Oren Michels is the founder and CEO of Barndoor.ai, the first and only Control Plane for the agentic enterprise. Previously, he co-founded Mashery in 2006 and served as CEO until Intel acquired the company in 2013. When it was acquired, Mashery-powered APIs were used by over 350,000 active developers in over 100,000 active applications, and counted among its customers many of the largest e-commerce, media, and data companies in the world. He is an entrepreneur, investor, board member, and advisor to technology startups in the US and Europe and has made angel investments in several successful companies including Uber, Pebble Post, Addy, Navdy, and eero.

Highlight Bullets

> Here’s a glimpse of what you would learn….

In this episode of the Ecomm Breakthrough podcast, host Josh Hadley speaks with Oren Michels, founder and CEO of Barndoor AI, about the growing role of AI agents in business operations. Oren explains how AI agents can autonomously perform tasks within systems like Shopify, Amazon, and Slack, while emphasizing the critical need for governance and trust. He introduces Barndoor AI as a control plane that enables secure connectivity and policy-based guardrails, preventing unintended actions. Practical use cases include email management, JIRA ticket handling, and financial forecasting. Oren advises listeners to start small, experiment with multiple AI tools, and develop strong problem-solving skills.

Here are the 3 action items that Josh identified from this episode:

Timestamps:

00:00:00 The Problem of AI Governance
Oren discusses lack of governance in current AI systems and the risks of AI agents forgetting instructions.

00:00:30 Podcast Introduction & Guest Background
Podcast is introduced, and Oren Michels’ background and achievements are highlighted.

00:00:44 The Rise of AI Agents in E-commerce
Josh frames the future of e-commerce as dominated by AI agents and introduces Oren as the guest.

00:02:06 Oren’s Perspective on AI Agent Adoption
Oren explains the rapid and slow pace of AI agent adoption, especially beyond coding tasks.

00:03:02 What is Barndoor AI?
Oren introduces Barndoor AI, focusing on connectivity and trust for AI agents in business systems.

00:03:40 How Barndoor AI Works
Details on how Barndoor AI enables granular control and governance over AI agent actions.

00:05:45 Security and Guardrails for AI Agents
Discussion on security risks, both from bad actors and unintended consequences by legitimate users.

00:06:33 Difference Between Barndoor and Other AI Tools
Oren explains how Barndoor adds governance missing from tools like OpenClaw and Claude.

00:09:24 Use Case: Email Management with AI Agents
Oren shares how he uses AI agents to manage and triage his daily email load efficiently.

00:12:04 Why Governance Matters in AI Actions
Explains the importance of restricting AI actions to prevent mistakes, especially in sensitive tasks.

00:13:00 Custom Rules and Granular Policies
Barndoor allows highly specific rules for AI actions, such as price-based restrictions in e-commerce.

00:13:58 Use Case: JIRA and Finance Automation
Examples of using AI agents for JIRA ticket management and automated financial reporting via Slack.

00:16:48 Enterprise Use Cases & E-commerce Optimization
Barndoor’s enterprise clients use AI for handling sensitive data and optimizing Amazon listings seasonally.

00:19:08 Customer Service and Contextual Communication
AI agents help draft personalized emails by pulling context from Salesforce and previous communications.

00:20:40 AI Agent Adoption is Still Early
Oren emphasizes that AI agent use is in its infancy and encourages experimentation in low-risk areas.

00:22:40 Personal Use Cases for AI Agents
Josh and Oren discuss personal productivity applications, like sports team management and scheduling.

00:24:14 The Evolving AI Tool Landscape
Discussion on the rapid evolution of AI tools, the importance of using multiple models, and specialization.

00:27:47 Future of AI in Business Operations
Speculation on the future: specialized AI tools for each business function, governed by platforms like Barndoor.

00:31:00 The Importance of Problem-Solving and Prompt Engineering
Success with AI depends on defining problems and giving clear instructions, akin to prompt engineering.

00:33:46 Actionable Takeaways for Listeners
Josh summarizes three action items: start experimenting, document processes, and stay flexible with tools.

00:36:44 Book Recommendation: Why Computers Think
Oren recommends a book that explains the probabilistic nature of AI and why it sometimes fails.

00:37:34 Favorite AI Tool and Personal Use
Oren shares his favorite AI tools and how he uses them for both work and personal learning.

00:38:49 Who to Follow: Aaron Levie
Oren recommends following Aaron Levie for insightful commentary on AI and business.

00:39:28 Where to Learn More About Barndoor AI
Oren directs listeners to Barndoor AI’s website and their personal product, Zenni, for hands-on experience.

00:39:45 Podcast Wrap-Up
Podcast concludes with thanks and a call to subscribe and leave a review.

Resources mentioned in this episode:

Tools and Websites
“OpenClaw“: “00:00:00”
“Barndoor AI“: “00:03:14”
“Superhuman“: “00:10:12”
“JIRA“: “00:14:32”
“QuickBooks“: “00:15:17”
“Plaid“: “00:16:14”
“Zapier“: “00:30:10″People to Follow
“Aaron Levie“: “00:39:49”

This episode is brought to you by eComm Breakthrough Consulting where I help seven-figure e-commerce owners grow to eight figures.

I started my business in 2015 and grew it to an eight-figure brand in seven years.

I made mistakes along the way that made the path to eight figures longer. At times I doubted whether our business could even survive and become a real brand. I wish I would have had a guide to help me grow faster and avoid the stumbling blocks.

If you’ve hit a plateau and want to know the next steps to take your business to the next level, then email me at josh@ecommbreakthrough.com and in your subject line say “strategy audit” for the chance to win a $10,000 comprehensive business strategy audit at no cost!

Transcript:

Oren Michels 00:00:00  There is no governance built into any of those systems. And to the extent it is, I use open claw every day and every day I’m reminded of how it’s supposed to act because it forgets forgets its thoughts. And of course, but things that you have told it before are not necessarily retained. And so if you were to say don’t ever do X, which I have said on various occasions, it will still try to do X, because there’s a lot of context that comes before, and it’s doing its best to come up with the right answer every time.

MC 00:00:30  Welcome to the Ecomm Breakthrough podcast. Are you ready to unlock the full potential and growth in your business? You’ve already crossed seven figures in sales, but the challenge is knowing how to take your business to the next level.

Josh Hadley 00:00:44  What if I told you that in the next few years, your biggest competitor on Amazon won’t just be another brand, but thousands of AI agents making decisions faster than any human could? And the real question is, are you going to be the one controlling those agents or competing against them? Welcome to the Ecomm Breakthrough podcast.

Josh Hadley 00:01:00  I’m your host, Josh Hadley. I scaled my own brand from 0 to 8 figures in sales, and now my mission is to take it to over nine figures on my journey to nine figures. I bring you the unfiltered conversations with the smartest minds in e-commerce. Past guests include Ezra Firestone, Kevin King, and Michael E Gerber, author of the E! Myth. Today, I am super excited to introduce you all to Oren Michels. Oren is the founder and CEO of Barndoor AI, the first and only control plane for a genetic enterprise. Previously, he co-founded Mashery in 2006 and served as CEO until Intel acquired his company in 2013. When it was acquired, Mash repowered APIs were used by over 350,000 active developers in over 100,000 active applications, and counted among its customers many of the largest e-commerce, media and data companies in the world. He is an entrepreneur, investor, board member, and advisor to technology startups in the US and Europe and has made angel investments in several successful companies including Uber, Pebble, post, Adi Navidi and Eero.

Josh Hadley 00:02:02  With that introduction, welcome to the show.

Oren Michels 00:02:04  It’s great to be here. Thanks for having me. Josh.

Josh Hadley 00:02:06  Oren, it’s great to have you here on the show. You have a fantastic track record behind you. And ultimately, like, we can see how quickly the world is changing right now, especially in the e-commerce space. And, you know, you and I were talking beforehand, but the speed at which agent e-commerce is going to be kind of coming in to the e-commerce space, I think, I don’t think anybody’s prepared for. And those that aren’t dabbling in the, in the AI space are going to be left behind. Is that the same sentiment that you kind of share here? Oren.

Oren Michels 00:02:34  Yeah, I think that we’re you know, I say that AI is moving very quickly and very slowly at the same time. So while we’re hearing a lot about agents and we’re hearing a lot about these things, most of the active use historically and really up until the last month or so, has been with folks writing code.

Oren Michels 00:02:50  You know, coding agents were really the first agents that that that were widely deployed, were now starting to see that move into other, other forms of knowledge, work in and around companies. And I think it’s going to change a lot of how we do what we do.

Josh Hadley 00:03:02  Yeah, I totally agree with that. Tell me a little bit more about barn door AI and what you guys are doing. And again, how that’s going to be playing a key role here in the a genetic commerce situation that we’re moving towards.

Oren Michels 00:03:14  Absolutely. So when I talk about an agent, everybody has a different definition of it. In my world, an agent takes action, it rights, it actually changes something in a system of record. It creates something, it updates something. It does work that we as humans otherwise would have to do. AI is also very good at suggesting things for humans to do. AI is great as a tool to like, tell you what to do and answer your questions through a chat interface, but to me, that’s not as interesting as having the agents take action.

Oren Michels 00:03:40  But in order to allow agents to do the things that we do, there are two things you have to have first. First, you have to have connectivity. The agent has to actually be able to connect to and take action on the system you’re using, whether it’s, you know, Amazon Commerce or, or Shopify or email or chat or whatever it is that you’re doing. And the second thing is trust. If you do not trust the agent to not do stupid stuff and to not make more of a mess than that, you’re going to have to clean up than it’s worth, you’re not going to use it. And so we started Barn Door to solve those two problems, to make sure that that that AI could connect to the things we use and that we as humans would trust them both from a traditional security standpoint of of bad actors trying to get access to things that they shouldn’t, but also perfectly acceptable usage of AI that has bad results because AI ultimately is not intelligent, it’s not smart. It is a probabilistic guess at every answer.

Oren Michels 00:04:37  That’s really how it works, right? That’s what these models do. And so you want to make sure that when these models are making their probabilistic guesses, that not only are they likely to do the right thing, but that they’re prevented from doing things that they shouldn’t be doing. And so these are called guardrails or policies or whatever you want to do. And barn door is a is a software package. It’s a it’s a service that we provide that allows you to very finely designate in what contexts and in what circumstances the agents you’re using are allowed to take actions on these various tools so you know, you’re able to work, let’s say, in Shopify. And you you know, it’s very broadly through us. You can create items, you can update items, you can look at orders, you can create orders, you can do all the various things that you would otherwise be doing in the interface. But you can do it programmatically and, you know, in plain language from cloud. And so if you’re going to if you’re going to do that, you probably want there to be rules so that it doesn’t take actions it’s not allowed to take.

Oren Michels 00:05:35  And so we provide that capability. And through us you have these two things that I believe are necessary for AI to succeed. And that’s connectivity and trust.

Josh Hadley 00:05:45  No, I love that. And I think that’s going to be even more important as we move into the world of people. Using AI agents to take action on their behalf is like, what’s the security risk that comes with that? And are there guardrails in place? So or maybe.

Oren Michels 00:05:58  You know, there are, you know, the traditional concepts of security are more around bad folks potentially penetrating and getting in and doing bad stuff that those risks obviously still exist in AI. Some argue they’re getting worse because people are putting AI’s onto these systems, but there’s also a risk from perfectly acceptable legal people doing things they’re allowed to do. But bad things, bad, bad outcomes happen, and that’s what you have to be able to guard against as well.

Josh Hadley 00:06:33  Oren, very interesting concept there with barn door AI. Very interesting. My question for you here is like, so what’s the difference of me using barn door versus just like utilizing open claw or just clawed co work or cloud code or something like that on my own.

Oren Michels 00:06:47  And you should use all those things. That’s that’s what I do every day. But the thing is that when you tell one of those tools to take an action, it does whatever it thinks the right action is, irrespective of whether you allow that. We call that governance. There is no governance built into any of those systems. And to the extent it is, I use open claw every day and every day I’m reminded of how it’s supposed to act because it forgets. Forgets it’s not. And but of course. But things that you have told it before are not necessarily retained. And so if you were to say don’t ever do X, which I have said on various occasions, it will still try to do X because it’s, you know, there’s a lot of context that comes before and it’s doing its best to come up with the right answer every time. And so and oftentimes it will hallucinate and do random things. So if you want to be able to say, I have an agent or I have a I’m using clouds to do this particular task, and I want to make sure in doing that task, it can take these actions, but only these actions and not other actions.

Oren Michels 00:07:42  You want to create that through policy. That’s the layer we provides. You know, open claw as an example, open claw is a very, very powerful piece of software. There are two ways that people tend to deploy it in one way. They put it on their regular machine or even its own machine that it has, and then they have that computer on it. It will have, let’s say, their email or access to their business applications or whatever. And it because it’s native and because it has root access, it figures out what it should do and acts in whatever way it deems appropriate using these tools in the same way that you and I would use them. But you and I have governance built in. We know certain things are stupid and we’re not going to do them. And, you know, if we do them too often, we’ll get fired or we’ll have bad outcomes. They don’t understand that. And so without a layer of governance, it is very difficult to trust them. The other way that people deploy open claw is what they call a headless deployment at a hosting facility, like a place like DigitalOcean, that has a really easy way to spin up online instances.

Oren Michels 00:08:40  And there it’s also hard because then open cloud doesn’t natively have access to anything because your your tools are not there, your email is not there or your Shopify is not there. All these tools are out there. So you’re able to, with Baader or with an AI or personal product, you’re able to connect that to all the various tools you need, and then you just tell open claw, hey, use the end skill, or use the barn door skill to connect to my tools, and it has everything available, and you’re able to govern what it is, what it can and can’t do.

Josh Hadley 00:09:09  So give me some use cases there with with the governance. Like what do you mean by, you know, making sure that it’s not hallucinating because I totally agree with you. You tell it one thing forgets the next day it starts elucidating, thinking of things that aren’t actually true or that you want it to be doing. So how do you guys solve that?

Oren Michels 00:09:24  So I’ll give you an example. You know, I am not a 8 or 9 figure Amazon seller.

Oren Michels 00:09:30  I will give you when I give you use cases, I might suggest a few that are interesting to the folks watching here, but I also don’t want to insult them by pretending I know how to do their job because I don’t. But we are as humans were good pattern matchers. So as I talk about how I use it, I encourage your listeners to say, oh, okay, well, that’s how he uses it in his job. That’s kind of like this problem with my job. So I can do something similar there. So, you know, with that preamble, I have an open clause that I use every day to help me go through all the news and newsletter emails I get. I get a lot of them because there’s a lot of things I want to keep track of, probably about 200 a day, and it traditionally takes me about an hour to go through all those each day. Some of them I read the entire version of it. Some of them I just want to see a summary.

Oren Michels 00:10:12  Some of them I’m sort of done with, but they don’t really respect unsubscribe. So they keep coming anyway, right? So I have open claw and open claw through. Our AI tool is allowed to do exactly two things. In this case it is allowed to read my email, and it is allowed to set flags in my email. That is the only thing it’s allowed to do. It’s not allowed to create one. It’s not allowed to send, it’s not allowed to delete one. So I’m not having to worry that if something happens, it’s going to I’m going to miss an important email or I’m going to have something sent to someone. I don’t want something to be sent to. And so every day Open Claw goes through it looks to all of my emails. I use a tool called superhuman. The categorizes news emails, so it’s further limited. It’s only allowed to look at news emails I don’t I only let it look at the one that superhuman categorizes as news. And so it goes through all of those.

Oren Michels 00:10:54  And it knows because I’ve trained it, it knows that there are certain senders that automatically get marked priority. For me. They have the priority label attached. There are certain senders automatically get an archive, this label attached. And then there’s the rest of them that I want to see a summary of each one in a row, so I don’t have to open the email. I can scan down and I can say, oh, that I want to read that one, that one and that one, archive the rest. And doing that for me every morning and then reading the ones I actually want to read has turned an hour into 5 to 10 minutes, and I got the rest of that time back in my day. And not only did I get the rest time back in my day, because it often takes so long, these things will sometimes stack up over days, and I will either not bother to read them because I am behind at this point, or it won’t take an hour. It’ll take two hours to go through, or it’ll end up being what I do over the weekend when I’d rather be having fun.

Oren Michels 00:11:44  And so, you know, it really has allowed me to take this very tedious part of my job, but very important part of my job, which is being aware of what’s happening out there and, and, and turn it into a pretty quick task.

Josh Hadley 00:11:57  I love that. So what’s the difference? Why why do you need barn door versus like letting the open claw just do its thing there and try to give it?

Oren Michels 00:12:04  Well, because if I, if open claw merely was was given access to my mail without the governance, then if I were tasking it with doing something else than that it would might think, think using that term, but it might take an action or infer that the correct action is to send an email to someone or to, you know, if I say to organise this or archive it, I want there to be a step where I scan down all the ones that’s marked for archive to make sure there’s not something that I want to see, but if it’s already deleted it, I can’t do that right.

Oren Michels 00:12:36  So I want to make sure that I let it take the action that will get me most of the productivity back without the risk of something catastrophic happening where a customer email or a, you know, an investor email ends up getting archived or deleted or responded to in a way that’s inappropriate.

Josh Hadley 00:12:55  That makes sense. And so ultimately, I guess in Barn Door, you’re just able to set up your custom rules then and then.

Oren Michels 00:13:00  It absolutely.

Josh Hadley 00:13:01  Integrates back and forth.

Oren Michels 00:13:03  Yeah. And and with Barn Door, you can be very, very specific. You can say, you know, in the e-commerce world we have a native integration to Shopify. We’re working on one for Amazon. But you would be able to potentially allow it to take certain actions, let’s say on you could dictate you can take actions on items that have a selling price of less than $25. You can do these things, but if it’s more than that, I want you to suggest and then I will. I will approve it each time. Or, you know, there are ways you can create policies around the context that can be fairly, fairly granular and really allow you to take action with, with AI in a way that that gets things done for you.

Josh Hadley 00:13:40  Yeah. Fantastic. Well. Oren, I would love to dive deeper into this rabbit hole. So give me some more use cases. What are some ways that you’re still utilizing it? Maybe even some of your clients are utilizing Murnau, because, yeah, I love the ideas of like, hearing how other people are utilizing AI in their own business and then applying that to myself.

Oren Michels 00:13:58  Sure. So my head of product here, co-founder, we use a tool called JIRA, which is very common in software development. It’s a tool for creating tickets and sort of organizing what you’re building and organizing all the issues that you have in developing software. And my head of product is a very powerful tool. He no longer logs into it. He does everything through Claude, through the integration, and he creates tickets. He looks at tickets, he closes tickets. He assigns them all these things he does. That’s his job. He can do them en masse, and he can look at at things that and delegate tasks to people based on, on the past.

Oren Michels 00:14:32  So he can say, you know, based on the tickets that our engineer Miles was working on before, this ticket is likely to be won for him because he’s able to analyze that and say, okay, I’ve got a bunch of tickets. Look at the past tickets and who should I delegate these things to? Right. So if you look at it and we’ve got dozens and dozens of different tools, my head of finance. You know, as a startup, one of the most important things by the most important thing is to never run out of money. So my head of finance gives me a weekly cash forecast and a weekly cash availability report, and it shows up in my slack. She doesn’t do it. She has an AI which basically goes out to QuickBooks, gets this information, goes out to we use Ram for paying bills, goes out there, picks up the things that are scheduled to be paid that week, creates it into a little table, and pops it in my slack for me every, every week.

Oren Michels 00:15:17  And so I know exactly what we’re doing with money. And she doesn’t have to think about doing that right. And so it’s that kind of of tool that allows you to actually go to these various systems, again, with rules that keep it from doing things in those systems that it’s not supposed to do, and actually create a slack and send it to me, her agent through barn door is only allowed to post to internal channels. We also have external channels in slack, where we have customers and other outsiders who we collaborate with, partners and such. Obviously I don’t want my cash information going to those people. I can trust that and hope that the agent will only put it in my slack channel. But if there’s a rule and there is in her case that her AI is not allowed to post to any external channel, I don’t have to worry that somehow my cash information is going to go somewhere else.

Josh Hadley 00:16:08  That’s an important one, obviously. I mean, what about logging? Are you having it scrape like bank account details and things like that?

Oren Michels 00:16:14  Absolutely.

Oren Michels 00:16:14  It’s allowed to see those. It’s allowed to see those through. Because, you know QuickBooks through plaid is integrated. We have a plaid integration as well. So we can see our bank account information. And it uses that information to be able to get the balances. So we know if we have to transfer cash or you know, we like yesterday I got a renewal on one of our CDs is coming up. And so I was able to say, well, how much of it do I want to roll into new and how much do I want to, you know, pull in for working capital?

Josh Hadley 00:16:36  Yeah. No, I love that use case. Especially just using like the read only access that comes into books or zero account. Great use case. What are some other fun ways that you’ve seen maybe some of your clients utilizing barn door?

Oren Michels 00:16:48  You know, it’s interesting. We do a lot of work with some fairly big enterprises that have internal systems that they want that they’ve created, whether they’re, you know, in one case, it’s a company that deals with a lot of PII.

Oren Michels 00:17:01  Another case, it’s a major hotel chain that has guest information, and they want to be able to use AI. Or in the case of a hotel chain, they’re using it for customization. They’re using it to look at historical things that that frequent guests have, let’s say, requested or have had issues with and proactively provide better service to them by being able to look at these, at these, at this history. Right. And so AI is incredibly good at looking at a lot of data and drawing inferences from it. So one of the things in e-commerce that I’m in addition to my work at Barn Door, I’m a board member at a company called autopilot that provides automated updating of of Amazon listings. So you basically you want to update the product description frequently because that helps you you raise in, in in the search results. And one of the things we found is that there aren’t just a handful of holidays every year. There’s seasonality that’s very subtle throughout the year that causes different buying behaviors. And so being able to, you know, know human is going to update these things as frequently and as tied directly to the tie, directly to what’s going to work in getting you better search results.

Oren Michels 00:18:14  As an AI. As you know, the humans can’t keep up, especially if you have a lot of assets to sell. And so being able to understand historical data and use that to ensure that you have the correct and most up to date and most powerful and, and, you know, relevant Descriptions. AI is incredibly good at that.

Josh Hadley 00:18:35  Yeah, I think that one’s super powerful, especially with like all the different holidays or you have like especially like you want to change your listing on Amazon to incorporate more seasonal keywords for Q4, right? Christmas gifts and gift for dad, gift for mom, etc.. But during the summer, you may want to be focused on, you know, maybe like travel gifts or things like that or other people. So I really like that use case, especially the e-commerce space.

Oren Michels 00:19:03  Absolutely.

Josh Hadley 00:19:04  What other things have you seen? I think you also mentioned customer service and a few others as well.

Oren Michels 00:19:08  Yeah. So I do a lot of emailing to prospects and customers and investors and others.

Oren Michels 00:19:14  And there’s a lot of those people in my life and I cannot remember all the things I’ve said and which person I’ve said one thing to each time I’m going to respond to them. So if I have a customer email that’s going to go out, the first thing I do is go to cloud and I say, I’m about to email customer X. I want everything from Salesforce, all my previous emails, and I want you to give me sort of the history we have with them. I want to see what the last email exchange was with them, and I’m trying to communicate X to them in the context of how I’ve done that. Please draft it again, and I don’t want it to send the email again. I want to make sure it’s it’s a customer. It’s really important, right. But by getting it that far, I’m just doing a little edit and sending it out. It not only saves me a bunch of time, but it also gives me the ability to have a better email because it’s going to be more contextual and it’ll be more to the point and consistent with whatever the background has been with that customer or investor that I’m dealing with.

Josh Hadley 00:20:04  Yeah, understanding that context and having like a deep memory for each contact, I think that becomes very, very important.

Oren Michels 00:20:11  It does, it does. And you know, it really allows you to it allows you to appear to really be personalizing things in a way that historically happened. You know, computers get a bad rap of sending these sort of, you know, form things out to everybody. That just just sounds, you know, like it didn’t come from a human in this way. You could actually have AI make it humanise you a bit more.

Josh Hadley 00:20:33  Yeah, I think it makes a lot of sense. Are there any other good use cases or ideas that you think our audience needs to hear?

Oren Michels 00:20:40  Well, I think the most important thing is actually that, you know, we’re hearing a lot of folks talking about agent AI and what they, you know, you should be definitely doing doing this and everybody’s doing it and all these sorts of things. And the truth is that it’s actually in its infancy.

Oren Michels 00:20:55  Most of the AI agent work that we’ve seen to date. Most of the use has been by people writing code. That was sort of the original. You know, turn it loose and have code do things. And it’s incredibly powerful that way. It’s relatively recent for other knowledge workers to be using it in the way that we’re now starting to see. And I think the key thing to figuring out what to do with it is just to get started and to find some ways to use it in the same way that my, my co-founder basically no longer uses the JIRA interface. At first it was weird for him because he’s been using that his entire career, but now he wouldn’t go back. and if you may actually start. If you’re not ready to jump in and turn it loose on your what pays the bills, perhaps you’ll start using it in your personal life. You’ll start using it with personal emails and calendar and helping you organize things or anything that you’re doing organizing family stuff. I was talking to a friend of mine who had the system that they had been using for their pickup soccer league.

Oren Michels 00:21:52  For whatever reason, they lost access to, and over the weekend, he just went to one of the AI tools and coded up a custom one specific the way he wanted. This is a sales guy. He’s not a coder, right? He’s a sales guy. And he had it up and running, and his pickup soccer folks were signing back into it 12 hours later. And and so, you know, these things that just seemed outlandish not so long ago have become a thing. And really, the best way to figure it out is to start using it in in situations that are not very, you know, high dollar or high risk. And then once you start seeing it, you understand what it really is and how it really works. You’ll then start bringing it into your work world, but the only way to understand it is to do it.

Josh Hadley 00:22:40  Yeah, I love that use case there that you shared of just like personal application. And in fact, that’s exactly what I’ve been doing as I begin like experimenting with AI, how I want to use it, and the applications that I want to use for our own brand.

Josh Hadley 00:22:53  But the personal use case that I just did over this last weekend was I created a couple custom GPT s that were very specific for creating a batting lineup and creating all the fielding positions for my teams, for both my softball team that I coach as well as my baseball team. And so it’s like it’s going to feed in what’s the batting data give me, like the statistics, and then it’s going to set up and configure the appropriate batting order to ideally give us a higher probability of winning and make sure that we have the right people in the right positions during the right innings. And so what used to take me again? Probably like an hour. You know, on a Friday night before our Saturday games of going through setting up the lineup. And it was definitely faulty. Like, now this is done in literally just a couple minutes.

Oren Michels 00:23:37  So well. And it’s doing no offense to you as coach. It’s doing a better job than you would do because.

Josh Hadley 00:23:42  100%.

Oren Michels 00:23:43  You know, the people involved. So you might say, oh yeah, that person’s really good or that person, whatever.

Oren Michels 00:23:47  This thing has the numbers. And so it’s able to make the right decisions as opposed to the emotional ones.

Josh Hadley 00:23:54  Yep, exactly. It rules out there’s no like, human, I guess, subconscious thinking here that goes behind it. Or personal relationships. It kind of gets removed and you’re just staring the data in the face, which is great.

Oren Michels 00:24:06  And when someone comes to you and said, why am I batting ninth, you can say, well, the numbers don’t lie.

Josh Hadley 00:24:11  ChatGPT told me.

Oren Michels 00:24:13  Exactly.

Josh Hadley 00:24:14  So on that note, I am curious to get your take on this, Orin. So there’s ChatGPT, which is I, I feel like recently fallen out of grace with how quick Claude has moved and things like that. Chap was the the originator and what most people were using, and now it’s almost like it. I almost didn’t even want to say I created a custom GPT. I feel like an old grandpa or dinosaur now at this point, because it is all moving to Claude. So give me like the lay of the land where you see things going in the AI space.

Josh Hadley 00:24:41  Like what tools should people be using and experimenting with right now?

Oren Michels 00:24:45  I think the easy answer is all of them, because just as with humans, we don’t only hire one kind of human in our companies, we hire the best human for this job, a different human for that job. And different tools are good at different things and different models, even within the tools are good at different things and they have different costs. And, you know, there are all kinds of ways to optimize it. But I think the most important thing to remember is that, like open cloud, none of us had ever heard of Open Claw a month ago. Two months ago. Right. One dude spent a few months writing this thing, and it’s changing the world. If that can happen, that’s going to keep happening, and you’re going to have new tools and new APIs. And exactly. Some are going to come into favor and out of favor. I use a wide range of models, and I use a wide range of platforms because I’m always trying to figure out which is the right one for that.

Oren Michels 00:25:31  And, you know, I, I was helping a friend who’s an author set up some stuff to, to track their books the other day, and they were saying, I want to use Open Claw for this. I’m like, well, actually that’s not the right AI for this. That’s that’s not the right person for the job. You can actually accomplish this better in cloud for these reasons. And so it does keep changing. I think that that, you know, OpenAI is a company with a bunch of really smart people, maybe took their eye off the ball briefly, but I wouldn’t count them out. But I also recognize that, you know, you look at something like Cloud Coworker. Cloud coworker is incredibly powerful. But if you look at the history of of software and infrastructure and these sorts of things, it’s really unusual for the company building the infrastructure to also build the best apps that are going to live on that infrastructure. This goes back, you know, you look at the phone companies who then became the internet providers, and they all wanted to stop being just, you know, dumb pipes to carry the carry bits around.

Oren Michels 00:26:26  They wanted to, you know, become the e-commerce company, become the the connectivity company, become the, you know, these sorts of things. And it never worked. Right. Because it turns out there were other people. They were really good at their infrastructure thing, and there were other people who were really, really good at figuring out what to do with that infrastructure. And I think we’re going to continue seeing that. We, you know, already there’s a fair amount of stuff being written about how you really get the most for your AI. You know, bang for your buck when you find a purpose built tool that uses AI smartly to do something, as opposed to expecting the generalist jack of all trades to be the master of whatever you’re doing. And you know there are there are AI tools that are really, really good at reviewing contracts. The large, generally large language models aren’t such tools, right? They’re not actually very good at reviewing and read loading contracts. And I think you’re going to keep seeing that.

Oren Michels 00:27:16  And so it’s it does become complicated and it does become messy. But that’s kind of how it’s been for a while. And this whole thing is at its infancy. Right. This is this is very early days. And, you know, there were there were other spreadsheets before Excel and Google Docs before I used Visicalc. I used Lotus, they don’t exist anymore. Right. So, you know, a lot of these things are there. They’re going to change over time. And really, really smart, insightful people are going to come out with products that that will really, really help us run our businesses.

Josh Hadley 00:27:47  So or if that’s the case, do you like what is the world look like? And again, anybody like who knows what AI is going to do. But like do you see this to where hey, if you want if you want to leverage AI for customer service, go find the software that does that best that it’s utilizing AI in the background. If you want to do something like have agents acting on your behalf on Shopify, go find the software that does that that’s baked in AI into it.

Josh Hadley 00:28:15  Amazon and like and so ultimately like your tech stack just becomes like hundreds wide. It’s like you’ve got one for HR and you’ve got one for customer service, etc. or are you thinking that things actually stay within more of like the cloud and everybody just gets like very specific skills inside inside of cloud? Like, which direction do you think? Like I’m going to end up being like most impactful?

Oren Michels 00:28:37  Yeah. You know, I think when you talk about a cloud skill or an open cloud skill, most of those skills are acting on software, right? They’re acting on systems of record. They’re acting on Amazon or Shopify or Mail or Slack or whatever. Right. I don’t think systems of record and software is going to go away. But there’s a there’s an executive I follow who basically said, what’s going to happen is not that these systems go away, it’s that agents are going to act on them 100 times more frequently than we as humans do, right? So when you think about the optimization for ads, as an example, I was an early buyer of Google AdWords when it first came out, and I would list the keywords I wanted.

Oren Michels 00:29:19  I would list how much I was going to pay for them. I’d put that in a spreadsheet and I’d upload it, and when I wanted to change my mind, I’d get new keywords and new numbers, and I’d put them in a new spreadsheet and upload it. Nobody does that anymore, right? That is all optimized to the second to the microsecond by these AI models to to most efficiently, you know, create this, this keyword market. And I think we’re going to be seeing a lot of that coming. You’re going to interact with these agents through something like cloud, and you’re going to govern these agents because there’s going to be bunches and bunches of them. They’re going to have hundreds or thousands of these things that are optimized to do different tasks. And you’re going to be governing those things through something like Barn Door, I assume will be barn door. Ultimately, you know, when you think about it, we’ve had what we in the software industry called low code or no code things where non coders can stitch together applications.

Oren Michels 00:30:10  That’s been around for a long time. There’s been you know, we have Zapier right now. There was techno there was cognitive. These concepts have been around for a long time, and yet no one’s stopping hiring engineers. It actually, you know, we still require people to think through it, even if it’s not. And write the code. There’s a there’s a skill around how you take a business problem and dissect it and figure out how to efficiently and effectively tell whether it’s a computer or a human to go do that job. That skill is not necessarily something that everybody has. And so folks who can see a problem and see the tools and go, wow, with these tools, I can solve that problem this new way, those. You know, that skill isn’t going anywhere. And we as business people will want to get the very best of that as it evolves to work it out.

Josh Hadley 00:31:00  Yeah. So ultimately, like you’re saying like just get really good at solving problems and figuring out the process that that backs into.

Oren Michels 00:31:06  And so the tools are going to change. The tools are going to change. I went to my undergrad at MIT and computer science. I haven’t coded in decades. When I was learning that the way we were taught, and I had taken some computer science classes at a different college when I was in high school, and in that college it was taught as here is the problem, here’s the algorithm to solve it. Today we’re learning language X, right. This algorithm to solve this problem in that language that we’re teaching us a language. Whereas at MIT it was we were using a language that nobody actually programs with on a regular basis. The whole point was, here’s the problem. How would you solve it? And how do you tell a computer to go solve it? And, and I think that that, that the more that we get good at that. And frankly, as a manager, it’s the same thing. If you have an employee who’s working for you and you want to mentor them, you know, you sort of have to be really good at creating instructions.

Oren Michels 00:31:59  And we call that in the AI world. We call it prompt engineering. Right. The difference between telling whether it’s chat or cloud to go do something, you know, if you write the right prompt, you’re going to be much more likely to get the thing you want than if you don’t.

Josh Hadley 00:32:14  Yeah. Well, I think obviously like the best communicators and those that are able to articulate ideas clearly and step by step, it’s almost like really good teachers. Like you’ve got to first be like a good teacher, a good mentor. And it’s almost just like putting it on steroids of like, hey, if you were able to onboard really good team members to your team to begin with and help them understand your logic and your thought process in the way that you would like, solve X, Y, or Z, then guess what? Now you get to like, magnify that by 100 times.

Oren Michels 00:32:40  Yeah, exactly.

Josh Hadley 00:32:41  Oren, there’s been a fantastic conversation. Is there anything else that you feel like the audience needs to hear that you haven’t yet been able to articulate?

Oren Michels 00:32:47  No, I just I would just reiterate, jump in, start with what you’re comfortable with.

Oren Michels 00:32:52  And I guess the other thing is that you probably have friends who are talking about how they’re using AI and how they’re using agents, and they probably don’t do the thing you do. They’re not they’re not Amazon sellers, right? They’re not they’re not e-commerce rockstars, but understand how they’re using it in their work and what they do. We’re very good at pattern matching. If you listen to how others are solving the problems they have. Think about those problems. It’s not as though, well, that’s different than what I do, but more what are the things I do that kind of look like that, that I can maybe use those ideas to solve, rather than necessarily saying, I want to find out how people are doing. A genetic AI in e-commerce. Because the reality is, if you’re listening to this program, you’re probably going to wind up being one of the first to be doing that anyway. And, you know, you open the program by talking about it. These agents are going to be the ones you control, are the ones you compete with.

Oren Michels 00:33:41  I would encourage you to be have them be the ones you control, because you will be competing with them before you know it.

Josh Hadley 00:33:46  Yeah, really well said there Oren, as we wrap up this episode, I love to leave the audience with three actionable takeaways. So action item number one is start playing with exactly what Oren said. Just start utilizing it in your own personal life, whether it’s for cooking purposes or whether it’s in my use case was for baseball and setting up a lineup every time. But like, figure out things that you do on a repetitive basis, and then go ahead and create a custom GPT, or create a cloud skill for yourself and just begin seeing how they work. Because start there because like, you don’t want to start with your financials and be like, oh, I’m, I’m going to set up a financial or accountant for me, and I’m giving them access to my bank account and things like that. That’s where the stakes are pretty high. And so start where the stakes are low.

Josh Hadley 00:34:32  Figure out how it works, how you give it feedback, how you course correct when it does veer off. You know, because like even the custom GPT I created, it definitely wasn’t perfect. Perfect. And I still had to refine it multiple times. Action item number two is create really good SOPs and become really good at defining the problem. And then the steps that need to be taken in order to accomplish that. So as things continue to evolve and you said this well, is that we’re still in its infancy. Yes. There is a lot of noise in the AI space, and almost everybody feels like they are behind. But the most important thing that you can do right now is get really good at documenting. Like, hey, this is the methodology that we use in our business to run Amazon PPC campaigns or to run meta ads, because everybody’s going to have a little different flavor as to how they do it. And the the opportunity is not, hey, I’m just blindly going to let Claude or open Claude, just start managing my PPC.

Josh Hadley 00:35:27  You’re going to need to give it the context, the rules, the guidelines, and so begin setting those up as though you’re going to have a perfect employee that follows your instructions to the tee, because that’s exactly what AI will be able to do. And it will run 24 over seven for you. And you don’t have to deal with any personnel problems. Action item number three here is to, you know, the tool itself that you focus on doesn’t matter so much. I think the most important thing is to stay in the game. Stay in the game of experimenting with Claude. Maybe it’s open, or maybe it is just staying with ChatGPT. There’s still a lot of different things that these different tools and software providers provide to us, and so the tools are going to change over the next six months, the next year, and definitely over the next 3 to 5 years. So the most important thing is just be using it. And as soon as things change, you’ll be at the forefront or in anything else you would add to that?

Oren Michels 00:36:18  No, I think that’s exactly correct.

Oren Michels 00:36:19  And, you know, I was corresponding with one of our users the other day who said they weren’t sure if they should commit to Claude over ChatGPT. I’m like, you’re not committing. You’re just using it today. And if the other one’s better tomorrow, then you can use it. And you know you can. You shouldn’t be worried about that sort of lock in concept.

Josh Hadley 00:36:36  Great. A great use case there. All right, Oren, as we wrap things up, we’ve got three final questions for you. Number one, what’s been the most influential book that you’ve read and why?

Oren Michels 00:36:44  You know, I’ve read a fair amount of books. I would say there’s a recent book that was that was written called Why Computers Think. And it basically breaks down the math, which I no longer understand from my undergraduate days of how these things work while there is math in it. What has been really helpful with this book is not trying to understand vector math, which I can no longer do, but to really understand that these are probabilistic systems, and it sort of breaks it down as to why it is that they hallucinate and why, you know, why it works, but why it sometimes doesn’t work.

Oren Michels 00:37:17  And I found that just to be it gave me a good grounding in understanding what these things are.

Josh Hadley 00:37:22  Excellent recommendation. I haven’t heard that book before, so now that’s a good, fresh one to add to the list. Oren, we’ve talked a lot about this today, but what is your favorite AI tool that you’ve been using and what’s your favorite way that you’ve been using it?

Oren Michels 00:37:34  Well, we’ve talked about, you know, I’m I’m enjoying Open Claw a lot right now because I’m using it. I think I talked about for the, you know, to get me time back, which I really appreciate, and to make me actually more aware of the things happening in my world. So that’s that’s pretty lovely. But I’m also, you know, it. It’s also just a wonderful tool to kind of go down some rabbit holes when I see something out in the world that I want to know more about. It explains to me the answer to that question. I, I noticed something at my house the other day.

Oren Michels 00:38:05  I noticed some trees that. That are starting to leaf out, and some other trees that perhaps. Look like they’re not as healthy. So I went to, I think it was Claude in this case, and I had a whole conversation about this particular species of tree. And, and, you know, given the location where we live and this sort of thing, what is likely happening with it. And I learned an awful lot. I in the end, it told me I didn’t have to worry about it, which was great. But but I learned a lot in a lot of the context and sort of how we got to where we are. And I really, I sort of appreciate being able to dive as deep as I can and learn more.

Josh Hadley 00:38:34  That’s a really another really good use case for personal application of just like incorporating it into your daily life.

Oren Michels 00:38:40  Absolutely.

Josh Hadley 00:38:41  Great comment. Third and final question here. Who is somebody that you admire or respect the most in the e-commerce and business space that other people should be following and why?

Oren Michels 00:38:49  I would say Aaron Levy, the CEO of Box Aaron on X pretty much every day is posting really interesting stuff about AI.

Oren Michels 00:38:59  He’s he is very much he’s the person I alluded to before who talked about how it’s going to be using the same software apps 100 times as much as humans. It’s a very interesting take because while he’s very, very bullish on it, he also recognizes that it’s in its infancy. That’s changing. And I I’m in this every day and I learn something from each of his posts.

Josh Hadley 00:39:17  That’s a great recommendation for people to follow on. And this has been a great conversation. If people want to learn more about you, they want to follow you. Learn more about Barn Door AI. Where’s the best place to do so?

Oren Michels 00:39:28  The best place is barn door AI and if you want to get started on your own, we have a product that’s at Zenni. That’s the personal version. Super easy to use. Just get a free trial, go sign up and play with it and and you will be able to do stuff on your own that I think will will really change your life.

Josh Hadley 00:39:45  Wonderful. Well, Oren, thanks again for your time today.

Oren Michels 00:39:48  Thank you. Josh.

MC 00:39:49  Thank you for listening. Visit Ecomm Breakthrough Comm for more information. If you’ve enjoyed today’s episode, the best way you can show your appreciation is by clicking the subscribe button and quickly leaving a review. See you again next time!