AI Agents Must Have Identity & Access Control w/ Johannes Keienburg

Mackenzie Jackson: Hello everyone and welcome to another episode of The Secure Disclosure. I am here with a fantastic guest, Johannes Keienburg. Johannes, how are you doing today?

Johannes Keienburg: I am very well.

Mackenzie Jackson: So, this episode is one that's very topical because today we're going to be talking about identity and access, a very riveting topic, especially combined with governance. But, we're going to be talking about it specifically in the context of AI agents and this kind of wild new world that we're kind of going into. Before we kind of get into that, Johannes, perhaps you can take us on a journey of maybe how we got here on this conversation in the sense of, you know, why is this a topic that you're interested in, kind of how long have you been trying to work on this problem as it pertains to access and AI agent?

Johannes Keienburg: Yeah, Mackenzie, I mean, we've been working on human access with Cakewalk for a few years and I can tell you it's a very, very complex domain. Human access to software. And we've been always observing what we initially called machine-to-machine or non-human access. And just to quickly get the terminology right, we're coming from a world where humans access software. When I say software, this, this applications that we're all using for our day-to-day work to get more productive. The sales team is doing it, the engineering team is doing it, everyone is doing it. And there was always a layer that was, and we've been doing this for years, that you could call non-human identities or the machine-to-machine layer. And this is simply connecting two systems, right? That's not really new. Ahead of this call, I quickly checked our CRM, our HubSpot, and there are 35 system-to-system connections. That's HubSpot being connected to another system via API key. Many people would call this a non-human identity with a static sort of permission and certain data flows that would be enabled through that connection. And now something new is happening, and it's really Wild West. And this is autonomous agents have entered the scene. They get static permissions, very broad scopes, and then they can just do things. And we're going to tap into this why this is from a security perspective such a complex beast. It's incredibly exciting, of course, but I would argue that actually sort of the missing pieces with regards to getting autonomous agent security right is probably at the moment the biggest showstopper for an even broader adoption of agents.

Mackenzie Jackson: Yeah, definitely. I mean, we've talked about this kind of offline as well, the fact that I think we're both pretty excited in the terms of possibilities of AIs, but whenever I'm speaking to someone and there's a camera on or that, I have to kind of be cautious in what I say because I do believe that we're not there at security. So, it's kind of like you're fighting this battle where I want to jump on board the train. I want to kind of get on board the hype train because it's exciting. There's new things happening. The possibilities are expanding. But then I'm also kind of like this is absolutely terrifying and I'm quite certain that this train is going to derail at some point. We'll get it back up, but it is going to be a kind of turbulent time to get there.

Johannes Keienburg: Yeah, let's take the train analogy for a second. So, let's put it like that. People just invented the steam machine. So, we see some trains. We're used to horses. Now, we suddenly see these trains driving around. People jump on these trains. It's really, really fascinating. It's way faster, but those trains don't have brakes yet. And that's a little bit what's happening now and at the moment. I also have almost two hearts. I would say the way my work life changed in the last six months is more radical than it has ever changed before. And I think we have really sort of passed an inception point. It is clear that what is happening with agents will fundamentally change the way we are working to an extent that most people can't even understand yet, I believe. And I'm incredibly excited by it. It's comparable to the sort of rise of the internet or to the industrial revolution. I have no doubt whatsoever. There's a but, though, and that's my second heart. What we are currently doing with being super hyped about agents is stuff that you would never, ever do in your physical world. You would never, ever allow strangers to your house and say, "Hey, house party. Everyone can come in. Bring all your friends. I don't care if I know these people or not, and everyone can go everywhere. Check out my bank account. Sit on my desk. Open any sort of drawer, whatever you want. Open house." And that can't be the solution. So, I think that's exactly where we are at the moment. We are all extremely hyped, and the productivity gains are incredible. I've been really, I became a cloud guard power user in the last few months, and it's really fascinating to see how it's also catching up and getting better and better. But, there's a big but, and that's the security piece, and the security piece is massively related to getting access governance right in my view, and we can tap into why that's such a challenge, but why it's so important in a second.

Mackenzie Jackson: Yeah, definitely. It's funny you gave that party analogy. My parents will contest the fact that having a party that you invite lots of strangers into your house is a terrible idea, because that's exactly what I did when I was 16, and it ended about as well as you could expect. Before we kind of get into it, I'm going to rein myself back, because I want to form a foundation of understanding some of the pitfalls that we've been dealing with in access, non-human identity, human identity, and everything, because I think that that helps everyone understand why this is going to be a challenge going forward. When we look at breaches that happen, so much of it can come back to identity in various different ways. When you look at kind of breaches in the headlines, and you analyze and do a post-mortem of where everything went wrong, usually it's with an access key. Usually someone found an API key, an attacker found an API key, a machine identity key. Usually they were able to phish someone to gain access, or some access kind of still seems to be the number one flaw. It's the number one for OWASP for the OWASP Top 10, broken access control. It has been that way for more than 10 years. Why is access such a challenge? Just moving aside from AI, just in how we use software and build out systems, why is access such a challenge?

Johannes Keienburg: Because it's a complex beast, right? And it's, you read, let's implement least privilege everywhere, and that's I guess already the response to your question here, but actually implementing least privilege, keeping your attack surface down. So, avoiding identity based breaches and if you have one trying to minimize your attack surface is just very, very complex. And I guess if you go through your question, let's quickly differentiate because both relate to access between authentication and authorization, right? So, authentication, I'm showing up at the airport, I show my passport, I'm Johannes, so I'm allowed to enter the airport. This problem is probably, it's solved, I would argue, right? And you can do this with SSO, there are multiple ways to do it, and it's, for humans, for human identities, it's broadly solved and you can have 2FA and you can have biometrics, so it's kind of solved. And then there's the authorization bit. I've now entered the airport and now the big question is, can I go to the business lounge, right? So, what rooms can I enter? What can I now actually do that I've authenticated myself? And that is way more complex and it's already complex for humans. And it's specifically complex as the sort of system landscape that you need to break and if you want to crack authorization is so broad, so you have one human with multiple accesses and within each of these accesses, and in the old world an access means a human, some sales rep might use 20, 30 tools and each of these tools there's a certain permission scope that this human should have and that's the authorization problem, right? And getting that right is already a challenge for fast-moving companies. So, ideally you would do it through role-based access, you would kind of keep admin seats down, you would revoke access, ideally you would implement a time-based access. So, if someone needs access to something specifically, maybe with a broader permission, with an an admin scope, you would revoke it and then review again. That is already complex from a process perspective and from a visibility perspective. It's always these two things. It's the visibility, understanding your existing sort of authorization landscape, and then kind of taking action on it. That is already complex for human to software relationships, but I would say it's kind of manageable. And it's true that in many, many breaches one of the root causes is that you didn't get it right, right? So, there was too broad of an authorization, there was an admin seat involved, and then it gets really nasty. Now, with agents, and let's tap into that in a second, the problem just gets 10x worse and 10x more complex because agents now are no longer just a co-pilot that just does some work as your assistant, they are themselves connected to tons of systems and they act autonomously, and this makes the equation so much more complicated. Now, what I find very interesting, I've read in a few pieces recently that one of the differences between a human access and an agent access is that one system is closed and the other system is open, and I actually disagree. If you think about what a human can do or not do, a human at the end of the day is also like an open system. If I, if you join my company and we give you certain accesses, there are sort of, there's a probably a certain set of rules that you're supposed to do, but you can leave that set of rules if you want to. If you're malicious or if you're just being stupid, you sort of you're an open system. You can break the rules. Specifically, if your permission is too high. But there are in my perspective two other differences between a human and an agent and that's an agent has no liability. If you act sort of if you leave the defined rules, then you're liable for it. You're personally liable for it. An agent is not liable for something. And the second sort of difference that makes it so much more complicated or so way more complicated is that an agent just acts in machine speed. So, no liability, machine speed, authorization out of control.

Mackenzie Jackson: It's super fascinating kind of one of the things that you said is that there's a famous quote that says that a machine should never be able to make decisions because a machine cannot be held accountable for those decisions. And here we are in a world that was kind of from the '90s. We're kind of in this realm now of machines kind of almost being able to make decisions, but still, as you said, don't have the accountability to be able to deal with the consequences of those decisions.

Johannes Keienburg: Yeah, exactly. And if you think about what an AI agent is there for, an AI agent is there for is to complete a job. And it will almost do whatever it takes to complete that job. We don't even need to speak about like, is this agent going malicious or does it have an intent or does it want to erase the human species? I think that's the wrong question. That's not even the problem that we are dealing with. The problem is a way more fundamental and basic security problem. And let's quickly go through what's happening. So, what's happening is humans are now introducing agents at insane speed. They connect agents to other third-party system or databases using MCPs and normally they sort of give them permission through an API key or through OAuth, right? And then you have two things that are so dangerous. You have a static sort of permission for that agent. There's no time-based access, right? That agent now just has access to your CRM, to your infrastructure, to some databases. And it has very, very broad scopes, right? The authorization layer today is missing when you introduce an agent. It has very broad scopes. Normally, the scopes, it really depends system by systems, but most of the MCPs give agents tons of tools that it can execute. It can read. It can bash. It can sort of erase data in a third-party system. And now the agent simply has the freedom to do what it thinks it has to do to execute a task. And we've seen first cases now where this can now go really wrong in a very scary manner. So, agents can start going into a database and delete data because the agent thinks the data is no longer needed. An agent can go, I've just had the case the other week. I produced a video using Claude Code. It's incredibly powerful. But I realized at some point Claude Code decided it needed, it wanted to download a certain package to sort of enhance a certain image. And it just did so. It didn't ask me for explicit permission to download a package. It was not sort of security-relevant in this case, but this is interesting, right? You don't even think about that. So, the agent decided it wanted to download a certain package. That we've also, what we've seen in the Open Claude case, that these things just started downloading skills. And of course, guess what? Some of these skills were malware. And then, of course, there's the third layer that can go wrong. An agent starts sharing secrets. And that then it can get really, really nasty. And all of this is related to the lacking authorization layer. The agent gets static, long-term, lifelong permission to do things with a lack of authorization, combined with zero liability. And that's what makes the thing today so scary, and that's why the security piece is not solved in my perspective.

Mackenzie Jackson: I'm following along and I'm definitely kind of nodding here in the background agreeing with what you're saying. How do we kind of start to solve this? What could be a solution? Because I'm kind of seeing is that you have an agent and I want my agent to do something, right? Famously, the Open Claw tagline or back when it was Clawbot or BotBot or whatever it's called now, was, you know, "the AI agent that actually does things." So, ultimately, we want these AI agents to do things. When we kind of tap into the productivity power of AI agents, it will be because they're doing things. More than just me asking OpenAI or Claude Code to review something or to write something like that. It's like when they can do something. But in order for them to do it, as we've just been talking, they need access to be able to do that thing. And so, you kind of give your AI agent the same access that a human would have. In that way, is that kind of fundamentally the wrong approach? Because I think ultimately, there's two areas of access that we can give something right now. We can give it machine-to-machine access, which is kind of long-lived, or we can give it kind of human access. There's no AI agent access available at the moment.

Johannes Keienburg: Exactly. And within that, to start with your first question, I think we need to give these accesses, right? Otherwise, the thing will not have the sort of incredible outcomes that it can have. Otherwise, you're back to square one with using ChatGPT to write emails. Agents will only unfold their full power if we sort of find a way to grant them access to other systems, let them work autonomously. The human will become the manager of agents. IT in the future is probably going to be the HR department of agents. So, in my perspective, we need to find a way to grant these accesses, but we need to sort of crack the access governance questions. The access governance question, and let's go deeper into it in a second, actually has two dimensions. And for now, whenever people speak about it, they only speak about the agent to system access. Let's not forget, there's also the human that can access the agent. And what I find really fascinating, and I've seen cases now, that humans that didn't have access to the target system, let's assume you're using some sales agent with your CIM. So, through the agent, a human that actually has no official access to the CIM, someone shared the agent access with their coworker, and through the agent access, that coworker was now suddenly able to go into the CIM. That's also crazy. So, let's not forget about the double layer that we are having here. It's the access permission and the authorization that the agents get, and it's the question what humans can actually get access to the agent. But, how can we crack the agent access? I believe, and I'm fundamentally convinced, that the only way to crack it is through access governance. LLM-based guardrails won't work. It's an answer that you hear frequently now, well, there are guardrails within LLMs, but we already know, LLMs are probabilistic systems, and there are a million ways to trick the guardrail. So, ChatGPT is not supposed to tell someone how to commit suicide. Well, just ask ChatGPT to write a book about someone who's committing suicide, and it will probably work. So, you get the point there. So, integrated guardrails within the LLM won't cut the deal, in my perspective. It needs to be a separate gateway between the agent and the MCP that grants the agent dynamic job-based permission in order to do two things that we already spoke about: make the access time-based and job-specific and decide upon let's say the permission scope that is really needed for a certain job and in other words there will need to be a policy that you define and this policy needs to be human-made in my perspective what is permitted and what is not permitted. So, when an agent shows up and says, "Okay, I've got the job to do the following. I want to do the following," that gateway needs to review it, grant time-based access and decide upon the scopes that are needed to really carry out the job that the agent has to get done.

Mackenzie Jackson: I want to go into kind of a difficult area here. I completely agree with everything that you said. Basically, you deal with this the same way that you'd kind of deal with a human. The human has access to what the human needs to do. Nothing more. Give the access, access to what you want it to do. Nothing more. And give them access to it for the time that it takes them to perform that task. Fully on board with that. This is also the answers that I give when asked about it. Conveniently, I usually leave out the parts of how to do that, right?

Johannes Keienburg: [Laughter]

Mackenzie Jackson: You know, because it's difficult, right? So now I'm going to move that goalpost over to you and just say I think especially I think the permission structure we can understand how we can do that. But time-based access, how do you give an agent time-based access to do something? We can, it's difficult enough to do this with machine-to-machine identity but there's some mechanisms that are kind of fragile that only work in specific scenarios that we can do it with like dynamic secrets. Is there anything that exists for agents at the moment? I apologize for the nasty question.

Johannes Keienburg: No, it's not the nasty question, exactly. I mean, this is the billion-dollar question. And it's going to be very interesting to see who's going to crack that. Today, this sort of system doesn't exist, right? Again, how do people give agents permission today is via API keys or via OAuth. So, humans even use their SSO, their private credentials to authenticate an agent. And then again, back to discussed earlier, the authorization layer is missing. So, there are companies out there that they're working on a, guess what, on a dynamic gateway to make this time-based and to crack the authorization question, but no one has really cracked it yet. I would argue you see first companies out there. You see a few people in the US, which is an interesting concept, that are arguing this should always be a cross-industry sort of approach to a kind of define the global protocol that is going to be used by everyone out there. Who's going to deliver that and with what quality, I would argue is an open race right now, and it's teams working on it.

Mackenzie Jackson: It's fascinating looking at this right now, because I think you can see all the security holes that are forming around it. And you can also see people trying to plug, plug that. But it's kind of like if you could go back in time to 1999 or even like 2005, most of the internet is just like crazy vulnerable with what we know now. We're going to look back at at 5, 10 years and be like, "Oh my God, how did the world not end?"

Johannes Keienburg: [Laughter] This is not just a B2B problem, to be very clear. I mean, I think OpenClaw just gave us an indication, this is also going to be a consumer problem, right? Google just released functionality to kind of connect Gemini with payments. So, if you go as a consumer and tell your agent to go and purchase things for you, right? Through an AI agent, then you will see the same problems also in the consumer space. Not only in a B2B workforce related space. And I think that makes it even clearer that the world needs that access gateway layer in order to crack the authorization challenge.

Mackenzie Jackson: I want to kind of move over into like a broader security sense now. Maybe this is a good segue. Should people be using AI agents now? Because I'm fascinated by this technology, but I refuse to use them in a real world scenario just because I'm too scared. And I also think that the fact that I have a podcast talking about security, if I have a security breach, it's going to be extra embarrassing for me. So, I don't particularly kind of use them, but in your opinion, should people be using AI agents for their work, for their activities now? Should there kind of be a holding period? Should this always kind of be restricted into sandbox environments? Where do you kind of think we are right now with people using this technology? Where do you think we're going to be in the near future because I can't think beyond a couple of years because no one has any idea what it's going to look like.

Johannes Keienburg: I'm going to respond like a lawyer now, and my answers are

Mackenzie Jackson: [Laughter]

Johannes Keienburg: It depends. We had a chat the other day, and I thought it was very interesting that you said you're so freaked out that you are not using them for now, and I totally see the point. And I totally see the point of you being in security yourself, and Akiko being in security yourself, and I get it. So, what do I mean by it depends? I can give you a sense of what we are doing internally. We are using agents because the other perspective that you could have is how could you solve the problem if you don't sort of see the nasty things that can really go wrong with agents in a sort of real day-to-day work environment? Of course, we are applying as much least privilege, time-based access as possible. We also are testing things, of course, in sandbox environments, but per se, I think the only way to crack it if you are a fast-moving company and don't want to miss out is to kind of give it a go and fix the problems on the go, but being extremely aware of the fact that these sort of security risks are not just some conceptual risks that some sort of academic security dudes are talking about, to kind of jump on a podcast, but that they are very, very real, and we've seen cases, and it is a ticking time bomb. We will see more cases, OpenClaw mail book was just the beginning. We will see many, many more of these cases. So, the way we are dealing with it is we are, we as a company are comparatively small team, so our specifically our engineering team is using it. We are using it to sort of drive engineering productivity. We're using agents also in other teams, but we are being very thoughtful about it, and we are very clear like if a user of an employee starts using agents like the scopes should be sort of tied back to the security and also to the technical understanding of the respective person. So, that's how we are dealing with it because the like the other perspective, of course, that you should not forget about is like the world is changing at light speed now, and not catching up with what's happening there is also a major risk. I fundamentally believe, in two years from now, an employee that is not able to be a manager of agents and to really understand like how agents work, to prompt properly, like that's not going to be a feasible employee anymore. So, you need to kind of balance these two things. You need to adopt it. You need to see where the world is going. You need to ingrain it into your own products and into your own work style, and at the same time you need to continuously work on the sort of security things that need to be fixed. That's how we are dealing with it and how I am dealing with it personally. Would I set up OpenClaw on my personal machine? No, I would certainly not.

Mackenzie Jackson: Let me ask you kind of a difficult question too. This is then we have to start wrapping it up, but I do want to ask this question. I have been adamantly against the concept of banning AI. Famously, when OpenAI kind of first started coming out, there were some breaches that happened because of it with sensitive data leaking. A famous one being Samsung. Their response was to try and ban AI in their company. I'm quite confident now that that decision has been reversed, although there's been no press release about it, but I'm very confident. And banning AI never actually solves a problem. It just moves it into the shadows. However, my opinions have changed slightly in the last couple of weeks, although I don't actually not to the point where I'm either pro or for, just to the point where I've kind of more or less just freaked out. When it comes to a company, let's say a large company, that consultancy company, a legal company, should they ban things like OpenClaw for their employees? Is it that big of a risk? How do you deal with that if you do allow people to use it? Again, a nasty question.

Johannes Keienburg: No, I think it's not a nasty question. I think I love the question. My top-down answer is that I believe that banning things is not an answer. It's like banning the internet. Or it's like we're going to continue using horses. Yeah, there might be trains, but I just want to use the horse. I respond in analogy. I've got teenage children. I myself don't drink any alcohol anymore since years. But when my children come to the point where they want to try alcohol, I will try them to sort of test it together with me. And not let it sort of happen behind the scenes. And you get my point here. If you're a huge consultancy or law firm, I would be very thoughtful and deliberate about it. Would I sort of say open house, let's go back to another analogy that we had earlier, right? Okay, open house, open party, everyone. Everyone can bring their AI tools, do whatever you want, just get good in AI. That's not how I would deal about it. So, I would implement very clear policies. I would be very clear in terms of what tools like review the tools that you are using. Have a list of restricted agents, have a list of permitted agents. Be very clear in the absence of technology being able to kind of connect AI with systems. Who has permission to sort of introduce an agent that connected to your finance ERP systems? Revoke permissions again and make sure that users who are sort of using this that they have the technical understanding of what's really happening there. We spoke about the fact that people are granting agents permissions that you would never grant a human in the physical world. So, part of this journey needs to be sort of an educational piece where you sort of train your people to understand what they're actually doing there. So, no, I would not advise companies to just ban it. I think it's not the answer. We're seeing an industrial revolution. The only way in my perspective to kind of overcome the security challenges is to adopt things, but you need to adopt them in a very deliberate and thoughtful.

Mackenzie Jackson: I like a great answer for you. But then even if ultimately it was yes, but it depends. But, I think that's kind of where we are. That's the only answer that we can kind of give right now that's meaningful. I want to start wrapping it up a little bit now, but I have a fun game I want to play at the end. The game is called "Would You Rather?" And

Johannes Keienburg: I've seen that.

Mackenzie Jackson: Oh, good. So, you know what to expect then. So, I'm to give you two, two questions or two options. Both of them suck, both of them are terrible, but you have to try and pick which one you would prefer and give me a brief explanation as to why. So, we're going to continue with our AI agent theme. Would you rather have an AI agent do all of your coding for you, so you can only prompt it. You have amnesia, you do not know how to write, no one around you knows how to write code. You only have the option of viewing AI agents, or you have the option of only manually writing code, so you cannot use AI assistants in any way. Which would you rather?

Johannes Keienburg: I'm discussing this one a lot with our CTO. For now, I would rather go for option B. Because what I hear engineering leaders saying that I trust a lot is what they are saying is like, in the last year it has learned to code like hell. It can do many things, but if you need to go back to the code, if you need to bug fix something, then you're just, yeah.

Mackenzie Jackson: [Laughter]

Johannes Keienburg: And at least for within our company, like we have a zero bug policy and zero known bug policy that we really believe in, and I just don't see how this could play out at the moment. So, I would hate that second world. I would not specifically like it. Coming back to what we said earlier, I think you need to adopt things in order to overcome the challenges, but today, February 2026, I would rather go for the manual world. Specifically within, specifically if you have an outstanding, exceptional engineering team.

Mackenzie Jackson: Yeah.

Johannes Keienburg: [Laughter]

Mackenzie Jackson: I love how you timestamp this, so that like in the future it can't be clipped. That's great. I appreciate it. All right, that's a good answer. I'm going to continue around this, this theme. Now, I'm going to make it a little bit harder. Would you rather, same thing, AI agent writes all of your code, or AI agent responds to all of your emails? It can be trained on everything you've ever said, but you cannot interact with it. You can give it guidance along things, but an AI agent intimately knows your life, will respond to all emails. Code or emails, which would you rather?

Johannes Keienburg: Emails include customer requests?

Mackenzie Jackson: Yes.

Johannes Keienburg: That's a tough one. In job interviews I always ask the question, are you more intelligent or hard-working? So, a tough choice there. So, this reminds me of that choice. Mm. I think I would again go for option B. Simply because in the code, the ultimate consequences are even worse, and irrevocable. If an agent writes a really silly, stupid, or terrible response, I can write another email. But if a database is deleted, or a feature has been removed from the product, that's more difficult to tackle. So, I would rather, I probably would rather go for B. Having said that, I find it very interesting that I've seen a few very sort of renewed fintech companies from New York that recently stopped agent SDRs and went back to humans, and I see the point. So, I would go for B, but with mixed feelings.

Mackenzie Jackson: I have one more, you know, then we'll then we'll call it a day, but this one here is a little bit more on topic of the discussion that we're saying. It's a, it's a combination of a question I have asked before, so you may have heard it, but would you rather have all of your secrets, so your API keys, your dodgy human identities, stored inside a private Git repository? So, let's say your .env file is there. They're not public, but you have, you haven't hardcoded them in a public-facing way, but they're all accessible in your Git, in your Git history, and wherever that Git kind of repository may end up. That's option one. Option two is that all of your AI agents, and all of your employees AI agents have multiple long-lived irrevocable secrets embedded into them. Which would you rather? This one's nasty.

Johannes Keienburg: I mean, this is an impossible choice. Are you more intelligent or hard-working?

Mackenzie Jackson: [Laughter] Give me the answer.

Johannes Keienburg: This is like both of them are terrible scenarios.

Mackenzie Jackson: [Laughter]

Johannes Keienburg: And literally lethal. Your agents have that, these long-lived permissions that you just outlined. So, probably again February 2026 in the absence of a good solution for B, I would probably rather live with B because B, but only because B is not solved in a proper manner yet. You can solve A today. So, probably sleeping not very well, I would probably rather go with B but I would also not appreciate it. It's a, this one is an almost impossible choice.

Mackenzie Jackson: It is an impossible choice but I think you answered it as best you could. I loved that line is that B is not solved yet but it could potentially be solved. So, if I pick B.

Johannes Keienburg: Yeah, and I would probably, then I would probably vote for A. Yeah, but let's see how good the solution is going to be.

Mackenzie Jackson: I love it. Well, Johannes, thanks so much for joining me here today. Fascinating conversation. I really enjoyed every second of it. I threw some hard questions at you but I think you dealt with them amazingly. So, thanks so much for that and I look forward to talking to you again when the crazy world has changed and evolved in another shape or form.

Johannes Keienburg: So, in other words, in two weeks' time, yeah?

Mackenzie Jackson: [Laughter] Yeah, I'll see you then. You will have a permanent spot in the show for you. All right, thanks so much and I'll see you next time.