Welcome and Introductions

Speaker 3 0:01

Welcome to ACP, the Amazon Connect podcast. This is the show that focuses on Amazon Connect and related technologies. I'm your host, Tom Morgan, and I'm joined as usual by my co-host, AWS solution architect and contact center consultant, Alex Baker. We are also joined this week by Bentley Paton, AWS Connect architect at Cloud Interact

Tom Morgan 0:35

It's time for another ACP. We are back in the ACP studio. Alex is with me. Hello, Alex

Alex Baker 0:40

Hi, Tom. How's it going?

Tom Morgan 0:41

Yeah. Good, thank you.

Podcast Timeline Mixup

Tom Morgan 0:43

I need to apologize already. We've only just started. This is I'm so bad at this podcasting thing, but I completely forgot that the alternate reality of podcast world is not the same as the real world. So last week was the one I did by myself and it obviously went to my head because yesterday we recorded an episode which you won't hear until after this one with somebody really interesting from AWS actually. But because they're very interesting, we have to get AWS to be okay with it, so we tend to drop those ones in at like a future date which is why this is the one you're hearing now. But when you do hear that one, you will hear me talk about, me being solo and then you being back. So actually, that was all out of order, and that's my fault. So apologies for that.

Alex Baker 1:23

Maybe you should have kept it quiet and just seen if anyone picked you up on it

Tom Morgan 1:27

I should have. But never mind, it's too late now unless we take it all out in edit, which I'm not sure either of us have got time to do.

NLX Goes GA

Tom Morgan 1:33

Anyway exciting one this week, actually. I'm really excited for this one. Yeah, it's a good one because we talked a couple of weeks ago about AWS buying a company called NLX. We then had a bit of a first look at a bit of a preview, and this month it went GA. And that's great, and we like to talk about things, and we can do that all day long. But we've managed to find somebody in Cloud Interact who's actually been building with it since it was preview and maybe a little bit before that as well. So joining us today is Bentley Paton. He's a Cloud Interact... i'm gonna let you introduce yourself actually, Ben.

Bentley Paton 2:03

Yeah, no worries. Thanks for having me on, guys. Yeah I'm Ben Pitt and I'm a solutions architect at Cloud Interact, focusing on mostly agentic conversational IVRs at the moment

Before CX Designer

Tom Morgan 2:14

Cool. So before all this announcement stuff, what would you be using to build out these flows?

Bentley Paton 2:21

So in the past sort of twelve to 18 months, it's been the, the out-of-the-box Amazon Connect AI agents. That was a big step up from the previous Lex implementation with Bedrock intent. But yeah, so the, the sort of out-of-the-box Amazon Connect AI agents before the agentic CX designer came through

Tom Morgan 2:42

Got it. Yeah. And they've been around a while, right? And then in the last, I don't know, a year maybe, they've added on that agentic AI capability, and it's Yeah, I don't know. It was fine, but it felt like a bit of a bolt-on to an existing system, I think is fair to say, right? And it works, and that's what we used. But,

Bentley Paton 2:59

yeah, I think having-- I think I've been working on AI agents within Amazon Connect for the best part of two years now. I think the original bolt-on of a Lex built-in intent was really quite clunky, and it didn't work too well. So I think that the Amazon Connect AI agents was a big step up. And then I think in the past sort of 12 months, we got the orchestration agents. They, again, was a big improvement. But obviously still bits that were lacking. And I think those are the bits that are hoped to be patched by the NLX acquisition.

Alex Baker 3:31

did that sort of midway through that journey, did it completely phase out Lex? I guess I, I'm probably a bit out of touch with it and which components you still need

Bentley Paton 3:40

No, so you still use Lex, but they introduced a new built-in intent the Amazon Queuing Connect intent, so it was specific to Amazon Connect. Before, I think it was a, a bedrock intent. And then that was all you needed. So you build a Lex bot with a single intent which was the built-in, and then you would build the AI agent on top. And it would just invoke through the sort of the get customer input block. And that was all you needed to do. Now, with the agentic CX designer, that is completely phasing Lex out of the out of the build process. So you don't need to build any Lex bots anymore or do any of that integration

Tom Morgan 4:17

And we first heard about this, Alex, I think earlier this year, and then we had the June preview. And this was an acquisition which is unusual, and I wanna say you, 'cause you've got the history of Amazon Connect cause you, you've worked in it for such a long time. I think this is their first acquisition of technology. Am I right?

Alex Baker 4:35

It's, yeah, I think it's the first that I'm aware of or the first that was publicly heralded as an acquisition. Yeah. So yes, it was quite unusual.

Tom Morgan 4:44

versus building it themselves. And so I think we're all really curious as to what that was gonna look like 'cause, acquiring a technology and bringing it in is a different thing to building exactly what you need, for exactly your requirements. But I th- that's the other thing. For most people there's gonna be some segment of people who previously knew or maybe even used NLX, and so they're gonna have a different journey. I think for most of us though, we probably weren't that familiar with NLX.

What CX Designer Is

Tom Morgan 5:09

I think, yeah, Ben, if you were explaining what the Agentic CX Designer brings to Connect, like in a, I don't know, 30 seconds or something

Bentley Paton 5:18

Yeah. So I think the, the main idea is that it brings your agentic and deterministic capabilities together in a drag and drop UI. So someone who's familiar with building contact flows in Amazon Connect will be familiar with this type of dragging up a box, connecting with something else. But it brings your AI capability in while giving you control over deterministic logic and removes the pain points of having to build Lex bots and do all of that configuration. Now it's it's one block in Amazon Connect, and then it goes out to your application, which is a drag and drop interface. And you just if you want user input, you choose the user input block and you build your slots within there. So it's a much more user-friendly way to build conversational journeys

Tom Morgan 6:05

Okay. I wanna dig into that in a minute and where it sits in the flow block.

Deterministic vs Agentic

Tom Morgan 6:08

But can you just give us an example of something you'd always make deterministic and something that you would happily hand off to AI?

Bentley Paton 6:16

Yes, I think a good example would be payments, for example. So say you've got a, an AI agent that can do a bunch of different self-service capabilities. It can answer basic queries, all that sort of stuff. You want that to be agentic. Doesn't need to be going down, specific strict paths. But then when it comes to making a payment, that's something that you wanna get right every single time. You want the same inputs, you want the same messages being relayed. That's when you would go into a more deterministic flow. The nice thing as well is that you can create a deterministic flow and provide that as a tool for the AI agent part of the conversation so that you know that when the AI agent invokes that tool, it's gonna follow a specific flow.

Alex Baker 7:00

Payments is quite an interesting one.

Integrations and Lambda

Alex Baker 7:02

Do-- can you also then integrate like third-party payment providers as well? Is that kind of within the, the CX designer part, you can go from that deterministic flow off to other things?

Bentley Paton 7:13

Yeah, so you can definitely still do integrations as you would with a contact flow. I think that's something where it feels like maybe it's not quite where it needs to be just yet. Obviously in Amazon Connect, you can drag and drop a, a Lambda function, make a Lambda function call. At the moment, you can't directly integrate with a Lambda function unless you expose a function URL. So it's more APIs via standard REST calls, or you can expose things like MCP gateways. I think the integrations part is a little bit more complex than you'd ideally want it to be. That's where it falls down slightly at the moment, I think

Tom Morgan 7:52

And is that kind of a bit of how it's been crowbarred into Connect, right? From its acquisition.

How It Fits Connect

Tom Morgan 7:57

How does it fit in the Connect flow? 'Cause presumably the call still comes into Connect, like it still comes into a Connect flow and then what happens?

Bentley Paton 8:05

Yeah. Yeah, exact same journey as before. So call comes in, you enter a Amazon Connect contact flow, and now instead of using the get customer input block which you would use for standard IVR or AI agent, you now use the agentic CX block. And as soon as you hit that, you basically go into your application within the CX designer. An application is essentially a group of flows that you build out. So it's a little bit disconnected because that one block in Amazon Connect, and then actually you need to go into the agentic CX designer to see what it does. But it does give you a nice visual representation of exactly what the flow's gonna do and it's a little bit cleaner than if you were to try and build that out into a contact flow

Tom Morgan 8:50

Yeah. Okay. That makes sense. So it's almost like, it's almost like you're going into another flow except the flow is owned by almost a separate process, a separate application almost that is the authentic CX Designer.

Bentley Paton 9:01

Yeah, exactly

Tom Morgan 9:02

Okay. And so we talked about backend systems being a bit clunky because of presumably-- because now you can do it in the CX Designer, which is presumably how they did it in NLX land before the, you know, before it got integrated, but also you could do it in the call flow. So now you've got two places, so that's feels a bit weird.

IDV and Data Capture

Tom Morgan 9:22

What about stuff like IDV, like identity verification? Obviously that's just again, it's a subset of like backend system checks and stuff, but would you still do that in Connect or would you do it in authentic CX Designer? Does it not matter? How do you-

Bentley Paton 9:35

th- I think ideally if you're building something new today, so you're building a new IDMV process, you're building a new IVR, I think the right thing to do is to start looking at building that into the agentic CX designer because it's clearly the strategic option. It makes capturing data very easily because, let's say we want user's date of birth, you just drag a user input block, add a slot for date of birth, which has a, a date option, and then you place a message, "Can you provide me with your date of birth?" That's really simple. Or you can use an agentic journey block, which you either give that slot to the AI agent capability, and then it can capture the information for you a little bit more fluidly. So I think, yeah, IDMV makes sense to do that within the agentic CX designer. The, the part that's just a little bit more fiddly is the integration to actually fetch that customer's data. So I think now the easiest way to do it is to build an API gateway with your Lambdas behind and then call that via REST. So it's not a major issue. It's just slightly different process to you're used to with just deploying a Lambda function and dropping that into the flow

Tom Morgan 10:49

Got it. Okay.

Alex Baker 10:50

Sounds like that will probably be a sort of roadmap thing as they, they integrate it a bit more and it's, gonna be a bit less of a disconnected UI and look and feel to it, hopefully

Bentley Paton 10:59

Yeah, I think so. I imagine that the, the Lambda capability will come in because it feels like it's a fundamental of

Alex Baker 11:05

Yeah

Testing and Simulation

Tom Morgan 11:06

I'm gonna end up turning this into our own personal tech catch-up. We were laughing just before you joined about, like, how so-- like, how ridiculously busy it is at the moment, and that we don't get a chance to talk properly. And like a lot of stuff that I do now is pre-sale stuff versus the actual actually building the thing, which is what I know you do a lot of. When I do the pre-sale stuff, I often lean quite a lot on some of the the test and validation features that come with the Agentic CX Designer, a lot of the, the simulations and the data you get through testing things. And that's great. I think it's great. That's what I tell people. But you're doing it. You're using it. Is it great? Do they work? Do-- Are they actually helpful?

Bentley Paton 11:43

Yeah I think it is really helpful. It's definitely something that's been missing for a long time. And it's really useful when you're building. You, in the agent 6 designer you can click test at any point. You can test individual flows. You can test the entire flow. You're able to inject what are called context variables in the designer, which would essentially be contact attributes in Amazon Connect. So you can inject those at different parts of the flow that you wanna test. Makes things really easy to just test on the fly. You can drag a, a new box in, click save, and then test on the side. And then if you've got a test case that you like, it's worked, and you think you'll probably need to repeat that more and more you can save that as an automated test. And that way you don't have to keep, typing or copy and pasting responses. You just go into your saved tests and click replay, and it will just step through, and it will tell you exactly which nodes were hit whether it passed based on when you saved it originally. And you can set that to run when you when you build your application. You can essentially have an automated test suite that says, "Next time you build, have you broken anything? Do all of the tests still pass?" So I think it's really helpful. The only thing I would say is that in my testing or building something for production you do still really need to test with voice if that's the intention of the, the application. Because voice

Tom Morgan 13:11

so all of the, sorry all of the, yeah, all of the testing stuff, all of the simulations, they're all chat-based. Is that right?

Bentley Paton 13:16

Currently, yes. Yeah. I'm hoping that maybe the automated voice will come in. I think that would be really useful. But yeah, at the moment it's chat

Tom Morgan 13:24

Got it

Debugging and Analytics

Alex Baker 13:25

Do you get to see each step of the way maybe why the agent made a certain decision or used a certain tool, that kind of thing? How good are the analytics within it?

Bentley Paton 13:36

Yeah. So when you're doing testing with the chat functionality you have a debug option and that will show you you can click through the chat. So say it's a quite a long conversation, you don't have to scroll through all of the, the different bits. You can click on a certain message and it will open up, the, what happened around that message. And then you've got individual blocks that you can click on. It says, agentic journey started decision made, that sort of thing. And you can click through and it's quite granular. So some of it's not too relevant, you're not that interested in it, but you can usually find out why something's happened. It's quite descriptive. So it's a little bit less abstract than some- sometimes things can be in Amazon Connect. And that can be really helpful. The only thing I would say again on that is that once you're doing live calls I don't think the analytics are as good. In the CloudWatch logs currently you just see an entry into the agentic CX designer. You don't see any of the steps from within there. I'm not 100% sure maybe I'm missing something, but I don't think you can currently see that in the designer itself either. So I think that that sort of audit trail of live calls is missing

Workspaces and Deployments

Bentley Paton 14:49

currently

Tom Morgan 14:49

and you said earlier when you were making changes, you'd publish the application, deploy the application. Is that separate from a call flow? So that whole process happens. Do you have to update the call flow as well, or does that just stay like the call flow is, it's just got that block in it, and then everything else in that block is totally separate?

Bentley Paton 15:07

Yeah, so within Designer you have you, you build a workspace. So a workspace might be for a business area. And then you have an appli-applications, and essentially an application is your entry and all of the logic that you want within that journey. And then within an application you have flows. So you could have multiple applications in one workspace or for one business area. And when you finish dragging and dropping, building your journey, you build that application. You have the option to deploy a development version. So you essentially have two aliases. You can deploy to development, or you can deploy to production. So that's all done within the CX Designer. then when you use it in the contact flow, it's a single block. You choose from a dropdown within that block what what workspace to use, what application, and then which alias. So it's quite granular. It allows you to do development and production deployment within the same environment. And it means that, yeah, you can update in CX Designer, and you don't have to touch your contact flow.

Tom Morgan 16:15

It, I don't know. Yeah it's one of those ar- it's one of those things that it was... I saw it in preview, and I hoped they might take away some of those layers as it went in, 'cause it feels like there's more than there's needed now because of its history of where it's come from. NLX wasn't just for Connect. You could use NLX in lots of different platforms, contact center platforms. So that whole concept made sense as a third-party tool that you would, you'd need, you need all those layers for what you're doing. I think when it comes into Connect, you wanna be able to jump from a contact flow straight into the, I guess probably all the way down to the flow of the CX designer really unless I'm missing something

Bentley Paton 16:51

Yeah, I think, yeah, at the moment it's a little bit separate. I imagine that over time they'll become a little bit more intertwined. And I really hope that they take the best elements of the CX Designer interface and the Contact Flow Designer and merge them and create more of a unified journey. 'Cause there's some things that people who build contact flows will know are really painful with the lines crossing and boxes interrupting lines. And then the agentic CX Designer it's a bit more fluid, but you've got a lot more lines crossing and less ability to get around that. So I really hope that they clean that up and bring them both together. But that's just something that I'm really particular about

IaC and Business Users

Alex Baker 17:33

I was gonna jump in and ask if there's any support for deploying via infrastructure as code or doing things programmatically. It sounds like it's geared towards and I get the impression that some of the intent was to try and make it a bit more accessible to the business user. Does that mean it's taken away from the sort of the developer community?

Bentley Paton 17:53

So I'm not 100% sure on the infrastructures code support just yet. You can export your flows as JSON files. So I'm sure that support will come if it's not already there. You've got the JSON files, and they seem to be fairly easy to read. So you can definitely get those and analyze those and pass them through different environments. The one thing that's a little bit painful at the moment, I don't know whether it's just user error or it's not there yet, but things like your data requests and context variables and slots, I don't think there's kind of an export and import functionality for those yet. So even though you can move your flows between environments you still have to manually go in and create all of those bits.

Alex Baker 18:41

What level of pain might that cause if you were migrating between environments? Are there loads to go in and do manually, or is it fairly trivial?

Bentley Paton 18:49

It, it depends on the complexity of the application. Depends how many context variables, how many data requests you've got in there. It does introduce, the ability for some, some human error when migrating between environments. That's a little bit annoying. But it shouldn't be... it's not, it's not days of work. It's maybe a couple of hours max just to make sure that you've got everything right. In terms of business users being able to use it and does that take away from developers I think it's a really similar case to contact flows. So you think that with a drag and drop contact flow interface that, anyone can pick it up and build an IVR and do what they need to do. I think that's, absolutely the case for simple IVRs and I think the case for simple journeys in the CX Designer. But I think that the real difference comes from having experience with conversational design. And that's where you probably, for a complex application or self-service, you probably still want a partner to build out the bulk of the application and then you can hand over to the business control over at this part it says this. You can change that or make minor adjustments." But I think yeah, I think you still need to understand conversational design. There's so many, fallback paths and when it goes to here, it needs to redirect back to this place. And that's all stuff that, you only get from experience building these things. But in terms of, proof of concept or simple IVR I think it's a game changer for that sort of thing. It's so quick to do it. No Lex bots needed. The sort of speed to prove things is massively accelerated

Tom Morgan 20:30

Are you finding from doing this with customers, are you finding that the business users are a bit more engaged in those kind of flows? Are they a bit more willing maybe to make changes over time or suggest things? Because it's-- It feels a bit like you're looking at the same flow in a little bit more detail, so you can see the moving parts a bit better.

Bentley Paton 20:50

Yeah, I think certainly with some customers they want to be able to do that. They want that control. I think it gives them more confidence where you've got more ability to have deterministic logic. People are wary about allowing AI agents to do the full self-service capability. And I think this gives them that confidence. And I think that yes the fact that they can visually see, when this happens, it does this that's getting them a, a lot more engaged, whereas before it's a bit more abstract. If anyone's got experience building complex Lex bots a lot of that the, the logic is within Fulfillment Lambda, which is essentially a black box to a business user. Very difficult for them to make changes to those. And now, they've got a visual of what it is. You don't, essentially it's a living document. You don't even need to break down in, a big document this is what happens when this happens, because they can see exactly what it is. And I think that's where, conversation design comes in. If you design the flow to be easy to read, then it's essentially documenting its- itself. And I think that's really valuable

Speed and Iteration

Tom Morgan 21:56

And is that true for sort of build and development speed as well? 'Cause people have been waving around some fairly ridiculous arguments like weeks, not months. But like there's all the other-- The stuff you just said all still needs to happen. But is it faster building out, this way than before or not really?

Bentley Paton 22:13

th- I think it speeds things up to an extent. And I think that, again, comes down to it makes proof of concept extremely easy. And that's amazing, but it's also dangerous 'cause, you build something for a customer in a few days to showcase what it can do, and then, when it comes to being a conversation about how to make the production ready it's not accelerated to the point where, you should be dropping things into production within a week. I think you still have a lot of things to consider, plenty of testing to do. And I don't think that is completely eliminated just by having a, a drag and drop interface to do it

Alex Baker 22:52

Whenever we're in the, the pre-sales process that Tom mentioned earlier, if we're talking about agentic AI or conversational AI, we would always tend to build in more of a, a lengthy sort of hypercare or iteration process afterwards. And I guess that's probably still the case by the sound of it. You wanna get it into production, test it thoroughly, get some live data going through it, but then you're probably gonna want to make some changes to it, perhaps more so than, j- just a, a standard contact center environment.

Bentley Paton 23:19

Yeah, absolutely. I think that's really key. And often, especially when it comes to conversational IVR for general inquiries and routing you don't know off the bat exactly what the success criteria is. You wanna get something into production that can answer questions that you know will be asked and then you wanna analyze the data to see exactly what's being asked and not answered, and then you need to iterate upon it. Different if you're just doing, a, a standard self-service journey where you know exactly what it's gonna do. But yeah, the I think 100% getting something small into production is really valuable rather than going for, massive development over a long period of time, putting it in and then finding out that actually that's not what people are calling about.

Tom Morgan 24:06

And but for new projects, you think you should start here first by default now? Like this is the kind of direction of travel for things?

Bentley Paton 24:14

Yeah, absolutely. I think that the strategic idea is that now all Conversational IVRs should be built in the CX designer. And I imagine that will be the steer from AWS

Tom Morgan 24:26

Make sense.

Pricing and Knowledge

Tom Morgan 24:27

And what about pricing? 'cause they were kinda quiet on that during the preview, and obviously NLX was a third-party tool that got acquired but it's now been rolled into Connect Customer. Do you know anything about the pricing for it?

Bentley Paton 24:39

Yeah, so I believe at the moment it's fully included in Amazon Connect customer as one of the AI capabilities. I'm not sure whether that will be the case forever. There is the option within the designer to choose from different models and also increase or reduce the token output for specific tasks. So based on that, that suggests that maybe token usage will come into consideration at some point. But as far as I'm aware, currently it's not an additional charge. I don't know if you know anything else about that, Alex.

Alex Baker 25:16

N-no. So I'm just looking there's not too many immediately obvious details on the pricing page. One I, I was gonna mention is it does call out that if you're using integrations like Agent Core or Bedrock knowledge bases, and I was just gonna ask about that. Is that still a thing? Is it still something you need to do? But the, the reason I ask is 'cause they're called out as separate costs, of course, 'cause they're different services.

Bentley Paton 25:40

So the CX Designer has its own knowledge management or knowledge base area that is currently separate. I think that most likely it will be brought together at some point so that you can use your existing knowledge bases, whether that's S3 or through one of the connectors. But at the moment it is separate.

Alex Baker 26:02

OK

Bentley Paton 26:03

and Agent Core, again, so with the integrations what I was hoping was that my existing Agent Core gateways would be exposed within the CX Designer, so they'd be easy to use. That's currently not the case. And I've not had any luck connecting to any of my existing Agent Core gateways so far. What I've done is migrated to API gateway and using data requests rather than MCP tools.

Alex Baker 26:30

Okay, cool.

Live Sync Preview

Alex Baker 26:31

One other thing just to call out that I've just noticed actually I don't think we've mentioned it yet a cool feature called Live Sync. Again, the reason that I call it out is to do with billing. So it says something that I found says it's priced at an additional one cent per minute of usage on top of the applicable channel rate. But real, real quick, have you played much with Live Sync, Ben? 'Cause I've seen a demo of it. It looks great. Not sure if it's

Bentley Paton 26:56

no, I haven't managed to

Tom Morgan 26:58

GA as well, I think, hasn't it? If it even is GA. I think it is, and that's where the pricing came from, right? But only just

Alex Baker 27:05

Yeah, possibly. Th-th-this is for anyone that hasn't seen it's where you're synchronizing a voice interaction with the agentic AI with something going on a website or an app. So you, you-- the, the two sort of parts of the interaction are in sync with one another, which looks like it would be a really cool user experience.

Tom Morgan 27:24

It's g- it's gonna be a whole episode once we get our hands on it, I think. Get some... 'Cause it's all very well talking about it, but I think Alex, I know you think the same we like to have tried it in the real world with customers before we bring it here and talk about it, and that's that's why it was so invaluable having you on as well, Ben. Because y- you, of the three of us, you're the one with the real world hands-on customer experience with doing the agentic CX designer today. So that's why we wanted to get you in. So that's yeah, that's been really useful.

Alex Baker 27:51

Definitely. Yeah. Thanks, Ben. Really useful stuff

Bentley Paton 27:53

no problem. Thanks for having me

Wrap Up and Thanks

Tom Morgan 27:55

Cool. We could, and in fact, we are gonna keep on talking about this I think because there's so many I've been trying to bring this episode to an end for the last 15 minutes. But we keep, there's just so many other things to talk about. But yes, it is time to bring this episode to an end. Thank you very much, Ben. Thank you, Alex, and thank you all for listening. Be sure to subscribe in your favorite podcast player. That way you won't miss the next episode. Whilst you're there, we'd love it if you would rate and review us. If you have colleagues that you think would benefit from this content, please let them know. To find out more about how Cloud Interact can help you on your contact center journey, visit cloudinteract.io. We're wrapping this call up now, and we'll connect with you next time.