the stuff that they write. And so give me a summary of it. Slack me a summary of what you get from it. And when I have a chance, I'll go to the Slack and tell you, you know, what I want. And then I'll send it out to go do, you know, to go find the information. But I'm also doing that with all of my major sales prospects. I figure out what those sales prospects have in the way of news and new things happening. And I have it create for me suggested follow-ups based on the news of the day. So there's lots of different automations, and you know, you're getting into deeper stuff depending on more use of internal data. So most of the customers we're dealing with right now are using a combination of common tools, you know, common SaaS tools like Salesforce and things like that, as well as internal data and internal systems that they've created MCP for. And so with those situations, they're actually taking internal workflows. I've seen one doing preliminary credit determination workflow or insurance claim workflow, right? I've seen another one that is a hotel chain who's sort of setting up for pre-arrivals for guests. They're setting up communications and instructions to the staff based on the information they have both internally from previous days and from the outside world that they're able to pull together and give to the staff. So these are things, you know, it's typically pretty specific. And that's the point. If it's a generalized thing, we've already got a tool for that.

Rajiv Parikh 14:43

No, that's great. I mean, you essentially get to serve a customer in a way that they prefer to be served. And it's very subtly delivered in a way that makes it more useful for folks. So now you've mentioned the notion of enterprise control plane, right? And that was a lot of the basis for how you started Ashri, right? So how does that need to evolve when the traffic itself is non-human? And how do we prevent a systemic pileup when you have tens of thousands of agents simultaneously executing their workflows?

Speaker 15:12

Well, yeah, I mean, certainly there are, you know, back from the days of API world, one of the first things you do is build rate limiting and right. So there's certainly rate limiting, and there's also managing, frankly, managing your LLM spend with how many tokens you're consuming, right? So there are all kinds of reasons.

Rajiv Parikh 15:26

That's very expensive.

Speaker 15:28

There's all kinds of things that we do to help mitigate those things. But ultimately, you also, by having visibility, you're basically saying to an agent, hey, go do this task. And then you're be able to watch through our system, you're seeing what is it doing? What are the MCP calls it's making? What systems is it accessing, what methods is it accessing? We don't see the data itself, but we see that systems are accessing and the methods so that you can look at it and say, does this make sense given what I've asked, or going off and doing a bunch of stuff it shouldn't be doing? And if it's the latter, you're going to start refining your instructions or putting more guardrails up so that it sort of sticks with the program, right? And so ultimately we want to see a lot of things happening, but we want to see the right things happening. We don't want to see a lot of waste.

Rajiv Parikh 16:11

Yeah, that makes a lot of sense. And so you have to build around that, build the right guardrails, you have to build the right indicators. And you pioneered API management, right? The category cashery.

Speaker 16:20

Yes.

Rajiv Parikh 16:21

And you've noted that API agents, there's the new API management, but with the P taken out.

Speaker 16:26

Exactly. We've taken the P out of the APIs a deed.

Rajiv Parikh 16:28

Yeah. So APIs are deterministic and only do what the program would do, while agents can reason, improvise, operate at machine speed, right? So what do you see as the fundamental difference in building that control plane for a deterministic pipe versus probabilistic decision-making entities?

Speaker 16:46

Fundamentally, there's a lot of uncertainty, right? These are things work probabilistically. And so you can expect them to do things that are unreasonable. I think that when you talk about the security concept of least privilege, it really matters with the ages. We intentionally over-provision or over-privilege humans because we expect, I'm, you know, I've been executive, I should not do stupid stuff, right? And so rather than having me come back and ask for permission every time, I'm over permissioned and I'm trusted. And whereas with the agents, it kind of goes the opposite direction. And you generally underpermission them and then underprivilege them. And then you say, okay, let's see how it's going. And when we ask it to do something, are they doing the right things? And if so, we gradually open up those privileges and permissions. And you know, it's interesting as I as I work with these AIs, that workflow is pretty calm. You'll say, go do this. And they'll say, gosh, you know, I tried, but I don't have the privilege to do that. Find me a way to do it. And just like, well, okay, if you're asking to do that specific thing, okay, that makes sense. Or yeah, why are you asking to do that? You shouldn't need to do that based on what you're doing, right? And so it's a combination of both the security folks who are deciding what these things were allowed to do, and the people using them who are saying, Well, my CISO is giving me these sets of privileges. I'm gonna have to figure out a way for my agent to get this job done in spite of what privileges I do or don't have.

MCP Governance And Least Privilege

Rajiv Parikh 18:04

Right. So then I see. So it's interesting. In a previous podcast, we had a person who's a CFO at a security company basically saying it was the other way around. It was like the humans were underprovisioned. We have lots of responsibilities on lots of rules on them, but agents seem to have open access. But maybe what you're seeing in the enterprise world is that there's a lot more caution.

Speaker 18:26

There should be, right? And so we've talked a little bit about MCP, and I'm not sure how familiar your folks are, but MCP is a way to take an API and explain to a language model, explain to an AI what tools it can use and how to use them. So if you start with a Salesforce API, you can do anything in Salesforce through the API. But when you put an MCP in front of it, and it's not necessarily Salesforce who does this, you know, Salesforce runs their API, but anybody can build an MCP for Salesforce, right? So when you build the MCP, you're saying, I'm taking this subset of capabilities on API, I'm exposing them out with you know the ability to do these things, and here's the instructions on how to do it. That's essentially what MCP is. So most corporate MCPs only expose read capabilities because without governance, they don't want to expose right, right? And so they may be well provisioned in the read department. But if you say, okay, update that opportunity, it's gonna say, yeah, I can't do that. And so we have to provide a different set of MCPs that have all these right capabilities, which we can do because we also bring the governance to allow you to decide who and what and under what circumstances, whether it's based on the human identity that we get from your identity provider, like an Okta or Entra or whatever, or if it's a non-human, if it's just the agent itself, but it's not necessarily tied to a human, we have to handle that as well.

Rajiv Parikh 19:44

I see. So the MCP in this case, the enterprise NCP has been the rate limiting factor. What you're doing is you're helping to adjust it by matching it to the level of governance that you want to have on a particular individual.

Speaker 19:56

Right. And you think about something like OpenClaw. OpenClaw has a bunch of security issues that people have talked a lot about, but ultimately we allow you to basically have a very, very sandboxed version of OpenClaw and using MCP, just expose to it the pieces that you want it to use. And we also do that to anything else. So, you know, there was a recent announcement that Amthropic had about the cloud co-work and the new capabilities, but that has no real governance on it either. And so most of the CISOs I've talked to are saying, you know, it's not a whole lot better than OpenClaw, to be honest with you, right? It's prettier and has a nicer shine to it. So, you know, we want both of those tools and everything like them to succeed. And we want that to succeed in spite of the fact that they themselves don't come with governance. And we think that's how it should be. Most CISOs are not going to trust your average AI company to provide governance. Letting the software vendor determine what governance is is not usually a best practice in most CISO offices.

Rajiv Parikh 20:52

It's better to have a single place you can go for that and a single place that controls all those things. So it makes a lot of sense. So here's an interesting one. And this is going to be a little somewhat theoretical. So bear with me. So economic researchers from MIT have identified a systemic risk called the measurability gap, where agents optimize for measurable proxy outcomes, but they completely miss unmeasured human intent, accumulating massive hidden debt. So there's this notion of we can do a tremendous amount of automation now in the way we couldn't do before. But the the previous constraint was what the human mind could handle, right? Our whole economy is built around what an individual or subset of minds can have. But now the ability to verify that becomes the essentially the rate limiting factor. So how does Barnbow's infrastructure ensure that we're just not governing access, but actually verifying the intent and quality of the work being done by these thousands of agents?

Speaker 21:46

Well, there's a couple of reasons. One is that when you look at how humans work, we don't generally drill into their brains and look inside and see what they're up to. They have work product and they have how they do the work. So we give them access to tools, we see how they're Using those tools and we see what they create with them. And we do the same thing. So with Barn Dor, you have a full set of logs and information that show all the various ways that AIs are interacting with the underlying tools. And so you can look and see if they're being incredibly inefficient or not. And often they are being very inefficient, and you can give them different ways of doing it. But you know, as you talked about that, I sort of saw another problem that I think is a parallel one, which is that the chat interface that most people are using today, because most people are not yet truly using agents that actually do jobs. The chat interface does a couple of things. That one is that it is a great tool for searching. So it finds you lots of things, but it also spends a lot of time telling humans what they should go do. And the challenge with this, I have a friend who's a marketing executive, and she was very frustrated because she she felt she wasn't doing a good job in her job. And I said, Well, why do you think that your company is successful, you're doing well? And she goes, Well, there's just so many things I'm not getting done. Like, you know, are you talking to an AI that's telling you all the things you should be doing? And she goes, Well, yes, of course. I'm like, no one can do all the things, right? And AIs think you can do all the things. You can't do all the things.

Rajiv Parikh 23:06

Or in you nail it. Every time you ask for something, it gives you this well-structured response. And then it asks, I can go deeper. You're going down this rat hole where you keep going deeper and deeper.

Speaker 23:17

No, and and yet before AI, we as humans always knew we weren't doing all the things. And that was just fine. So I think that part of what you really want to do is prioritize it and also figure out where the low-hanging fruit is. And I was on a panel at a conference last year with a journalist, uh, Quentin Hardy, was leading it. And the way he put it was he said, what you want is you want your AIs to do the jobs that if your kids had your job, you wouldn't want them doing. Right. It's like, get the stuff out of my life that is unpleasant enough that I wouldn't want my kid doing that if he had my job. And I thought that was a good place to start.

Rajiv Parikh 23:54

I think that's totally a great place to start. Do all the stuff that I hate doing and don't be in a spot where you just feel inadequate because you're not doing it up.

Speaker 24:03

Absolutely.

Rajiv Parikh 24:04

All right. So since AI agents can make dozens of database calls in seconds, what specific architectural shifts must an enterprise make to move from uh who are you access to uh what are you trying to do access?

Speaker 24:18

When you think about how access has been controlled up until now, and you look at these great companies like Okta, who do identity, it really is you are this person, you are still employed here, therefore you get access. And maybe you get access with some restrictions, you only get to go to this part, you only go to that part, but it's all based on who you are and without the context of what you're trying to do. And what you have to be able to do with the AI, and you can do this with barn door, is understand the context that if the AI is performing a particular task that it identifies and within the scope of that task, it's allowed to do more, then you can create policies that intercept that and that will allow it for that task. But it would be specifically an agent that's operating doing this specific task. You're like, okay, that agent that is charged with doing that task. And I think that that this is one of the reasons we're gonna have lots of agents is for exactly this purpose, is that you want to have the ability to be very narrow and say, okay, this agent's job is to go do these things, and when it does that, it's great. And then there's probably some other agent that goes and tells this agent what to do. And that's fine.

Rajiv Parikh 25:26

Right. And so you create this network or layered set. Are you finding that when you're talking to your customers or your potential customers, prospects, that this notion of who are you versus what are you trying to do? Are you leading them into that or are they thinking about that already? Basically, your technology helping to lead them there.

Speaker 25:44

The folks we're talking to for the most part have a pretty strong idea of what they want to do. What we haven't really had yet is gosh, wouldn't it be nice to just have MCP in our company and see what happens? Very little of that. I think it's coming. What it's more is we have a very specific project. It probably involves both internal and external systems, where the internal ones we've built MCPs on the external ones, you know, we're using existing ones or we use barn doors. And there are specific workflows and specific tasks that they expect to do that in order for those things to happen, access is necessary and governance is necessary. And so when we've done these and the deals that we've won, these folks have had a pretty strong idea of what they're doing. And it's across a broad range of companies. It's you know, hospitality, it's you know, you name it, it's all over the place.

Rajiv Parikh 26:32

And they have this notion of like it's not just the who are you, right? Which is a security. Right.

Speaker 26:36

Oh no, absolutely.

Rajiv Parikh 26:36

It's not just the no, no, I have to make sure I'm enabling the what do you do? What are you trying to do? And they're coming in with that framework of the case.

Proving Value With Venn.ai

Speaker 26:43

Right, because the AIs don't have internal governance. I'm a human, I know what stupid things are. AIs don't know from stupid, right? And so they will interpret the words you gave it in a probabilistic way and go attempt to go do that by whatever means that it can muster. That requires governance.

Rajiv Parikh 27:01

That requires governance. So you've dubbed 2026 the year of the ultimatum for the C-suite. Either keep AI read-only and miss out on great ROI, or give it right access and face significant or maybe unprecedented risk. So for a CEO listening right now who's terrified of taking the leap to write access at scale, what is the very first low-risk, right use case you would recommend they automate today to safely prove the value of an agentic workforce?

Speaker 27:29

Well, there's a couple of things they do. So we talked a lot about Barn Door, which is our enterprise product for companies to do this with a lot of detail. But we have a second product called Venn. And Venn is the single user, Venn.ai, and it's the single user version of it, which you can use in your work, but you can use it personally as well. And so you can basically go to Venn.ai, you sign up, and you know, I have it talking to my personal Gmail, my personal calendar, you know, various personal things. And I ask it questions and I have it put things on my calendar, and I get used to interacting with these things from cloud. It finds stuff for me I can't find. And the the magic is when it's doing it across different tools. So you're able to say, go find that document, go find this email, do these things, create a new document. You're able to tell it to actually take action across multiple things at the same time. And I encourage that because once you start doing it personally, where the risk is perhaps lower, you're going to wonder why it is the people in your company can't do this for real. And that's why we put Ven out. We put Ven out as a means of letting people start to experiment with this and really get an idea.

Rajiv Parikh 28:36

They don't need the full enterprise version to get started, which I think, you know, from a go-to-market standpoint, I think is brilliant. You know, I've asked my team about this. We're a marketing firm, but we do a lot of AI development as part of helping to do the work that we do faster, better, higher quality, et cetera. And I asked them, you know, how do you buy what you buy? How do you guys decide on what tools you could use? I give them wide berth to play because this is moving so damn fast. And I'm like, Do you ever talk to an enterprise salesperson? No. They see a bunch of videos, they read a bunch of things, they go test a bunch of things on the website, they look at reviews, and then they go play. And I think that's a really smart way to go is not get yourself so slotted in the enterprise category. Give people the chance to get value right away.

Speaker 29:24

Right. Well, it's also fun because if you are open claw curious, as I am, it's really very seamless. Whether you're running open claw on like a Mac Mini or something that you have in your office, or if you're running it virtually through someone like DigitalOcean, it's super easy to just connect it right into that and off you're going. And that allows it to talk to whatever you want and allows you to be very, very careful about what it can and can't do and grant it more privileges gradually.

Spark Tank Broadway Innovation Quiz

Rajiv Parikh 29:48

That's a great way to get it started and make sure it doesn't go out of control. Yep. Awesome. Well, thank you so much for that, Oren. We're now going to move to what we call the Spark Tank.

Speaker 29:58

The Spark Tank. Okay.

Rajiv Parikh 29:59

Welcome to the Spark Tank. Today we're joined by Oren Michaels, the co-founder and CEO of Barndoor AI. But Oren is the ultimate multi-hyphenate. While he holds an electrical engineering degree from MIT and was a key strategist at Intel, he's just as comfortable in the wings of a theater as he is in a server room. From his time as the CEO of the legendary Groundlings Improv Theater to producing Tony nominated Broadway hits, Oren understands that the most successful productions, whether in tech or onstage, requires a perfect blend of structure and spontaneity. Today we're putting that rare intersection of technical precision and creative vision to the test. We aren't just looking at show tunes, we're looking at the revolutionary business pivots and technical firsts that turn the great white way into a global economic powerhouse. So, Oren, are you ready to prove your agentic intelligence is as sharp on the history of the stage as it is in the security of the cloud?

Speaker 30:53

I cannot wait. I cannot wait.

Rajiv Parikh 30:55

Here we go. Well, behind you, you do have your good night, good luck, uh Broadway. I do.

Speaker 30:59

I do have my poster back there, yes.

Rajiv Parikh 31:01

We are going right to that world. Who knew you would have that on the spark of ages? So here you go. In 2016, the original cast of Hamilton achieved a revolutionary feat in labor relations that changed how Broadway intellectual property is valued. What was the outcome of their negotiation? I'm gonna give you four choices. A, they secured a profit participation stake in the show's long-term royalties. B, they were granted the right to block any future casting choices for their roles. C, they received lifetime passes to every Disney owned property. Or D, they became the first cast to be paid entirely in Bitcoin.

Speaker 31:37

It was actually A, and it was a really big deal. It's a big question across Broadway. So you have when you create a new show, the initial cast, how they perform that kind of becomes the model for every cast beyond that. And so they're creatives as well. And historically, those casts have not gotten participation. And this was a first, and it was interesting. Of course, Lynn Dunwell was both a performer and the writer. I'm not sure if he was one of the producers, but he's a performer and a writer. And he felt that they should have this right. And so he actually helped coach the rest of the cast when they went to the lead producer, Jeffrey Seller, to get this done and they got it done.

Rajiv Parikh 32:14

That's amazing. Totally now that actors are traditionally work for hire. The Hamilton cast argued that their workshops and improvisations help create the characters and the brand. And by winning the share of the profits, they moved from being employees to being equity partners in a multi-billion dollar entertainment property.

Speaker 32:30

They did.

Rajiv Parikh 32:30

All right. Great answer for that. Number two, the 1943 production of Oklahoma is credited with many firsts, but its most significant contribution to the business of Broadway was the invention of the new vertical revenue stream. What was it? A. The first tiered premium seating pricing model. B the first original cast recording ever released as a multi-disc set. C. The first nationally televised making of documentary to drive ticket sales. D. The first use of corporate merchandise booth in the lobby.

Speaker 33:03

Huh, that's a good question. I think of Oklahoma as being the first true, I guess Show Mo was the first true book musical, but Oklahoma really brought that revelation past. Gosh, I'm gonna say it's not there's there was no making of.

Rajiv Parikh 33:17

Yep, so you're on the right track, musical.

Speaker 33:20

Yeah, so the the cast recording, I would imagine.

Rajiv Parikh 33:22

There you go. B, that's the correct answer. So before Oklahoma, people didn't take the show home in high fidelity. So Decca Records took a gamble on recording the cast, creating a recurring revenue model that allowed a show to be profitable worldwide, even if it was sold out in New York City. Great instincts, your show instincts are really showing through. Here's number three. Before 1957, the book writer, director, and choreographer worked in silos, which show fundamentally disrupted this power structure by legally establishing the role of director-choreographer, ensuring movement was integrated into the code of the script rather than just decoration. Okay. I found these hard. So hopefully you get this.

Speaker 34:01

I'm sure this will be hard as well.

Rajiv Parikh 34:02

Here we go. A guys and dolls. B Fiddler on the roof. C West Side Story or D cabaret?

Speaker 34:11

I'm gonna say West Side Story. Uh you you sure? No, but I'm I'm gonna guess that.

Rajiv Parikh 34:16

Okay. What was your logic in picking Westside Story?

Speaker 34:18

So I believe Cabaret was more recent than that. And I wouldn't see Fiddler as such a big choreograph show. So it made more sense to pick up the West Side story.

Rajiv Parikh 34:30

Well, you nailed it again. Correct answer, C. Jerome Robbins demanded total control of the direction and choreography of West Side's story. This merger of roles meant that for the first time, dance was used as the primary storytelling vehicle for plot and character development, not just a break in the action.

Speaker 34:45

There you go.

Rajiv Parikh 34:46

So you got all three right. Congratulations.

Speaker 34:50

Thank you. I'm not embarrassed. I'm glad to know.

Career Twists And Intel Lessons

Rajiv Parikh 34:53

No, there are times where I've had other folks here and they get all three wrong. And I feel bad because we didn't, I didn't, we didn't get them the right questions. So I love that you nailed it and provided us a really interesting nuance into your background. Okay, so let's talk about what sparked you. So did you always know you wanted to work in technology? Was there like a specific moment that sparked you, your sparked your passion? How'd you discover it?

Speaker 35:15

No, not really. I mean, I grew up, my mom was a very early computer programmer on IBM System 360 and such. And so I was pretty good at math and that sort of thing. So she started teaching me to program with punch cards on an IBM 360 when I was in elementary school. And, you know, one thing led to another, and I progressed with that and did a bit more and a bit more. And actually, during that same time, I started doing theatrical lighting design and technical directing and getting involved in the theater. What I really wanted to do was go become a professional lighting designer for theater. That would have been what I would prefer to do. But my parents were unwilling to support that particular educational path, but they liked the educational path that involved this MIT degree. So that was the path I ended up taking. But I did a lot of theater. I did, I think, a hundred shows as an undergrad. And that's why I've sort of stuck with it, right? I've weaved in and out of it back in the ground these days. And since then with some nonprofit stuff and producing a few things on the side while I was running tech companies. And then when I got to New York after selling the asteroid, there's a group of about a dozen or so of us who straddle the tech world in Broadway and we all know each other. And a friend of mine introduced me into that group. So I started playing it in Broadway.

Rajiv Parikh 36:23

Was it the going to MIT? Is that what got you more into technology? Or was it when you're in the theater and you were doing the technical aspects?

Speaker 36:29

No, but it was all. I mean, I was a tinkerer. I was always a tinker, right? And so that would involve taking things apart. That would involve, you know, building things growing up. And and I think what sort of evolved is it was a weird path. So I came out of undergrad, spent two years writing assembly language code, hated it. Went to business school, came out with this degree that was destined for the entertainment industry. You're in LA at UCLA. Yeah, UCLA. So ran the groundlings for a year and realized that running a nonprofit theater was not all that necessarily what I wanted to do with my life. And so I got married at that point. My new father-in-law had a little manufacturing company. And he had gotten sick. He got Parkinson's and had to sell the company. And it was as small, it was an eight-person company that manufactured pollution control equipment. So I came in and helped him sell it. So now I had sold a manufacturing company. I got hired to sell another one. So I went into another manufacturing company and did that and got that company sold. And along the way, you know, sort of met a few people and wound up consulting for a third manufacturing company. And the guy who owned that company also happened to own a lot of wine who decided to sell it on this new thing called the internet. And that was back in '97, I think. So I started advising him and ended up going in and running that for him, a company called Winebid.

Rajiv Parikh 37:45

Wine bid, you and I have something in common there. I worked with Russ Mann at Wine Bid.

Speaker 37:50

That was after my time, but I was there for a couple of years, and that sort of got me into the internet thing, right? And so suddenly I was now sort of bricks and mortar with it. We were running this online auction company, right? And so that was early, that was 98, 99. It was very early days for internet stuff. And once I got done with that, it was 2001 and there was a there was a bust on. And so I got involved in another technology company as COO and ran that for a while. And you know, this other opportunity again. I sort of moved more into more tech and less, you know, physical goods during that period and then the mastery thing. So it wasn't like intentional where I said, I'm gonna go into the tech business. It just sort of evolved that way. It was a lot of twists and turns.

Rajiv Parikh 38:33

That's pretty cool. So after successfully scaling and selling mastery to Intel, you experienced a massive culture shock, right? You were a scrappy startup, failure is encouraged. Now you're in a semiconductor company where it's a billion dollars plus to build a chip and release it. Can't screw up on anything before a tape out. No mistakes, no failures. So how how did you navigate?

Speaker 38:54

Yeah, not very well. So yeah, it was an interesting concept. So I will preface this by saying Intel is an incredibly ethical company. They pay us everything they said they would. They took great care of my team and of me. So I have no complaints about Intel at all. But it was not a good home for a software company. We were a sales-driven culture. Intel is not that many true thousands of salespeople, but they're not all writing orders, right? They're they're influencing sales. It's a much different kind of motion.

Rajiv Parikh 39:22

They're doing design win kinds of stuff.

Speaker 39:23

Yeah, there's a much different kind of motion. And we were sort of a fish out of water there. And so in these large companies, the bureaucracy is very significant, and understanding, you know, how to get things done was very tough. And we were purchased, the our deal closed in the middle of them changing CEOs. And the new CEO came in a week after our deal closed. So it wasn't really his deal either. The whole thing was complicated that way. And ultimately, what I did, I felt there was a path for us, even inside Intel, but I didn't know how to get things done at Intel. And the sort of the coin of the realm in Intel is headcount and budget. And the more of that you have, you know, the more power you have. And so every time I was suggesting something, people would take that suggestion as if I was looking for more headcount or more budget. I didn't want either of them. I had a countdown clock. I knew I was out of there in two years. And so I finally went to my boss's boss and I said, look, the only way I could be helpful here is if you take away all of my headcount and take away all of my budget, I'll report to you and we'll try to make this work. But I wanted to be explicit that I'm not trying to get these things. I don't want them. And let's see if we can make a go of it. And we did that for a while. It, you know, it still was not all that good for me. So ultimately, I left a little bit before my my two years were up. They they sort of released me from that. And the company stayed there another year or so. They sold it to TIPCO and then it's still being used today.

Rajiv Parikh 40:48

Yeah, it's pretty amazing. The part about Intel that I think was probably interesting is that you saw the enterprise environment. Oh, very much. Zero tolerance for failure. Is that influencing some of what you're doing at Barndoor AI? I mean, you're trying to prevent catastrophic failures, AI agents, you know, going out of control.

Speaker 41:05

It's a different kind of zero tolerance for failure. So, you know, the due diligence that Intel put us through was very much a zero tolerance for failure. But I don't know that they really got a solid understanding of what we were like because they mostly focused on zero tolerance for failure. I like to say technical debt is another word for customers. And so I want there to be technical debt because that means we're moving quickly to meet the needs of our customers and we'll pay that debt down when we need to. But ultimately, it's important to move, particularly in the world of AI that's moving as fast as this is. And back then it was mobile, which was the thing that was driving all these API things, right? You had to move incredibly quickly. You were going to have technical debt and you were going to have organizational debt. You know, all these things were going to be something that you'd have. And we did those things. And you cannot do that when you're building a fab. You cannot do that when you're building a new chip. This is not the same kind of concept.

Rajiv Parikh 42:02

It's pretty amazing. So it enables you to think about it, how to build the next company, but also that environment in which you're working with folks who think like that. So it's interesting.

Speaker 42:10

Absolutely. And it does bring it to us, though, because we were surprised among our first major POCs have been financial institutions and other super privacy places where we're doing, you know, private cloud or on-premise and talk to and all these things. We had to have that out of the gate. We could not show up with the here's the version we did in our garage and let's see how it goes, kind of thing, right? And that was why I went to Crosslink with the ask I did for the size of C route I asked for, because I knew this was not a company that there was a, you know, a $5 million version that you could come out with an MVP and have anybody credibly use it. And that turned out to be a good decision.

Mentorship And Founder Advice

Rajiv Parikh 42:48

Makes a lot of sense. Yeah. Something more about you. You got your degree in electrical engineering from MIT. So do I. You've admitted you haven't written code in a very long time. It sounds like though you're playing with tools that play with code. So how does your foundational engineering mindset give you a unique lens to evaluate new paradigms without getting bogged into the code?

Speaker 43:06

The education I got at MIT, I had taken some computer science classes at another university when I was in high school. And it was a very different concept. So at the other university, it was here is a problem, here is the algorithm to solve that problem. Today we're learning Fortran, write that algorithm in Fortran. And that was sort of what computer science was. Whereas at MIT, it was here's the problem, figure out how to solve it and write it in Lisp, a language nobody uses for anything, pretty much, that is really, really hard to use, because it trained you to think of how you were solving the problem and doing so in an elegant way, as opposed to just, well, I'll take this and translate it into that language. And so it really gave you a means of thinking and of analyzing and of taking apart problems and really de emphasized which language was the flavor of the day. And so while I'm not adept at coding the languages that the kids Use today, I am adept at sort of dissecting a problem and figuring out what a way would be to tell a computer to go solve that problem, even if I'm not the right one writing the code. And so, like, you know, today we were, I was working with one of my engineers on how we're integrating Ven into OpenClaw. And it was like we were talking through how he was doing it. I'm like, but here's what the user is going to want, and you should be able to do it this way. And he was like, Yeah, that would work too. And so I'm able to have those conversations even if I'm not reading this code.

Rajiv Parikh 44:29

More fundamental ones, right? Yeah. More to principles. So usually we ask our guests to name a historical event or person or movement that inspires you. And you answered Steven Sonnheim.

Speaker 44:38

I did.

Rajiv Parikh 44:39

What about him? Lights you up.

Speaker 44:40

Well, in addition to the fact that I love his shows and that the first Broadway show I ever saw was the original cast of Sweeney Todd, and my Broadway production company is called Fleet Street Productions as an homage to that. What I liked about him is he was really, you know, we talked about Oklahoma that sort of brought a new style of music in and such. But he's written so many very, very different shows that have taken different kinds of music and different, you know, you look at something like a little night music, you can compare that to a Sweeney Todd, you compare that to Merrilly, right? These are just such incredibly different concepts, and he was able to do all of them. So that's one thing. And he's also an inveterate or was an inveterate creator and solver of puzzles. And I do like a good cryptic, and I like, you know, some of the stuff that that he was into. So I like that. And he also was just an incredible mentor to people who came up, whether it's you know, Lynn Manuel or Jonathan Larson who wrote Rent. And there's a lot of people in this business who I've had the pleasure of, I haven't I never met Sanheim myself. I saw him at events a couple of times, but never met him. But so many people who I admire and love to work with today worked directly with him and got so much out of doing it that you know, being a mentor of that sort and being comfortable enough in your own skin that you can hold others up is just something really to admire.

Rajiv Parikh 45:59

I appreciate that. Thank you for that. If you could be guaranteed to be really good at one thing you're currently terrible at, what would you choose?

Speaker 46:05

Face recognition. I cannot recognize people's faces. So if I run into someone out of context, I will not know who they are.

Rajiv Parikh 46:12

All right. But we're not going to see you wearing those Facebook glasses, right? The Ray Bans.

Speaker 46:16

I don't wear them, but that is an awesome use case for it because it would be life-changing for me.

Rajiv Parikh 46:21

It would be. It would be. What's the most interesting thing you've learned recently from a random internet rabbit hole?

Speaker 46:28

Oh gosh, a random internet rabbit hole. Well, I'm rediscovering because I'm getting open claw up and running. I'm rediscovering terminal, which I haven't used in forever. So I've had to go back and figure that out. But a true random one, I'll go into rabbit holes about people or events in the news and just find, you know, all kinds of stuff that tie people together. I really enjoy doing that.

Rajiv Parikh 46:50

Yeah, I do as well. It's like the first time I I could now like watch a sh movie or something and just tell me three things about this person and yeah, exactly. And I'm like, how do I remember this person? And now I could do it and not lose track of what I'm doing. So it's pretty cool. If you had to choose a theme song that plays every time you walk into a room or an what would it be and what energy are you trying to bring?

Speaker 47:11

Oh gosh. You know, I listen to a broad range of music. I don't really have a hype song that I've or you know, walk-on song I've ever done. It's a tough question. I don't think of music.

Rajiv Parikh 47:21

Imagine you at your next SKO and they're getting ready to announce and put you on stay.

Speaker 47:28

Don't know the song yet, but we have composers working on new songs for us. It's a great pair of composers called A Great Big World, and they're doing our musical Silver Linnings playbook, and we're starting to get some really cool songs from them. So I'm optimistic that it will be one of the songs from that musical.

Rajiv Parikh 47:44

That's gonna be your hype song. If you could add one subject to a high school curriculum that wasn't there when you attended, what would it be and why?

Speaker 47:51

Well, if I could add it back, what was there and doesn't really seem to be as much anymore, it would be shop class. I learned so much in learning to build things. I think we all should know how to make stuff. I think that that's super important. It's not really part of the curriculum anymore.

Rajiv Parikh 48:05

It's a great rotation.

Speaker 48:06

One minute you're in home ec, another minute you're working with wood, another with metal and yeah, it was it was super important for everything I ultimately did in both personally and professionally to understand how to build stuff. And so I think anything that causes us to actually go build and make things or repair things and really understand that the physical world around us, I think is is really, really important.

Rajiv Parikh 48:31

I agree. I mean, it's it's good to understand how these trades work and it's just a great way to navigate life and it gives you these amazing references to draw upon. If you could go back and witness, but not change what an ordinary day from your past, what day would you pick?

Speaker 48:46

You know, I would go back to when I saw my first Broadway show. I would go back and see the original cast of Sweeney Tatica, and that was just transformative. That was great.

Rajiv Parikh 48:54

Probably changed things for you in a fundamental way.

Speaker 48:56

That was like it opened my eyes to a whole different concept of what art and theater and things could be. In which way? It was one of the very first shows that really transformed a theater technically. And I was into you know, building stuff and lighting stuff and such. And it was an order of magnitude more technically sophisticated than anything that had ever been on Broadway or you know, theaters. And it was just incredible to behold.

Rajiv Parikh 49:19

That's great.

Speaker 49:20

And it became the model. Now we now we see it all the time. But many of the things we take for granted in theater right now owe their origin story to that, to that production.

Rajiv Parikh 49:29

Amazing. Final question: what one piece of advice you would give a budding young entrepreneur?

Speaker 49:36

Execution's everything. If your idea is so terrible that you can't tell people about it and worry they're gonna steal it from you, it's probably not that great of an idea. You should have a reason why you are the right person to go do that. And if you are the right person, you execute, you'll succeed. And then, you know, choose your co-founders wisely.

Rajiv Parikh 49:54

I love that. Great, great advice, Warren. So thank you for being on the show today. Uh appreciate getting to know you, introducing you to our audience. So really appreciate the time. You you had a lot of great insight about how you look at the world, how you look at investments, how you look at the theater, and just really appreciate having you here today.

Speaker 50:13

That was a fun show. I like your format. I was something different, which is a good thing to have in this world.

Rajiv Parikh 50:19

I have to be a little bit different and special. Thank you.

Speaker 50:21

Thanks, Reggie.

Rajiv Parikh 50:27

All right, thanks for listening. If you enjoyed the pod, please take a moment to rate it and comment. You can find us on Apple, Spotify, YouTube, and everywhere podcasts can be found. This show is produced by Anand Shah and edited by Laura Ballant, production assistant by Taran Talley. I'm your host, Rajiv Parik from Position Squared. We're an AI native growth marketing company based in Silicon Valley. Come visit us at position2.com. This has been an F Funny production, and we'll catch you next time. And remember, folks, be ever curious.