Tom Arbuthnot: Hi, and welcome back to the Teams Insider Podcast special one this week. This is a live recording from Comms vNext I grabbed my friends, Kevin Kieller and Craig Durr, and we had a conversation about what's going on in the market, UCaaS, and particularly AI, and the different AI options across the various UCaaS platforms and where Copilot and teams fits in that competitive space.

Really interesting conversation with Craig and Kevin. Thanks to them for taking the time. Many thanks to Eric who set up the, podcast set up. . And thanks to SCB Global who sponsored the podcast at Comms vNext, hope you enjoy the show. 

Hey everybody. Welcome. It is the first live podcast from Coms vNext. Really excited thanks to Eric for setting up this amazing rig. Hopefully we're coming through loud and clear. Many thanks to SCB Global who are the sponsor of the podcast as well. Really appreciate them helping this setup come together.

I've got two awesome guests and this will be going out live for Comms vNext and also on demand on the Teams Insider Podcast as well. And Kevin, do you wanna start off by introducing yourself? 

Kevin Kieller: Sure. Thanks for having me, Tom. And yeah, definitely thanks to to Eric as well. This is a super impressive setup.

So I'm Kevin Keeler. I am a co-founder of NAWC, half Analyst, half Consultant. I. And I also lead the BC Strategies group, which is a group of independent analysts and consultants. So I'm excited to be here. This is my second year at Comms vNext, and a great community event. Great. Great to be here.

Craig Durr: Awesome. And Craig? Hey everyone. Craig Durr. I am the Chief Analyst and founder of the Collab Collective. It is also an analyst firm focused on workplace collaboration and communication. 

Tom Arbuthnot: Awesome. We are live at coms vNext, we've just come off the back of the keynote. And unsurprisingly there was one from Microsoft CVP there, and it was a lot of AI, right?

Big surprise. AI is the talk of the show. If you're not at coms vNext, you're definitely missing out. It's gonna be amazing. Couple of days. Awesome show. Really love it. And it's a team show, but actually it's expanding into a. Teams and AI and modern work because that's the way the world's going, right?

We're not just talking about UCaaS anymore or I feel like we're not, we're talking about the wider conversation. And on that note, Kevin, you are doing a session later on comparing the different approaches that Big UCaaS are taking on AI. Maybe you can give us a little bit of insight as to what's happening.

'cause obviously I mostly see the Microsoft world, but it'd be interesting to hear your perspective on what WebEx are doing, what Zoom are doing. 

Kevin Kieller: Yeah, no, that's, when I pitched this session, as you say, this is a Microsoft focused event, but, one of my beliefs is that it's great to look at what the other vendors are doing.

And so certainly I'm looking in this session at, okay, so Microsoft Copilot, of course but, zoom, AI companion, WebEx, AI assistant Google Gemini for the, workspace and really, looking at. What are some of the experiments that the vendors are doing? Because this is certainly a time, we heard from Juan in the keynote, this is a time to experiment. I think that, I often say, AI is going to determine the winners and the losers don't be a loser. So experiment with this and, be aware of some of the capabilities and try things out using.

Some of the different, models. Specifically I think, zoom is doing a great job of keeping things simple. One of the things, I think Elia gave the keynote that Enterprise Connect and he said he's focused on the four Ss. He said security, simplicity. And Smart was smart.

Yeah. Yeah. Which was the ai. Yeah. And I said, I'll give you two of those three S's because really Microsoft is, security sophistication. Because, when Microsoft solves something they'll solve it not one way, but they'll solve it five different ways. Yeah. And I, we see this in all the different co-pilots that kind of surface in different ways and have different contexts and different scenarios.

So one of the things I'd say look to Zoom is they're always focused on simplicity. And they've brought their AI companion, you can tie it into Outlook and Exchange if you're using that. So it can have that context. They've, they just released zoom tasks, so it'll automatically pick up.

It's kinda like Microsoft to do, but, out of a meeting it automatically fills it in. So they're doing some of the. The automatic stuff. And I think that's interesting to look at, right? So I'm gonna highlight some of those examples. WebEx is with the AI assistant I'll say less capabilities.

Now they're focused more on, they always talk about experience and part of their experience. They do constrain what you can ask their AI assistant, it doesn't show up in as many spaces. And they don't really have some of the document handling that Zoom or Gemini and Google Workspace will have.

But so WebEx is is on that experience, but there's some lessons to be learned there. And then, with Gemini, and I know you, Craig, you came back from the Google event. They're doing a great job with the models like in, in 

Tom Arbuthnot: some insane, I, the 2.5 Pro stuff is like chart topping at the moment.

I feel like Google are doing all the hard work and they're not landing the product conversation at the end. Yeah. 

Craig Durr: I would agree with that as well too. One of the interesting ideas that came out of this past week when I was talking to some of the people at Google is we're at this point of transition.

So that's why I'm really excited to seeing what you're talking about where these AI implementations have been tools and now they're going to a view of being what I call a teammate, where they're starting to work, a little bit alongside. And to your point, this is where I think Google has leaned in and said, we can help it make, be a teammate, but I need to see it.

I wanna see it come together. And frankly, this is where I am in the 2025 timeframe is. Why waiting and to see how this transitions to be more of a coworker with myself, right? And who can help land that. And I agree with you. I think Zoom is doing a good job of that in the interface. I can work with it.

It's persistent. Yes, I can bring it across different workloads and different contexts. And, but I think I'm looking forward to this. I think where you guys 

Tom Arbuthnot: think that the. People are gonna make a platform decision based on the ability of the ai or do you think they've already made a platform decision and they'll go with the AI that aligns to their platform?

Because it feels like the AI competition is super hot at the moment. I'm not seeing many customers flip based on, Google is twice as good as Microsoft or Microsoft is twice as good. 

Kevin Kieller: But here's the thing that I try to, make the point of, is I think there's been a lot of model competition and I think that truthfully, with your assistant.

It's less about the model and it's more about the context. Yes. So retrieval, augmented generation, it's if you have an assistant, if you have a human assistant, they need to know about your business. They need to attend your meetings. And and that's where, it's Microsoft's race to lose, especially for people who have so much of their business content in SharePoint, in Exchange, in Outlook.

And so I think what we're seeing is, and we're gonna increasingly see is. All the other vendors having stories about how they integrate. If you have SharePoint documents, we'll find a way to integrate. And I know both WebEx and Zoom are leaning on Glean, for example, as a, as an interface to be able to pull context because it is the context that I think is gonna make it more important.

And two years ago, I think at Enterprise Connect, I did say that. This AI piece is really going to be a competitive differentiation for the UCaaS platforms. 'cause let's face it, for the most part, the features are commodity. They all do a fantastic job. Yeah. Yeah. 

Tom Arbuthnot: I think that's fair. They've the basics.

You can pick anyone and they all do a meeting, they'll all record it. They all have decent video. They all have chat and context, yeah. I think it's similar, 

Craig Durr: I think where it's gonna be a battlefield. Is gonna be what I call contextual flow, or it's the idea of the cognitive load. So to your point, I think we have a lot of these great tools that saying, yeah, I can do these great summaries in Doc, I can take these meeting trans transcripts and create something from it.

But right now our individual workflows are still cut and paste. So from that cut and paste idea, we still have to pick up things, move it out and so it's a cognitive load to what we're doing here. And so what I'm looking for is finding those tools that are making it more seamless. So 

Tom Arbuthnot: working and that cut and pasting is important, right?

Because Yeah. Techies will lean forward and take the pain. They'll be like, I know this thing on open AI or ChatGPT is slightly better than this thing over here. Best to breed. Best to breed but like real users, like you're supposed to be making it easier for them, not giving them a job to do.

Craig Durr: So those are the two areas I think we're looking about. I don't think people are flipping as of yet, but I think the opportunity to win is where Microsoft or Google or even Zoom, can keep you within that same thought process without having to move between applications and what have you. The Zoom model, the federated AI model is really interesting 'cause it behind the background, it's using both perplexity, it's using anthropic, it's using open AI

I don't get to choose from each time, but it's doing that. So I think that's where the opportunity is now to offset that though, I. I think there's a movement also to create a better agent to agent model. Last week there was an agent to agent protocol that was introduced by Google. 

Tom Arbuthnot: Yeah. The MCP staff.

Really? Yeah. 

Craig Durr: Yeah. MCP is really great. It's a wild west area, though. We can talk about that moment. But even above MCP, they were trying to create a protocol. Just agent Oh I haven't 

Tom Arbuthnot: dug into that. Announced by, saw the, announced they had a whole load of logos of Salesforce and others who are playing the game, right?

Yeah. 

Craig Durr: Again we're still learning a lot. If 2024 was the experimentation, learning how to do it, I think 2025 is how do I make it into a teammate and a workflow that's 

Kevin Kieller: And there's lots of work to be done there because, today agents show up really well in demos, but they don't scale to production.

And, even I love what Microsoft's done with the SharePoint agent, it's really easy to create an agent based on a document site. But you still can, and you can paste the link into a teams group chat or a teams channel, but you still can only have just like one agent. So if you created like an HR agent and then when you put another agent and it says, Hey, you can, I can only have one.

That's not how the real world works. And this to your point, the keeping you in the flow, I think that's what we're lacking now. There's great. Examples, but to build trust with end users. Tom, to your point, if you're a techie, you understand how this works. But to build trust with an end user, the context has to be the same.

So just because I'm in a meeting, for example, I'm in a Copilot meeting, I, it's not web grounded. And so if I say, Hey, let's plan a meeting in Toronto, where I'm from for next Friday, and if I wanna say during the meeting. What's the weather gonna be in Toronto next? Next Friday? Copilot won't let you do that.

Tom Arbuthnot: And that's fine as a techie because you understand. But for real people, particularly with Microsoft's branding of Copilot the inference from Microsoft is, it's one thing and it's not, it's 50 things with Brad but 

Craig Durr: hasn't that updated. So with their 50th anniversary, didn't they up announce that Copilot now is gonna.

Implement memory so that Yeah. 

Tom Arbuthnot: On the consumer side at the moment. Oh, okay. So again, this is the, this is where the branding is tough. They have Microsoft Copilot. Yeah. And it's somewhat deliberate. I think this branding of we have all the cutting edge stuff happens in Consumer first because that's faster and flow free and moving, and then they backwards it to the enterprise.

So we will get it on Enterprise. But yeah it's different. And this is again, mixing and matching. I hear this in my enterprise customers. They'll see a YouTube video and be like, oh, I saw Copilot could do this. Oh no, that's. This Copilot, not that Copilot. 

Kevin Kieller: And that's where, like for example, zoom is grounded the same in all the contacts.

So the AI companion sidebar, if you're in a meeting and you say, what's the weather gonna be in Toronto next Friday? You get right. So I think that to regular end users, that's really important. That consistency, like already AI is probabilistic, which people don't understand what that means. It means you're gonna get different answers, different times.

And in my session, I show some examples that are both entertaining, hopefully, but also somewhat scary because it's great when you can tell it's wrong. It's not so great when you're relying on a complex answer like a mathematical thing. And sometimes it's right, and sometimes it's wrong, but it always looks.

Really impressive. So it's 

Craig Durr: easy to be misled. You know why I'm laughing about that is as an industry, when we first introduced this idea, we said, you know what? That's happening. It's just the AI hallucinating. We gave it a nice pretty name. Yeah. It's hallucinating. It's okay. No, it's wrong. It's wrong.

It's telling me to, take the red pill instead of the blue pill, and it could be a, yeah. Totally wrong. It could be totally the wrong answer there, but it's hallucinating. We'll let it go with that. 

Tom Arbuthnot: Yeah. Give it a friendly name. 

Kevin Kieller: Yeah. But this, overall, I think the thing is this is exciting times. There's tremendous progress happening. Week it, it is almost weekly. Yeah. And certainly within six months. The challenge, I presented this at I. Enterprise Connect. I talked about this, but then just between, a month later there was a bunch of updates I needed to do to make sure, yeah.

That I'm being accurate. And even then, something could have come out today, and so it's very difficult to keep up. But, once again, experimenting is important because it keeps getting better. And even use cases that maybe it did a poor job on. You know what, maybe next month, maybe towards the end of this year, it's gonna make a phenomenal difference if you can leverage this effectively.

And that's really outta my session. I hope people take away, regardless of the AI assistant, there's some tips and tricks that I'm gonna share to say, hey. These things are gonna make it better and they're also gonna make it safer, for understanding when it might not give you the right answer.

Tom Arbuthnot: I, I think the other thing that I'd like to see landed in customers is that they get, with Zoom obviously, they get the AI included with Microsoft, they get the M three five Copilot chat, which is not the automatically Work grounded, right? But is included in the license, is enterprise secure. So it's like ChatGPT, but.

Enterprise for them, and that's all including the license. And I still work with so many customers that haven't pushed that out to their users and given them a valid option that isn't a consumer tool for ai. And you can do a lot of stuff in that. You can upload documents, you can play with 'em. It's not as automatic as the full $30 license, but I feel like lots of enterprises are not un unlocking even the basic value yet of what they get in the box.

Kevin Kieller: This has been the thing with, with the Microsoft 365 suite of, there's, people didn't even know the tools that exist there, right? Yeah. And more so now with, the AI features the thing that I worry about is with different licenses, if you've got teams premium, you get certain things, then you get Copilot.

I think about in large enterprises, supporting the end user base. It's very difficult when you don't know if they have this feature. Yeah. And I think, while Zoom AI companion comes at no additional cost, all of the vendors are trying to find a way to monetize, because both the training, but also the inference cost.

Tom Arbuthnot: Yeah. There's real cost there. There's 

Kevin Kieller: real cost. Especially these deep research models like the Nvidia keynote they talked about. I'm so 

Tom Arbuthnot: excited about that coming to it's super exciting, 

Kevin Kieller: but it's jensen Wong was talking about how it could be a hundred times more expensive than inferencing because you might try it 10 different ways 

Tom Arbuthnot: And Microsoft put an early stake in the ground there saying that it won't be at any additional cost.

So open AI. ChatGPT, you get I think you get 10 shots as a $20 license. And you get 120 as a 200 license. Now Microsoft's saying they're gonna put all that reasoning, deep research stuff in the Copilot license for no additional cost. It'd be interesting to see how they weather that additional load.

Because 

Kevin Kieller: they're already charging. But even if you look at Zoom, like Zoom's coming out with a new $12 per user per month license, that's zoom AI companion personalization. Oh, interesting. So they're trying to monetize that. Yeah. You got RingCentral coming out with their a.

Air AI receptionist. Yeah. Which is usage based, right? Which is some of the Microsoft agents as well. I like usage based. I'm all about, no, I don't like it. I don't because. 

Tom Arbuthnot: No, no win, no fee. If I'm using it, I'm paying for it. If I'm not. No. You know what's 

Craig Durr: even different is have you seen what Zendesk does?

I know to get off of that, they're actually doing it based upon results. If you're not happy with the results, 

Tom Arbuthnot: so does intercom. Intercom does a if the agent answer the person's question without the human getting involved, you pay us a buck or 50 cents or whatever. And if the human has to get involved, no fee because we.

The agent didn't do the job that human would've done. 

Kevin Kieller: Yeah. The reason I don't like it is it discourages AI adoption, and one of the problems now is you see this with Copilot and other AI assistant, people try it out and they do things. Write a poem, write this email as a poem, and then the. If they don't integrate it into their day-to-day workflow after a month or two months or whatever, that drops off.

Yeah. And so as an organization, when you have what I call an unbounded liability, like some Azure service cost thing, you can be surprised. Like that's 

Tom Arbuthnot: Yeah. That is, you're entirely fair to counter and make that enterprises hate variable cost. 

Kevin Kieller: And that used to be, you'd buy the E5.

Because it was, it was all Yeah, predictable. Everything. The 

Tom Arbuthnot: everything license. Yeah, exactly. Yeah. 

Kevin Kieller: And even like power apps, like Visual Basic, I used to love and create all kinds of things. Power apps became, it was gonna cost you and and so that usage based, I understand why vendors do it, but at the same time, you then have this unbounded liability if you really create something that people use.

Too often 

Craig Durr: So let me take a step back and then reframe this. Are you two saying that you think there's a great Microsoft story but it. It's getting it tripping over its own licensing and getting it implemented. Even you're talking consumer to the 

Tom Arbuthnot: three different levels here. Like marketing and messaging is the tough thing.

I think like the licensing, you've got the $30 and you've got the free. Yeah. And you've got the consumer. I think if you survey enterprises now, none is probably harsh. Very few could accurately articulate what goes where and why and what would be the best use case. That's, and it's tough like we.

I know Kevin does loads on this. Like we spend our time trying to follow this and decode it, and it's hard. It is when it's your, like you're trying to do it with customers. So if you're and enterprises get Microsoft talking to 'em directly and they get some support, you go down a tier to SMB mid market, they're relying on YouTube and newsletters to understand it now, 

Craig Durr: but there's promise coming from conversations Alyssa had that there's gonna be a simplification in the licensing.

Have you heard anything similar or do you see that possibility on the horizon? 

Tom Arbuthnot: I can't see it changing that drastically other than the Copilot chat being free. Okay. But the I, I think it's funny to me that is free and quite powerful and it's not being actively used as much as I thought it would be.

'cause I think because everybody thinks Copilot is $30 as a user a month, and that they stop the conversation there. What do you think, Evan? 

Kevin Kieller: Yeah I do think Microsoft is tripping over themselves. I think it's their race to lose because there's so much of your business information and, Satya, since when he took over has built trust, like PE organizations, trust, Microsoft with their data.

I think this is a challenge that Google has, right? Yeah. Even though. I'm sure they play by the exact same rules. They don't use your data to train their model. And, all of the vendors say the same thing. They, what I love is the vendors say it like they're unique, but they're all doing the same thing.

Zoom isn't training the model with your data AI assistant but the Microsoft complexity, both from the marketing standpoint and I also think that the, this agile approach, like changing the user interface in like even teams. Even making it better, but when you keep changing it and making it more configurable.

Yeah, so the licensing is configurable. The UI is configurable to such an extent that when I'm trying to support you. I don't even know what options you have. 

Tom Arbuthnot: Yeah. 

Kevin Kieller: And so you've got the teams' premium licensing, which I have a huge problem with 'cause it's an eclectic mix of random things.

Tom Arbuthnot: That's my session tomorrow. I'm going through all the teams' licensing and that's gonna be three hours. I I've really, I took a long time to put together, but I think I've broken it down into English now. But to your point, it's like teams is not included in the enterprise bundles anymore.

So you add teams enterprise, then if you wanna add team's phone, then you wanna add team's premium, and then you're still not there if you want the premium experience, which is adding Copilot on top, right? So like it gets complex 

Craig Durr: quickly. And then some of the features are available in teams premium and some overlap also, if you have the Copilot licensing as well too.

So it's, I think where they're where it's their battle to lose is this, and I'm gonna give it based upon personal experience with your idea about experimenting with the AI tools is right. You and I have time to, to dive deep into anthropic, dive deep into this, and it's a matter of actually committing time and then not having that be a lost cost going forward.

So what I'm saying is I might be working with OpenAI. I've trained it, I know it knows my documents. I know how to use it. I think in the Microsoft ecosystem right now where I don't have feel like I might be a consistent, does that Copilot there in teams know what I did before in my chat and my sidebar, does it have context to what I had with within that Excel file or not?

Because I don't know. Yeah, and the answer 

Tom Arbuthnot: may be depends sometimes 

Craig Durr: because I don't know which one I'm hitting. Am I hitting the consumer one? Maybe my enterprise employee, the other license or not as well too. I think that's a little bit of the hesitation as the non-technical user would have. It's interesting you say 

Tom Arbuthnot: that something sparks in my head, which is it's not beyond the realms of comprehension, that the Copilot could be trained to intelligently reply to be like, you are asking me about meeting content, but you're in word.

I don't have the ability to reach that yet thing. Interesting. It doesn't do that today. It just tries and fails or doesn't fail. But it could potentially be like, here's the limits of my ability. I can only see chat for the last 90 days. That's what I do. 

Kevin Kieller: But And that's a good point you bring up.

'cause this scares me, which is the limitations aren't well understood. No. So if I say, Hey, summarize all the important info emails from Tom, and it only goes back for a period of time, but we don't know what that period of time is. 

Tom Arbuthnot: And I think we can lay that challenge back at Microsoft very fairly because I've asked this question multiple times and be like, this needs to be in docs.

Like just 'cause if you explain it, we understand the limitations and we can work within those limitations. But if you don't explain it, I'm frustrated that I say find the emails from Kevin and it only finds the most recent ones. 

Craig Durr: That's the sunk cost challenge. That right there is an end user. You have some time and energy you put into trying to understand this.

And have to contextually switch or go back and forth or not get the right answer becomes frustrating. And then you move on to the next tool. And then you say to yourself, I don't mind cutting and pasting because at least I know this one opportunity over here with this one tool is consistent and I can just bring it into my Word doc or my Yep.

Excel spreadsheet. 

Kevin Kieller: And I think that this is what is in track. Like I have an example, it's like sometimes you use AI and it saves you a bunch of time. Sometimes you use AI and it's a waste of time, and then you do it manually. Like image creation is a good example of that, depending on the tool you're using.

Yeah. It's like you can be in PowerPoint and try to do something and then just go half an hour later 'cause you get sucked into that and then you go find the stock photo. Right? 

Tom Arbuthnot: Technologists are like putting ourselves within the same camp for a second. Technologists, analysts, whatever that sunk cost of.

It didn't work. It's I've learned something. I know where the limits are for a business user trying to get their job done, who's got way too much to do and waste their time, right? It's taking time away from them and they're frustrated. So that is really tough. 

Craig Durr: This is what I think is key, bringing it to a humanistic level right now, and this is I just wrote an article and I think it's gonna be published in No Jitter.

We're at a point where AI actually. Becomes a agent of trust within my workflow. If I have the ability to trust that a AI agent, that I'm asking the question and I believe the answer that I think it has the right context, I then start to get this cognitive release. Great, I can trust you to do my work, so I'm gonna move on to the next thing.

Now, if as an individual I relieve and have that element of trust, I probably become more productive with the three of us as a team, right? So now you start trusting me. We have a. A common AI agent, facilitator, we trust it. We now become more productive. We have, again, this cognitive load leads up, right?

And eventually that rolls to where it impacts the organization, but it starts with individual trust of the tool and how you're using it. That the information coming back is correct, that I'm not. Losing time. 'cause I'm thinking, I feel like you've 

Tom Arbuthnot: got the beginning of a scoring matrix there. Yeah.

Where I a hundred percent trust the meeting summary because I've seen it work often enough to be like, yeah, it's 90% the way there, but I trust it's right. Whereas if I ask Copilot, what are my upcoming meetings with Kevin? I trust it about 50% because I see it. Miss things. It's, 

Craig Durr: this is exactly the line of thought I did through this article here.

And so building trust, a new collab collective matrix. Exactly. But that's it. It's really interesting because it, something as very technical as AI ultimately can lead to this cultural understanding of business culture, of trust. And it starts with individual. If I trust what the technology is telling me, then I feel comfortable.

Then Tom trusts me that I'm giving him good information based upon the workflow I'm doing. And then our department is trusted by another department because we're producing this is where AI can be powerful, but we're right now where people are stumbling and we're all hitting these use cases within this Microsoft ecosystem.

I don't think we have that level of trust yet. We're like on the cusp of we want it so bad. It's so powerful. It can pull, if it can really do this Excel worksheet that I believe is spot on and it just saves me 30 minutes. Beautiful. But I'm building up the trust still.

Kevin Kieller: And I think as in the humanistic mode, like there's important lessons for us as humans, in terms of how we use this. For example, I've talked to project managers. They're like, this is great. I don't have to go to my meetings and I can send out the project summary, and I don't, and they don't review it.

That's a real problem. That's not gonna increase trust 'cause it's going to get something wrong. Yeah. For example intelligent recap, which you pay for teams premium. I have a huge issue with this fact. Ooh, you can't edit it. That's a total fail on Microsoft's part. Microsoft Total fail because. Yeah.

It like, and it stays around. Yeah. The 

Tom Arbuthnot: pitch is always have a human in the loop, but you are not letting the human be. There you go. Yeah. And if it's 

Kevin Kieller: wrong and it's gonna stay around, like that's a terrible use of ai. Granted, it says it could be wrong or whatever, but you need to be able to correct it.

Yeah. Whereas like the facilitator agent that's taking notes in real time, that's a great use. 

Tom Arbuthnot: Yeah. 

Kevin Kieller: That's a co-edit. Yeah, that's a co-edit. It's building trust. 'cause if we see something wrong, we can correct it, show your work. Yeah, so I think that us as humans, we have to use some of this stuff.

So please don't send out a meeting summary without reviewing it. 'cause it is on you. You're responsible for that, right? So 

Tom Arbuthnot: Or I'd say the other way to look at it, I completely agree with everything you said, is if you are a hundred percent automating it, then. Everybody that gets that meeting summary should be like, this was AI generated without any human review.

Expect errors like, like we've made a business choice to take the 80%. 

Kevin Kieller: Hey, I think you should review it. It's if you take minutes in a meeting and say 

Tom Arbuthnot: You definitely should, but like the reality is you get smaller organizations is. 80% minutes better than nothing they had before, and they accept the error risk or do they put a human in and get to a hundred percent again?

Craig Durr: That's again, building the trust. And that it's, the point in time is this a mission critical meeting that's taking place and do we believe it's taking place or is it just the three of us updating it once a week? Are weekly? Yeah. Yeah. And going from there. 

Tom Arbuthnot: So relative to the importance of those notes.

Yeah. If it's a multimillion dollar project, maybe the PM should be paying attention. Yeah. Yeah, exactly. Exactly us shooting this shit over UC. Maybe we can accept an error rate. And 

Craig Durr: so this is where the whole transition in this year going to agentic ai, this is where it's gonna be more autonomous, right?

So I want to be able to not only build up that trust, what's taking place with that one to one transactional. Idea taking place with the ai, but now if I'm gonna be talking to the agent and it's gonna be more relationship based, I'm gonna trust it to do a two, three step process. That's where it's becoming.

Tom Arbuthnot: And this is a real technical challenge, right? Because actually if you build a scoped agent to do one thing very well, you can really reduce the hallucination and increase the rag and get a really high level accuracy. Like all this agent does is it does a thing to B thing. But to our conversations today, what we all want is one thing that we can ask multiple things.

We have multiple agents in the chat. We can say, book a meeting in this scenario, and we can say, summarize a meeting in this scenario. So what we want as users is one thing to do it all. And the currently, the smartest technical way to do it is to break it down into. Individual workflow. Oh, yeah. 

Kevin Kieller: And that, that kind of pushes the problem off to the orchestration, right?

'cause then some, the orchestrator has to figure out what agent to invoke. Yeah. So like I create an HR agent, you create an HR agent, you create an, like which one do you invoke, right? And then when the answers don't agree. Which is better. How do you decide, so yeah, you just put 

Tom Arbuthnot: another agent to orchestrate the other HR agents than another agent.

Kevin Kieller: Yeah. So I think, this is, it ends up being challenging, but it's, it's exciting times. I wanna leave people with experiment. Yeah. Because there is definitely use cases that today will improve. Your business, it'll, it'll make you have more enjoyable work. But you've gotta figure out in your particular workflows what those use cases are.

Craig Durr: And I would then, I would build on top of that and say, I think the takeaway I would like people to leave with is think about it in a humanistic trust point of view. Start thinking about this not just as a tool, but as a virtual teammate. Both, either at a transactional level or as it becomes more relationship and doing multi decisions.

That is where I think you should measure your experiments as being successful. It produced a great answer once did it produce that great answer two, three times. Yeah. And do I find it? And then once you start trusting it, I think you can start allowing it to be bigger in scope and growing and how you utilize it.

Awesome. 

Tom Arbuthnot: I hope every listening has enjoyed that kind of story around where we are, where we're going, and some of our thoughts. Like I, I think it's such an exciting space and for UCV and Collab, I feel like it is the right topic to be talking about. Now, thanks again to SCB label for setting up the podcast.

Rick and Eric and the team Craig, Kevin, always a pleasure to have you on the show. Every time I have you on, it's always a good conversation. So getting us all together was great. Just let people know where they can find you if they want to find out more. 

Kevin Kieller: Yeah, so well, LinkedIn is a great place.

Kevin Keeler, I'm sure you'll find me there. 

Craig Durr: Same thing for me, Craig Durr on LinkedIn. You can also go to the website, which is Collab Collective. We have a collection of blog posts and a lot of up-to-date information in the UC space. 

Tom Arbuthnot: Awesome. Thanks everybody. Thanks again and look out for the next show.

We're from Comms vNext.