This week's group discussion topic for Tuesday, October 6, 2026 will be:
Artificial Intelligence (AI) may prove to be the most significant technological change since the commercialization of the Internet in the 1990s, or the invention of the semiconductor. Early AI pioneers to-date have framed AI as a tool to automate email workflows with features like summarizing emails, drafting and sending auto replies, organizing incoming mail, and more. Other efforts underway include introducing AI in to the analysis of inbound email for spam, phishing, and malware identification or parsing log files to identify patterns and watch for abnormal conditions. Where do you feel AI is hitting the right targets for pain points experienced by end users and email admins, and what are the opportunities still waiting to be addressed?
The mp3 audio recording for this week's call (October 6, 2026) is now posted: https://nextcloud.zetalliance.org/index ... if6zxaBAiR . Navigate to the 2026 folder, then the 10-October folder. You are invited to download a copy to take with you to listen to later.
Our next topic for group discussion will be on Tuesday, October 20, 2026 and will be announced a few days in advance of the call. We encourage you to join us on the next call (https://zetalliance.org/contact/). New call participants are always welcome, even if you initially just want to listen in live without speaking on the call.
This is an AI generated transcription of the Tuesday, October 6th, 2026 Zeta Alliance call mp3 recording. The transcription may include minor word translation errors, or have assigned incorrect speaker names with voice auto detection. If in doubt as to the accuracy, refer to the call's mp3 recording. Thank you Matthew Francis, of In-tuition Cloud Services, for assistance with generating the AI transcription.
[00:00:00] Randy:
Welcome to the Zeta Alliance call for Tuesday, October 6th. My name is Randy Leiker. I'll be your host for today's call. On the call with us is John, Matt, and Rollins. And for today's topic, we're going to be discussing artificial intelligence and how it integrates into email and the hopes and dreams of what it could be and the pearls and dangers of what it might not be. So today we want to talk specifically about with artificial intelligence being the most significant technological change easily since the commercialization of the Internet in the 1990s and possibly even the semiconductor even decades further back. Early AI pioneers have been pushing AI as a tool for automating email workflows and features like summarizing emails, drafting and sending auto replies, organizing incoming mail, and more, to basically take some of the burden off of managing a mailbox, really. But on the back end, we've also seen efforts to integrate it into things like spam and malware filtering and parsing log files to look for patterns, watching for abnormal conditions, automating a lot of the mundane server administration work that goes into running an email server. So the topic today really is going to focus on where do you feel like AI is hitting the right targets for pain points experienced by email end users and email admins, and what do you think are the opportunities that are still waiting to be addressed?
[00:01:34] Matt:
I'll offer myself up. Thank you, Matt. I think it's hitting the right points with regards to helping admins review logs, you know, tally up stats, things like that. It's a great help in investigating issues and triaging problems. I don't think it's hitting the mark. I think it's hitting the mark with actual detection of spam and so on. I think there's a lot of improvement that can be done there. I mean, SpamAssassin doesn't even have an LLM module incorporated into it, as far as I'm aware. I think rspamd has support for it now, but there's a lot of work to do there. And I think it's won greatly in the spammer's direction in terms of them using it to get spam delivered, and we're not doing enough with it for defense.
[00:02:32] John:
And I would agree with that. I would agree with that also. It's an arms race with the spammers and filtering tools that are out there as it is right now. We have some things that are going through SpamD, as you said before. Stalwart has some interesting things that they're trying to do with LLMs and add-ons to be able to find out fuzzy logic on spam. I think there's a lot of work that needs to be done there, as it's coming in. I do see AI being used mostly for day-to-day maintenance type of redundant tasks that you would normally do, verifying backups, looking at patterns of what's coming in to find out whether or not we've got people that are abusing things on the inbound or outbound side. As far as using AI to be able to reply to emails. It winds up becoming very interesting when you involve AIs in that scenario, is that the AIs wind up talking to one another and it just creates a lot of traffic. And the end user, the actual individual who needs to read the email, never reads the email because it gets lost in the mix. So we don't see it happening from the productivity side where people are finding that it's useful. When they've got their CRM tools and they're using... You know, like they're using HubSpot or something like that that's mining their email as well. That's really not doing anything any deeper than a standardized API or a filter that we would set up normally. I'm not seeing anything all that exciting in that world yet. I do wish I could see something better on the phishing and the spam filtering, though. That's really where I feel AI could do the best for helping productivity and reducing noise in systems.
[00:04:27] Randy:
Yeah, and I think it's a true irony that AI has made it so easy for the bad guys to craft exceedingly good phishing emails. It really has been the catalyst for a lot of that innovation that's happened. You know, long gone are the days where phishing could be spotted with misspellings and bad grammar choices and that stuff is just, it's in the history pile at this point. So it's kind of funny. It's kind of funny that we're talking about using AI to fight AI, but that's kind of what the world seems to be coming to. All right. Rolands, do you have anything on that topic?
[00:05:06] Rolands:
Yeah, I can add that what I see that only like platform as a service where we have some antivirus or spam protection, there is no product. What you can deploy locally and use LLL at least, I don't know, like we are using SpamTitan in front of our Zimbra and yeah, they have some fancy kind of service what they provide, but we cannot deploy them locally and for us, it's like we use Zimbra because of local deployment, and everything is like in our hands, yeah? And yeah, that's why, yeah, that's what I can add kind of.
[00:06:08] Randy:
Yeah, and I think, Rolands, to add to what you're saying, another tricky part of this whole thing with integrating AI with spam filtering and phishing, as John was pointing out, is that you've got AI token usage to consider too, you know, with the incredible metric increases that we see in spam and phishing, you can burn through a lot of AI tokens in a big hurry, and it's not something that, you know, you really have a whole lot of control of because if you're going to use it, you're basically committing to have to dump a bunch of funds into that or reserve a bunch of funds to do that so that you literally never run out of tokens, especially if you're running it in production, and I think that's something that probably isn't well understood.
[00:06:53] Rolands:
Yes and no. No, I think because, like, you can queue those emails, right, and if, like, from my perspective, I see that you have some box where you have a local LLM, yeah, I don't know, virtual machine or whatever, and then you have some service using that local LLM, and then that everything, you know, happens in your hands, kind of, and if you, okay, if you get a lot of emails, then it will, the queue will grow up, but, yeah, you need to manage that, maybe after a week, you will understand, okay, I need better hardware for my local LLM, yeah, or, yeah.
[00:07:52] Matt:
I think, yeah. I think it needs a different approach slightly. In that, you know, a lot of the scanning we do is at SMTP time, so we can issue a deferral or a rejection of the email to the original sender, but with, you know, LLM scanning of the mail and the resource usage that it requires, especially if you're under a DDoS attack or something, you just can't do it at SMTP time, so what do you do in that case? I think. The answer is to deliver the mail as normal without that scanning, and then post-scan it afterwards, and perhaps move it from, you know, in Zimbra specifically, I'm talking about here, is move it from a user's mailbox to their junk folder, so that you haven't, you can't delete it, because you've told the original sender that it's been delivered, but you could put it in their junk folder, for example. And that could be a post-delivery process, so that you're not reliant on the time limits that you have at SMTP time.
[00:08:58] Randy:
Yeah, and that's a very good point, Matt, because we do have to consider here that SMTP on a lot of servers does not allow for long periods of time to pass before a scan completes, and maybe that's even longer with an LLM and AI basically getting involved. As Rolands's pointed out, that's a good point, Rolands, about running your LLMs as a local resource rather than the cloud on somebody else's data. data center, because I guess that does give you a degree of control around the token usage, definitely. But to Matt's point, I wonder if we eventually will see AI be integrated into spam and phishing filtering as kind of a tiered structure, maybe. Perhaps we use our conventional defenses, our SMTP protocol tests, and, you know, looking for good and bad behaviors in SMTP sessions first, and then all of our conventional spam assassin filters and all that. And maybe at the very tail end of the process, maybe once an email has gotten through all the way through that, it still looks good. Maybe that's the tier at which an AI gets involved and looks at the email as like a final pass or something. That could be an interesting application, maybe something awaiting
[00:10:12] Matt:
to be explored. Yeah, or have been the initial, you know, have multiple passes of AI on the email. You know, the first one, for example, could look at the hello, the sending IP, and so on, and try and, you know, determine the likelihood of it being spammed from that, like RBLs do, and so on. But just to a greater extent by looking at all of the headers that it sees. And then only look at the data part later on if it needs to.
[00:10:43] Randy:
Right. And we kind of saw the same evolution happen with even email storage. You know, it used to be, you know, ages ago that you could block all the mail on the same disk, and you ran out of disk space, you went out and embed a bigger disk. But of course, as time went on, that strategy doesn't work anymore. Nobody can ever buy a large enough disk for all the mail. And so now we've got things like tiered storage, you know, where we've got multiple levels happening. And I could easily see that happening with something like this, with the spam and the AI filtering and all that. So that makes sense. So aside from spam and phishing filtering, just to jump off of that for a moment, what are some of the other things that you're seeing happening? What about the backend experience? What about, and this is something John touched on a minute ago, but what about the analysis factor? Something I personally have seen is, especially for newer email admins, they lack that depth of knowledge about maybe how protocols work or what the norms are on the internet, you know, between servers of exchanging email. And AI could be potentially a big tool for helping them analyze problems, or really anybody, for example. So I think it's a great question. I think it's a great question. I think it's a great What do you guys think about that? Do you think that that niche is being filled, or is there a long
[00:11:58] John:
ways to go on that? Two things on that point. Just a generalized question for everyone who has been using AI, and that's everyone on this call really is the question, is are you finding that you're relying more on AI to do the tasks that you used to have to think through? And is that a benefit or a disadvantage? I think it's a big question. I think it's a big question. I think it's a big attraction to you to do that. Like, there's a lot of things that I'm doing now, and I'm having to catch myself in realizing I'm getting lazy, that I'm allowing the AI to do things, to do some of the tasks that you really do need to remember. Because as you said, Randy, when you've got a new administrator, or you've got somebody who doesn't have the depth of experience, they're never going to gain that depth of experience if they don't do it. If they're relying on an AI to do it, and something doesn't fit the pattern that they're using, they're not going to be able to do it. That even you can describe in a prompt to an AI, then what do you do? You don't have the knowledge to be able to look at the system and find out what the system is doing. Are we losing ourselves in
[00:13:03] Randy:
that? I would say that we're definitely losing an element of critical thinking, no doubt about it. And you could make the same argument that that started even earlier before AI with the advent of internet search engines. How many times have you heard stories of somebody working on a project sitting there constantly running Google searches trying to find answers that they don't know? I think AI has made that dramatically easier because now you can ask questions in natural language. And it definitely kills off that critical thinking part of it. And I even saw an article the other day that was making that same argument for programmers and coders who are increasingly pushing back against AI because they're finding that it takes that critical thinking part away and that curiosity factor away. Because if you talk to most programmers that work day in and day out in the field, that's really what drew them to the field in the first place. They love that problem solving aspect. They love the critical thinking aspect. They love the curiosity part of it. And AI just kills all that because literally it's writing the code for you. It's trying to deduce and figure out the logic of the problem for you. And yeah, that could very easily happen on the email side. What do the rest of you guys think about what John asked? Matt, Rolands, Jason?
[00:14:26] Rolands:
I can add, I think that if people are lazy, they will be lazy. And there is people who like to understand things and they will understand at the end, with or without LLM. What I can add is that we had some problems, not the email related, but some website. And it's like a minute when you can debug all the process, all the logs, everything, and get some results and then find the issue. Before, it can take maybe hours to understand what's happened and why. And then you can, I don't know, get your website back up and running. Because of some syntax error or whatever. But yeah, I look at LLM and all the AI as a tool. And that tool gives us more power, kind of. And yeah. AI is an excellent tool for log parsing. No question about it.
[00:15:50] John:
It does an excellent job when it comes to that. We've had, you know, a couple instances within the last couple of months. And it was able to run through, you know, gigabytes worth of logs and find the pattern, find the problem within, inside of a minute. You'd never be able to do that as yourself. I've just found it to be very good for that. Where I have a concern about it is that when you start relying on AI, like you were saying, Randy, to do the critical thinking for you, that's where you wind up losing out. Because even then, if you're feeding it all the logs, you still have to tell it what to look for.
[00:16:26] Randy:
Exactly. Jason, Matt, do you guys have thoughts on this?
[00:16:31] Jason:
Yeah. Yeah, I agree that it can have impact on critical thinking. But what, and I agree that it is a tool. And that's how, that's what my approach is. Say, you know, when we have an issue, if we utilize, or if I utilize AI to find, tell me, find and resolve the issue. I'm going back during the postmortem. And when I'm writing the root cause analysis, because we have to do that here, you know, because we have a whole PI team, and we have to, we, anytime there's an issue, we have to, we have a postmortem, we have to write it up. So I use that process to kind of help understand what happened, rather than just writing it up. Rather than just taking the quick fix as a quick win and just moving on. I go back in, I dig in to the evidence and kind of understand what happened, you know, rather than just brushing it off as, all right, we solved it, let's move on to the next thing. So it's just, I guess it's just more, it's whatever the personal preference is. If you, if you want to understand the problem or figure out, you know, what's going on, you can do that. What happens, you know, that the person will do that. But use the tools that's afforded to you to quickly resolve the issue, but don't lose track of actually having to understand what happened. So, you know, that's my approach.
[00:18:08] John:
I'm sorry, I'm sorry, I'm fully willing to admit that I've used AI to be able to write the postmortems.
[00:18:19] Jason:
Polish it.
[00:18:30] John:
Uh, so I can compose an email that's, that's pretty eloquent without having to use the AI. But I've lately, what I've done is I've kind of switched to writing emails or writing any kind of message that I needed to write with stream of thought and then feeding it to the AI and having the AI clean it up for grammar and flow. Do you use it, use AI for that as well?
[00:18:52] Jason:
I do a lot of times. It just depends on who my audience is and what I'm writing. It's all going to be situational. But if I'm writing something and it's going to be seen by, say, VPs and directors in our company or somebody external, I do polish it up more than, say, just a normal email. But I kind of do that, too. It's like, you know, just here, this is what I want to convey. Can you draft an email or something? A lot of times I just throw that in a copilot. That's about the only thing I use copilot for because it's in Teams,
[00:19:34] Randy:
So, Matt, let's hear from you. What do you think about that? Do you think that there's benefit in using AI with log file analysis? What do you think about the pitfalls and benefits of it?
[00:19:44] Matt:
Absolutely. I completely agree with what John was saying, that log analysis. You know, searching for patterns, stats, you know, it's amazing at that. It's so useful, such a useful tool for investigating problems. It often gets the diagnosis wrong. So you do need to check what it's telling you and take a bit of a pinch of salt. But it's extremely helpful.
[00:20:18] Randy:
And I've even heard stories from other fields as well, like healthcare, for example. Very good example. You know, I've seen stories of radiologists talking about the ability to use AI to look at years of previous, you know, x-rays and MRI scans and all sorts of things for patients and be able to spot just the things that they're doing. And I've even heard stories of radiologists talking about the ability to use AI to look at years of previous, you know, x-rays and MRI scans and all sorts of things for patients and be able to spot just the slightest abnormalities across different scans. And I think really that's the use case for this in email logs is the ability to see those long-term trends and aggregate that together and recognize those patterns that a human just simply can't easily, you know, understand. So, another aspect of this that we haven't really gotten into too much is what about on the front end? What about the end user? And John just briefly touched on this. Where he was commenting, you know, about AI is literally starting to talk to each other through email, which kind of diminishes the benefit. But do you feel like AI has a place in the end user's mailbox? Is that overkill or is it a benefit or is it going to basically kill email as we know it as a useful tool? What do you guys think about that?
[00:21:38] Matt:
I haven't given it access to my emails. You know, I only use it to compose, to help compose text sometimes. But I still control it. I control and edit it. So, in terms of users actually doing that, I mean, we get a lot of more detailed support emails now, I think, where users have used AI to look into the problem themselves and then contact us and provide the information that it's telling them to send. But in terms of actual users, you know, completely moving their email flow over to AI, yeah, I don't know about that.
[00:22:18] John:
In the CRM realm, and I know Jason's involved in that as well, right? How do you use it on your side? Because we use it for tagging, sorting, and organizing. We've got some customers that are using 365 and HubSpot, and they want to be able to use AI to be able to generate dashboards from information that comes through on emails. Haven't found it to be all that. Useful yet, although it's not that far away. And the question would be directed to Jason, because he's in the CRM business as well. What do you think of that?
[00:23:01] Jason:
I haven't been too much on the CRM side of things lately, or for a while. But the way we've approached it, we've actually, so we have three pillars of business. We have our. . . We have our tech side, which is our, like, services side with CRM, hosted email, imaging, compliant text messaging, all kinds of other stuff. And then we have our compliance pillar and our actual financial services pillar. So we're like a. . . Our focus is like a turnkey solution for, like, RAs, you know, out of the box, have everything. And so initially we. . . Started. . . Started dabbling in AI, you know, like AI, like for notes assistance and things like that. And then we actually have built our own back-end AI called Denali that's going to stretch across all of our platforms. And so we've taken approaches where we're going to offer AI in. . . In flavors, so there's, if you will, like, there's varying degrees of integration that you can, that our users can add on or subscribe to. I haven't dug into those or been a part of those discussions, so I don't know what the varying degrees are with that. But that's kind of our approach as far as, like, because some people want a lot of AI. Some people want a little. . . So we want to kind of give them the choices that they would like. But we're considered, they want us to consider us like an AI forward company internally and externally and our offerings as well. So we're kind of all in and all over the place. You know, we use it in some workflows. We use it in aggregation of. . . Data in regards to positions and things of that nature on the financial side. So we have our hand in a lot of pots, and it's too much for me to know at the moment. But we haven't really implemented it in any on the email side. Of course, I don't get any development resources, so it's all on me.
[00:25:45] Randy:
But I probably don't. . . I'm sorry, Jason.
[00:25:50] Jason:
I know that probably doesn't answer the question fully, but that's just kind of where we're at with our email or our AI offerings.
[00:26:01] Randy:
Okay. Sorry, I keep speaking over you guys. Go ahead, John.
[00:26:06] John:
No, I was. . . If you have something to add, add. And then I'll. . . Because I was just going to talk about CRMs and ERPs. Sure.
[00:26:15] Randy:
So I think one of the dangers with giving. . . AI access to a mailbox in general is obviously a loss of control. There are plenty of things in popular media about that. I count easily 10 or 15 movies I've heard of or seen that always have a scene where the character is asked by the AI, would you like to grant me access to all your passwords of your accounts and your email and everything so that I can do things on your behalf like making purchases and responding to emails? And, of course, that's always the turning point in the movie where very. . . bad things follow for those characters, usually caused by the AI. But then there's comical aspects of it, too. I could think of a movie where a character gets a new phone and is asked by the AI on the phone, would you like to read the terms of service? And the character says, well, no. And the AI replies, stupid, and continues on and takes over his life, essentially. So there's that aspect as well. So for regular end users, even outside. . . outside of a business setting, a business email setting, you have to kind of think about the risk aspect of it as well. Does it make sense to give AI that level of control? Do you trust it that much? And, again, does it diminish the benefits of what email is really all about?
[00:27:36] John:
I had this discussion not that long ago with a large networking vendor who shall remain nameless, but they're a multibillion-dollar company, and you would definitely know who they are. I was talking about using AI and putting AI in their networks to be able to determine where network faults and problems would be. And the biggest question that we had right off the bat is, where's the data leakage going to go? Because they were talking about even using it. They were saying, oh, we're going to have our own LLM. That's great and all, but you're taking a financial institution or a hospital system or something along those lines, and you're telling it every single endpoint and all the details to be able to get into them. You're putting the security risk involved in that is just insane. Talk about a way to be able to have somebody be able to breach it by way of a chatbot or something, which has been proven as well, that you can go into a chatbot and have one of these audio chats that are hitting you from one of the UCaaS providers. And you can do something simple like just say stop script, and the thing will stop the script if they don't build it right. And then you can turn around and start asking the questions, and the LLM will happily go ahead and give you all the information that you're not supposed to have any remote access to whatsoever. So there's a lot of concern about how you implement AI, what you allow it to see. What we do, we have the CRM and the ERP systems that we build out. We're a SaaS provider for that, and we work with a lot of other vendors. And we find that a lot of these plumbing vendors, I'll use the plumbing as an example because they just were the most recent of the bunch, want to be able to check inventory. On certain providers, and it was allowing, they wanted to be able to try it where they would use the AI to read their email and then be able to go on to the systems that they work with on a regular basis, the vendors to be able to prove stock and get pricing and things of that nature. And it turns out that the information that it was getting back was wrong, just because it doesn't know, unless you tell the AI every aspect of your business and every aspect of your business, and every aspect of one of your vendors' businesses, it's not going to be able to know where to grab the proper information to give you that dashboard or to give you that report so that you can turn it around, put it in a quote, and send it out to a customer with any kind of reliability.
[00:30:03] Randy:
Or worse, it takes it upon itself to go out and get that information via hacking, which we're starting to see more in the media now about various hacking between AIs.
[00:30:13] John:
Yeah, and the hacking on the AI side was also, I've delved into that a little bit. Got some, it's not really inside information at all on that one, but the hacking attempt that you've heard about in the news, it was prompted to do that. The AI did not decide on its own to go outside of its network and try to hack the other network. It was told to. They have the actual prompts that were in there that told it to do that. So the things that you hear in the media, just in general, any kind of media, you have to take it with a grain of salt because they'll give you part of the story. They won't give you the whole story. They won't give you any of the story. They'll just give you a story. There's a lot of things that are out there that we don't know about at this stage. Just leave it there.
[00:31:08] Randy:
So how about others here? Rollins, Matt, do you think AI has a place in end users' mailbox for this type of thing?
[00:31:16] Rolands:
I was thinking. I was thinking about this some time ago. I know that there is some people who give access to their email to some agents, OpenCloud or Hermes or whatever, but I see a lot of problems there because I don't know what he can decide. And I cannot set some guardrails there. At least I don't know how to set them. If you give right permissions to your mailbox, it will do everything, right? And there is some maybe problems with, at least with Zimbra, for example, that you cannot give some permissions. I don't know, like to sort your email, for example. And that can help. And I think that will be our future, that all the LLMs, the agents will help you, some kind of. And that's why, like, what I see that all the included kind of, like Microsoft has Copilot or whatever, where it helps you with exact product what you are using. And where Microsoft sets some guardrails on that and permissions and everything, kind of. So, I'm kind of waiting on Apple when they will, I'm in Europe and we don't have even the newest Apple intelligence and like where it's, at least I hope that there will be some guardrails and that I can trust, because like everywhere it's about the trust, like, do you trust it or not? Yeah, at the moment, I don't trust it. Yeah. Like, if I use for coding, infrastructure coding, I will generate the code. I will put that in Git, will review, and then I will put in, like, production or test or whatever. I will use that code, yeah. And I don't see, like, how it can be done in email.
[00:33:38] Randy:
Do you perhaps see it taking the form of a permissions model like we've seen in the app stores for phones and tablets?
[00:33:46] Rolands:
Yeah. But if your agent will ask you permission every time, that will be, like, not interesting, kind of, yeah.
[00:34:00] Randy:
So, the risk there is too much permission, maybe like what we saw with early releases of the Windows user access model, where, you know, every single little thing was a prompt, you know, nagging users and getting a lot of negative publicity because of that. And causing people to go, "Well, screwed, I'm just going to run as administrator for everything because I'm tired of seeing all that stuff." But on the other hand, maybe guardrails can be created in predefined roles to minimize that prompting. So, maybe we have to go through that same growing pain period like we did with operating systems, you know, where we saw an overabundance of prompts until finally that kind of leveled off to something where you weren't just overwhelmed with it. So, does that sound fair, Rollins?
[00:34:46] Rolands:
Yes, but in email, there is no permissions, like, there is, like, two permissions, you have permissions or you don't. And, okay, maybe you can give, like, read-only permissions to your mailbox, for example, but then it's read-only, then it's kind of, yeah, it can generate some, I don't know, it can read your email, for example, and then generate some reports, some, I don't know, but it cannot do the, maybe, sorting or something, yeah, what you want to do with your email, or maybe there is some emails where you want to make some workflow with the automatic replies or, yeah, and then use LLM into that. So, how to do that, at the moment, I don't know, yeah.
[00:35:50] John:
You've got a scenario of using the right tool for the job, too.
[00:35:56] Randy:
Yeah, and I think there's strong tie-ins between your point, Rollins, and John, in that, you know, in addition to a permissions framework in order to safely allow something like that to work, then you also have to consider what John was talking about with, where is your data going? You know, how do you control that? Is that ending up in some LLM model out there? You know, is that private to you? Is it not? How do you create those restrictions? And I know a lot of people's default answer to that is, well, I'll just run the LLM locally. Well, yeah, that might work to a degree, but that leakage part is so difficult to define, much less control at this point, especially in this early stage. So, Matt, do you have anything to add to this? Where do you think we go with it?
[00:37:00] Matt:
you know, there could be AI used on that platform, for example, to protect them from scammers. And I don't just mean like typical phishing emails that we already try and block, but more advanced analysis of the mail over a historic time. So, you know, for example, you get these romance scams where people have been tricked into thinking they're in a relationship with someone overseas and then being asked to send them money and then more money and so on. And your typical spam filter won't pick that up, but AI could, and it could do that with a guardrail in place where it's only able to read messages and not respond and not act on any messages except to alert you. And maybe something like that would be a good first step for users with AI, especially on this one platform that I run with these vulnerable users.
[00:37:59] John:
I've seen a lot of clueless 30-somethings as well, so I wouldn't limit that to an age group.
[00:38:04] Randy:
No, absolutely. But it's a really good point, Matt, because I think that is a use case that you of us as email admins, I think that's something many of us don't really think about very often, is once we feel like our job has kind of finished with receiving the mail, scanning it for all the usual stuff, and then handing it off to the user's inbox, we're like, okay, it's in your inbox. That's it. But yeah, that's a really good point is what happens if we took that next step and said, hey, now we're going to give you some assistance in the inbox itself so that you can make more intelligent decisions about legitimacy of this email, or should you trust this, or is this notification from a so-called government agency trustworthy, or is it a scam? Stuff that is very, very difficult to do with conventional spam and phishing filters. So yeah, that's a very valuable use case, and probably one that I think is probably not being met very well today. Have you seen evidence otherwise, or is that
[00:39:06] Matt:
just something you think that we need to work into? As I said, I think it's an evolution of
[00:39:13] Rolands:
the scanning technologies we use today sure I want to mention Apple again I think yeah just sorry but yeah like they're like marketing and everything like thinking what they provide that your security like first and then and then that all the AI happening on your device or or in a kind of cloud but it's again with your everything and then you can get the context like all your emails or whatever you have and that AI can protect you actually like I can see where I can trust that and for my grandmother or whatever I can give that and and trust that there will be some solution. But yeah, there is no solutions yet, but I hope that there will be something like that.
[00:40:19] Randy:
Yeah, so that's definitely an opportunity to be addressed. And I think that really moves us past the basic defenses we have today, like adding a banner at the top of an email saying, be careful this email originated outside your organization. Because if you think about it, We're still pretty dependent on rudimentary things like that to basically control, or I shouldn't say control, but to guide a user as to how they should interact with a message. And it would be very interesting to see an integration perhaps where an AI looks at a message and the path that message took and maybe the longer history of emails from that sender or from that mail server or that network and basically added that extra context. and that feels like something we're definitely missing today so it makes sense um it also comes
[00:41:11] John:
back to using the right tool for the job whether or not email is the best place to be doing your workflows to begin with that's and a lot of companies a lot of a lot of businesses in general don't do that they don't analyze what they're using their data for and email is a communication platform that's all it is you shouldn't be using your inventory you shouldn't shouldn't be doing your pricing, you shouldn't be doing your workflow management through the email client. It's just not the right tool for the job in the vast majority of cases.
[00:41:40] Randy:
Yeah, and it could be, too, that what we end up seeing at some point is a separation of the two worlds. Maybe we see some type of AI integration into email, as we know it, both on the front end and the back, but then we see a separate site, just kind of a channel that runs in parallel to that, which might be some kind of API of some sort that AI is used for doing all of that workflow stuff. And really, we have that today, but I don't think it's very clearly defined. And especially with things, I think Rolands's brought it up with OpenClaw, you know, that is an interesting early attempt at truly integrating AI into everything you have. Again, that cautionary tale of how much control is too much control versus how much permission do you give. That open call feels like it's very much at that front line, that horizon, basically. And so it'll be interesting to kind of see where that goes, if we can have the degree of management and permission control and permission model in general, you know, what's allowed and what's not.
[00:42:51] Rolands:
I just had an idea, sorry, that for example, if you give your agent read permissions to read your email, yeah, but when agent replies, it adds that it's agent, yeah. It's not like, he's not sending on behalf of you, but he's sending as agent, yeah, and he says that it's your agent, kind of, like, and yeah, maybe we will be there somehow. It's like your private secretary, yeah, or something like that.
[00:43:36] Randy:
And who knows, maybe that spawns an entire separate industry of agents trying to detect other agents in email. So that obviously diminishes the benefit of email, but still, I can see that happening. All right, so we are closing in on an hour here. So let's go ahead and kind of start to wrap this up here. Anything else you guys would like to add about where you see AI going, things that are not being addressed that could be?
[00:44:04] Matt:
I think it opens up a whole new security exploit avenue, and that needs to be addressed, especially if we're passing in so much potentially malicious data into the LLMs, and there's a whole host of things that can go wrong there. Thank you, Matt.
[00:44:23] John:
Concerns about bias within the AI. I agree with Matt on that. That's a perfectly good example right there, is that the AI is going to start having its own opinions based on how it's trained. And it's going to wind up doing things that you don't intend.
[00:44:37] Randy:
Sure. So a bunch of collateral damage, essentially, as people are early adopters of the technology. Would that be fair to say?
[00:44:43] John:
That would be fair. Plus, where Matt was talking about compliance and governance of data, people have to be educated on what the differences are between a public LLM, a private LLM, and your ability to be able to control data leakage between the two and where you're doing your training. I find that the vast majority of people don't understand that. I don't fully understand that yet, but there are some security concerns that should be of utmost concern to anybody. But what we find out right now is that you've got customers out there, especially, well, not even smaller businesses, but you find businesses in general will take their company secrets and post them to a public LLM and not think twice about it. And that information has been learned and spread about. And I had a conversation with a customer the other day. He was taking his HubSpot data and sending it to ChatGPT directly as part of the $20 a month plan. Reality of it is, is he thought he was getting a good deal. And I immediately turned around and told him, well, your data is being mined. And that changed the whole topic. But getting that understanding out there and the evolving way that AI works right now is where we need to be and where we need to be paying attention over the next few. It's a matter of months. It's not years anymore. It's a matter of months for the next evolution in AI.
[00:46:06] Randy:
And I think if you look at how poorly all of us have handled our management of our privacy and the abuse of corporations in using that data, sometimes with our consent, sometimes without, sometimes with knowledge, without, this whole thing with AI and sharing and adding and contributing our data to AI to supposedly get better results out of it is this whole privacy thing all over again, but now exponentially worse because now it's not just our private information that's being used. Now it's company secrets. Now it's stuff that we're voluntarily turning over and truly can be used for stuff that our personal data never could before from a privacy perspective. So, Jason, Rollins, what do you guys think? Closing thoughts on this? Rollins, anything from you? I can add only
[00:46:59] Rolands:
that it's our future that's main I think and like for me I have a problem with permissions and I don't see like how to resolve that at the moment and yeah I like what AI agents are doing and how they can help and then even when when you give him to think and I related kid yeah kind of it's it's alive uh but but yeah when when you give to think and then sometimes yeah it gives you very very good answers and then and then and something new you can learn from that and yeah i see only good things but like with with all technologies there is a problems like email permissions that's a that's a one big problem i think and uh uh with other things to like like uh when and how you give access uh to do like like that uh that agent can can can do on behalf you and yeah like everything at the end is who is responsible for those uh uh actions and uh at the moment it the one who runs right and and that's a pub human um that's yeah that's that
[00:48:46] Randy:
is a very good point because that's something that's showing up in the news now is who is ultimately responsible for these things? Is it you, the user? Is it the AI company? Who is it?
[00:48:57] Rolands:
Yeah, I like that picture when OpenAI says, our agent or Swarm hacked the planet. Yeah, everything is good. And the same with Anthropic. But when some hacker hacks something, then it's a jail, right? And yeah, that's kind of the interesting world where we live in.
[00:49:23] Randy:
Yeah, there's some very sticky issues there to be figured out. And I think when we start adding email into this mix, especially end user email, when it's allowed to take actions on somebody's behalf, yeah, that picture gets very, very murky about who's doing what and who's responsible for what part of it. So there's no doubt. All right, well, we are at about an hour, so we'll go ahead and conclude this one. Thank you for listening in today, and our next topic will be scheduled for Tuesday, October 20th. As always, all are welcome to join these calls and contribute, and I appreciate your time for listening. Have a good day.
