I Categorized Every AI Tool and Harness So You Don't Have To!

0:00Good evening everybody. Welcome to the

0:0112th episode of our webinar series. It's

0:04hard to believe we've been doing this

0:06now for a year, but uh thanks for

0:08everybody who follows along. Tonight

0:09we're going to do a little bit

0:10different, little [music] bit faster.

0:11It's going to be a little more rapid

0:13fire on things, but this is going to be

0:15our uh kind of our vendor roadmap. So So

0:18think of this is like a vendor roundup

0:20of who's who in AI technology, and we're

0:23going to cover through the uh couple

0:25different layers that you have inside of

0:27the different tech stacks. So uh we're

0:29going to start at the you know the

0:31highest level which is uh where I call

0:33it the the surface level, which is this

0:35where the humans actually interact with

0:37it. You're going to then go to the uh

0:40harness level, which is the construct

0:42that talks to all the different uh

0:44elements, and then we're going to talk

0:46about the model. So we're going to cover

0:47a lot of stuff tonight. Going to have a

0:49bunch of uh slides here, which is going

0:51to be a little different uh than what we

0:53normally do. We got a lot of stuff to uh

0:55cover and look at here. So

0:57Um I'm also going to define a number of

0:59phrases uh like harness I just used uh

1:02for you tonight. So this one should be

1:04pretty instructive and uh give you a lot

1:06of information about where we sit here

1:09at uh the middle of the year in 2026

1:12with AI technology. So kind of a a where

1:15we are and uh where I think we're we're

1:17going and and what is the industry doing

1:20right now and uh how I see it ending up

1:23through the next couple of months. So

1:25um let's talk about the frontier board.

1:28So [snorts] these are the biggest

1:29models. Um this is going to be uh you

1:32know at the top of the heap today as I'm

1:33going to call it is going to be

1:35Anthropic. Right behind it is going to

1:37be ChatGPT, uh which is OpenAI. Behind

1:41that is going to be Google and all of

1:42the Gemini series.

1:44X and Grok are going to come in below

1:47that. Meta uh is still out there.

1:50They're

1:51changing up what they're doing with the

1:53models, but I'm still going to put them

1:54in here at uh this level. Um New Spark

1:57uh is is good, but they also have a lot

2:00of adoption of the Llama models because

2:02of their open source. So, even though

2:03they're frontier that then degrade and

2:06put out some of their own models, I

2:08still have to put them in into some of

2:09the frontier model realm here for Meta.

2:13Deep Seek Deep Seek AI the

2:15before pro

2:17is actually very cost effective. It's a

2:20good model

2:21out of China also Alibaba is

2:25out with the Queen 37. It's Queen 37

2:29Max. It's a very robust. It's very

2:32competitive. So, it is one if you're

2:34looking to save token cost and do a lot

2:37of workload, it's one to consider

2:39putting on to your own hardware or using

2:43a provider that can run this open source

2:45open weighted model for you and let you

2:48save a lot on the tokens. And coming in

2:50here at the end of the frontier board,

2:52I'm going to have to say

2:54Mistral out of I believe they're out

2:56it's out of France is is where they're

2:58out of. But medium 35

3:00they were lagging a bit and if I'd done

3:03this frontier board a month ago, I

3:05probably wouldn't have put them on here.

3:07But

3:09the medium 35 model is

3:12you know definitely a good EU and and

3:15last month we talked a little bit about

3:16data sovereignty

3:17sovereignty and um

3:20you know being that it's a EU model and

3:22it should for anybody who's operating in

3:25the EU and looking to encapsulate your

3:28data and get a really good high

3:31functional model, that's one to to look

3:33at is going to be the Mistral model.

3:36So,

3:37let's talk about you know that that's

3:39really your your level of what model

3:43you're using. And and honestly today the

3:46differences between Chat GPT and Claude,

3:50I used to recommend one for the other.

3:52It's less about that. There are a lot of

3:54parity between all of these. It's now

3:57about speed, token cost, reliability,

4:00and and even those those open source,

4:02open weight models that are coming on

4:04that are

4:06free to use. You have to use your own

4:08hardware or get somebody to to put it on

4:10GPUs for you, but they don't

4:14have a lot of the deficiency that they

4:16used to. So,

4:17those given models are now very robust

4:21and and able to go into a variety of

4:23workflows. So,

4:24um

4:25So, let me give you

4:27here real quick three rules to

4:31to layer one. So, that's that that that

4:33model. So,

4:35what I look at when we're coding our

4:38technology today, um

4:40don't necessarily standardize on one

4:43model. We route. So, we're using like

4:46open router. Perplexity has some

4:49interesting stuff coming out that is

4:51doing the routing for you. So, they're

4:53kind of making Perplexity you want to

4:55look at here over the next couple weeks

4:56and months.

4:58And you know, let's say let's say here

5:00August, September, October. Keep an eye

5:02on on what Perplexity's got going on

5:04because they're giving you an option

5:05where you can use Perplexity and use a

5:09open source model with a very low token

5:11cost. Um

5:14you want to you know, we talked about

5:16you know, route don't

5:18route don't standardize on your your

5:21model route to it.

5:22The frontier model's pricing is got a

5:25difference between you know, in some

5:27cases the the latest Fable 5 from

5:30Anthropic has about a 3x three times

5:34spread on Opus in in some of it. So,

5:38what you want to also do is you know,

5:40you know,

5:41rule number two here,

5:43buy at the tier that the task needs. So,

5:47you don't need to throw everything into

5:48Fable 5 if you're trying to just parse

5:50PDF documents or Word documents,

5:53uh something of that nature. That's

5:55That's not what you want to do. You want

5:57to use the right the right, uh you know,

6:00model level inside of that frontier. So,

6:02you're going to route to the right

6:03model, but you're going to then pick the

6:05model that is best suited for that type

6:08of task. And number three,

6:11um

6:12think about the open-weight models as

6:15well. Not because it's better, just

6:18because it gives you leverage against a

6:203x cost. So, that Quinn model, that, you

6:24know, Alibaba models, um

6:26uh Kimmy, there's a number of them that

6:29are not classed as a frontier model,

6:31they're classed as an open-source model,

6:32open-weight model. Think about in your

6:35long-term plan how you can leverage

6:38those so you're not boxed in and looking

6:40at high cost. Give you a side note, um

6:45token cost has gone really high. There

6:47have been some people that got some free

6:48tokens to go ahead and plug all of your

6:51Slack channel into, uh

6:54uh into a model so that you can have

6:57Slack be summarized and have these

6:59agents write in Slack. I know for us, we

7:02burned through our free allocation uh on

7:04a product called Victor very quickly.

7:07So, I couldn't believe I I It was $500

7:09or $1,000 worth of uh uh usage that they

7:12gave us, and um our development team

7:14blew through that very quickly. It was

7:17being used in different use cases. I'm

7:18not saying it wasn't valuable use cases.

7:21I'm not saying that we didn't get

7:22hundreds of dollars worth of value out

7:23of it, but I will say that that $1,000

7:26or $500, whichever it was, went a lot

7:29quicker than we expected, and um we've

7:32also heard of companies who got the same

7:33type of deal from Claude and Anthropic

7:36directly and burned through $4,000

7:39worth of tokens in a in a week by having

7:42it read and index and put into Oracle

7:45all of the messages that have happened.

7:48I don't know about your team, but we

7:49share documents. You know, when I get a

7:51document headed my way to approve for a

7:53quote or a proposal or something, a lot

7:55of times that's on to the Slack and

7:57Teams channels. So, if you're plugging

7:59in where it is looking at every revision

8:01of this, it's great to be able to go

8:02back and find that, but it is also uh

8:05something that can be quite costly.

8:08So, something keep into in mind. So, the

8:11um

8:12next thing we're going to talk about

8:14is a term that you may not hear about

8:16unless you're watching geeky podcasts

8:18like I do. Um it is called your AI

8:21harness. So, it's a new terminology

8:24being used. We've been using it for

8:26probably the last 6 to 8 months, but I'm

8:29near now hearing it a little bit more in

8:31the mainstream and it's something I want

8:33you to to understand when you hear us

8:35talk about the harness. So, this is the

8:39uh software layer wrapped around the

8:41model that turns your uh request and

8:45reasoning into action. So, uh it could

8:49be straight-up automation. It decides

8:51what the model can do and how the model

8:55is fed information. So, if you got rag,

8:57you know, you're feeding in additional

8:58information into it, or are we taking

9:00just this prompt request in, or are we

9:01taking prompt plus a a source file of

9:04company data and memory? That is all

9:07your harness. So, um

9:10it's one of the things I think you need

9:12to understand exactly how your company

9:15is using its harness and whether you

9:17have put thought into it, or is it

9:20something that uh is is just happening

9:23behind the scenes. If you're just using

9:24chat, you're you're not likely to have

9:27to worry about it as much, but if you

9:28are doing group where you have a whole

9:30bunch of people using unified chat, or

9:33you have company data going into it, you

9:34got digital twins, um

9:36this is where you want to monitor

9:38because when something goes wrong, it's

9:40going to be in the harness a lot of

9:42times. So, uh, you want to have things

9:44in place in your guardrails to catch

9:46what goes wrong, uh, in in that harness

9:49structure.

9:50So, let me give you kind of the

9:53anatomy here. I'll pop a a screen pop up

9:56here of the anatomy of your harness cuz

9:58I think that makes it a little easier to

10:01understand. So, it's, um, it's the

10:03tools. What can you reach? So, um, you

10:05can add some of your own data in here.

10:08It's going to be some of the

10:09off-the-shelf, some of the MCPs. So, are

10:12you going to have it reach out to

10:14something like Jotform, um, is something

10:16that you're can create forms for your

10:18website. You can have Claude in in this

10:21case use the MCP part of your harness to

10:24create those forms on the fly and then

10:26add them to a like a WordPress website,

10:29uh, if you if you want to do that. Um,

10:32the next piece of the harness is going

10:34to be the looping capability. Loops,

10:36there's a lot of YouTube videos out

10:38there about what, uh, people are doing

10:39with loops. Loops are one of those

10:41things you, um, should be using, but

10:43it's also something to keep in mind it

10:45can, uh, extend and and run a lot of,

10:48um,

10:49token costs, uh, very quickly. So, you

10:52want to look and and make sure what

10:54stops it, what starts it, what steps are

10:56in your loops to make sure you have

10:58those under control. Um, your your

11:00context policy going to be how you wrap

11:03it, memory, what is persistent, what are

11:06you keeping, what does it remember now

11:08from 15 minutes ago or 15 hours or 15

11:11days ago. Um, you know, that type of,

11:13uh, activity.

11:15Sandbox is a new term. We've heard about

11:17how a couple of the AI models, uh,

11:20OpenAI and Claude both got out of their

11:22sandbox. Sandboxing is something I, uh,

11:25definitely recommend doing. It sets some

11:27boundaries around. You put it into a

11:29container. You put it in some VMware.

11:30Boy, make sure it's locked down cuz if

11:32the big guys have it escaping from the

11:33sandbox and doing things it shouldn't,

11:36um you and I probably have it happening

11:38as well. So, it's something to keep in

11:39mind. Uh permissions.

11:42So, you want to set a um a gated

11:45control. So, you want to make sure what

11:47are the parameters

11:50of what the harness can do, what it can

11:52access, what it can't. Such as stay out

11:55of the company payroll uh you know

11:57information that might be our

11:59SharePoint. You want to put it into an

12:01organizational unit and and have it only

12:03be able to get to it. Then you use um

12:05you know 0365 DLP to make sure that it's

12:08not trying to get into it. That's you

12:10know a very common thing. Um but uh at

12:12the at the core, observability is a big

12:15thing. So, uh observability, we do a lot

12:18of logging. We like putting things

12:20through a seam SIEM. Um this is

12:22immutable log that is like elastic log

12:25uh Microsoft Sentinel. Um we use a lot

12:28of stuff from from CrowdStrike. So, um a

12:30lot of our our stuff is being monitored,

12:33especially on our workstations with the

12:35CrowdStrike seam to be able to make sure

12:38that we are logging what the uh AI

12:41agents are reaching. So, even if one of

12:44our developers has written some uh

12:46product that is using Playwright to then

12:49fetch pages, that gets logged into the

12:52seam. So, that's something you want to

12:53consider doing is is make sure you've

12:55got that. So, um now you know that how

12:59how's this all play together? So,

13:02this is where your model

13:05your your uh you know Anthropic or

13:08OpenAI plus your harness, your controls,

13:11is what gives you the agent. So, it's

13:14it's what ties things together. That can

13:16be your chat agent if you're just chat

13:18interfacing or if you're feeding files

13:20and and stuff into it to have it process

13:22and do things automated, that's your

13:24automation agent. So, um,

13:27this is what, you know, brings together

13:29and and how this ties together in the uh

13:32AI world. And um

13:35you've heard me talk since I think

13:37episode one about

13:39automation. And um you know, there's a

13:42lot of different vendors here and uh

13:45this is one of those that on on

13:47LinkedIn, on um messengers, with

13:51friends,

13:52I get asked a lot and um you know, the

13:56the thing is, here are the the main

13:57players.

13:58Zapier by far is the easiest, is the one

14:03that's been around the longest. Make, in

14:05my mind, is the most stable, um a little

14:09easier to to use, but I will say that I

14:12believe that Zapier's got more

14:14integrations at this moment and has got

14:16some uh good documentation and has

14:18really stayed in the game. Um it is one

14:21that we a couple years ago used kind of

14:23heavier. We did transition to Make and

14:26then we transitioned to n8n. If you're

14:28not developers, then you don't want to

14:31um you know, be be creating your own API

14:34calls uh and you want to have good

14:36automation around things like MCP,

14:39these are your players in that space. Uh

14:42us as developers, we actually have been

14:45uh exceeding so um it's kind of a badge

14:47of honor. We've blown up Zapier and and

14:49hit limits with it before we've hit uh

14:51limits with um make.com. We purposely

14:54never went to a place where we'd have to

14:57worry about limit with uh n8n. If you're

15:00doing multiple, you know, high

15:01parallelism, a lot of processing, a lot

15:04of interactions per per second, then uh

15:07you are at the point where you want to

15:09do direct API. If you are doing multiple

15:13automations uh and you're doing four per

15:15hour, let's say you have an automation

15:17kicks off when an order comes in or

15:19appointment is booked or or of that

15:20nature, these products are excellent for

15:22that and um I am going to say that uh

15:25keep playing with these, learn learn

15:27these. They are great at giving you the

15:30superpower of being able to automate

15:32things today at a lot easier pace.

15:35Um

15:36now the holy grail of uh automation is

15:40that real high agent where it's doing

15:44things for you and you've you've seen us

15:46talk about it on in previous episodes,

15:48but here are your main players here.

15:50It's uh Open Claw, Nemo Claw. If you uh

15:53have a CISO, it um you know, that's the

15:55one you want to play with. It does

15:57require some Nvidia hardware cuz it's a

15:59an Nvidia fork and everything. It is uh

16:01very optimized for that ecosystem. I

16:04play with this one a fair amount. I do

16:06like the uh Nemo Claw. Um Open Claw is

16:09uh

16:10uh more wide open and more acceptable,

16:13uh but uh it is uh also uh requires a

16:16little more config. And then Hermes is

16:18the uh new player on the on the block

16:21and uh one you hear a lot about. Hermes

16:24has got uh uh nice uh

16:27uh eclectic interface to it. It it's got

16:30the ability to do a lot of this looping

16:33and to be able to talk to multiple MCP

16:35things. So, you can have it do a lot of

16:37your automation such as when a um

16:41let's say somebody fills out a Facebook

16:43form, you can use Make to grab that,

16:45fire it into your Hermes agent, fire off

16:49emails, fire off text messages, fire it

16:51into your Salesforce, fire a letter that

16:54uh or or draft email that goes into your

16:57email box to approve and and send off

16:59when you are ready to talk with that

17:01client. So, those are the type things

17:03that these automations work very uh easy

17:06and transparently with.

17:08So, that brings us to now the enterprise

17:10tier. So, this is the tier of agents

17:14that uh if you're a bigger organization

17:17and uh have compliance in your regulated

17:19space, these are the ones I want you to

17:21take a look at and be mindful of.

17:24And you know, some of what I've

17:25discovered on the people I'm interacting

17:27with, some people that I'm interacting

17:29with that are watching this, you guys

17:31are trying to be or or are

17:34you know, being an agency for clients.

17:37You're also, I've noticed and been

17:39interacting with people who want to be

17:41that AI person.

17:43They want to be the the AI go-to inside

17:46of a company. So, these are ones you're

17:48going to want definitely have in your

17:49back pocket and and will be brought up

17:51in meetings and and somebody on the

17:54C-suite's going to ask you about them.

17:56So, these are the ones you definitely

17:57need to know. Microsoft Copilot.

18:00This is one we would have written off a

18:02couple months ago, but in the regulated

18:04space and with E7 licensing that they've

18:07come out with, it is coming around

18:10around the regulated space. So, it's one

18:13that is

18:14back in the the spot here just because

18:17we are seeing Copilot actually start to

18:19do some stuff and be be

18:22you know, functional. I hate to say it

18:24was not functional for a while. It is

18:26back in the function. It's got good

18:27controls around it. So, it is at top

18:29here.

18:30If you're a sales organization,

18:33Salesforce Agent Force is now at the

18:37point where it is good at helping you

18:40find the data that is hard to find

18:42inside of Salesforce. So, it is

18:45expensive to run Salesforce today. You

18:47want to get the maximum value. The Agent

18:50Force is definitely something that helps

18:52you with that. I'm impressed with some

18:54of the

18:55agentic stuff that Salesforce has been

18:56putting in place. So, it it definitely

18:59receives

19:00some credit on here. From your IT

19:03management system, ServiceNow and the

19:06ServiceNow agent is really good. It

19:08helps triage tickets. It helps auto

19:10respond to tickets. It can help

19:12uh

19:13pull together the things that a level

19:17two, level three tech need to know by

19:19the time a ticket transfers over to

19:20them. So, ServiceNow has actually made

19:23some real inroads. I got to give them

19:24credit for how their ticket handling

19:26systems are are good and they're doing

19:29well in the enterprise space.

19:30This is going to be one that all of you

19:32write down because you're going to go,

19:34"Who?"

19:35If you haven't been to a health care

19:39uh tech conference or software

19:40conference, you've not heard of UiPath.

19:43Uh U Uipath?

19:45Uipath?

19:46Uipath? I don't know uh exactly how to

19:49say it. That's funny. It's It's here on

19:51the screen. UiPath.

19:53Uipath. Um

19:56these guys have got a big suite of

19:59technology. It works well. The nice

20:03thing that they are able to do today,

20:05instead of having to tie a bunch of

20:07things together with various APIs, they

20:11started doing uh APIs built, you know,

20:13they they they they built their

20:14communication layers inside. Their

20:16systems talk to each other. They're big

20:18in the uh health care space. So, if

20:21you're in health care, you're in legal,

20:23you're in fintech, this is a site you

20:25should probably go look at. And by the

20:26way, none of these companies are paying

20:28us. We're not taking any money from any

20:31of them. We are not sponsored. We get

20:33nothing out of this. This is my honest

20:36opinion.

20:37Uh as always, I'm going to give it to

20:38you straight. I'm going to tell you

20:39which companies are good to look at,

20:40which ones I I would stay away from.

20:42This is one I would take a look at and

20:44keep. They're very expensive. So, um

20:47they are not for the small business

20:49today. I don't know if too many of their

20:51platforms are going to fit that, but

20:52they've got a great suite. You should at

20:54least look at them and look at how

20:55they're doing implementations, and look

20:57at how they are helping out some of the

20:59health care companies out there. That's

21:01how I come across them. As I know some

21:02some companies use them. Um

21:04the The one on here is uh around

21:06governance, so it's a

21:09Workato Workato is how I'm going kind of

21:12love dot com domain names.

21:15This is your governance model, so

21:16control plane is what you're going to

21:18think here. It is it is a little pricey

21:20as well, but it is also a very good

21:24control mechanism. If you're talking

21:26about a couple hundred employees, this

21:27is one you might want to take a look at

21:29for for that. So,

21:31and then I had to throw down here, if

21:33you're not a Microsoft shop, if you are

21:35on the other side, which we keep seeing

21:37this grow a little bit, so the people

21:40that used to use not Microsoft and not

21:43Google are going to Microsoft and Google

21:45for for email and and document hosting.

21:48So, the people that are built on the the

21:50Google stack is going to be your G Suite

21:53and your Google Enterprise services. We

21:55do like a lot of them. It is where

21:58having knowledge of the Google APIs is

22:02very helpful because being able to

22:05provision certain things. Some of their

22:06tools are a little bit behind

22:09from a usability standpoint, but from an

22:11IT standpoint, there is so much

22:13flexibility and things that you can

22:15integrate into the Google stack and do

22:17automated with a little bit of command

22:19line development.

22:21But,

22:22here's our positioning grid. So, this is

22:25how I see the

22:28atmosphere right now of the different

22:31the different areas. So,

22:33is this a magic it's not. It's Eric's

22:37quadrants. So,

22:39this is how we look at things. So,

22:41as you know, for autonomous right now

22:44from enterprise standpoint, that's going

22:46to be your Agent Force. If you if you

22:48got Salesforce already, don't go buying

22:49it just for Agent Force

22:51or Co-pilot.

22:53Then, in your deterministic layer down

22:56here in the the lower left, that's going

22:58to be your automations and Zapier and

23:01Make.

23:02N8N is over on the self-hosted. I didn't

23:05throw

23:07models over here in the self-hosted

23:09section. That is something I do predict

23:11that will be more coming around. I know

23:14I self-host on a couple of Nvidia

23:16assets. It is something that is coming

23:18about. I do think that we are leading

23:20edge or bleeding edge in in that realm

23:22of being able to do some of this. Other

23:24companies may follow suit and do this in

23:272027.

23:28But

23:30the reason why you do is the

23:34you know, if you're if you're up here

23:36and playing and then it is the risk

23:38gradient. This is where those automated

23:41agents are sitting, which is this upper

23:44right quadrant and it's the combination

23:47of what you're hosting with your

23:50automation and it requires some

23:52self-hosting, but it is well worth it

23:55from the automation and the lifts that

23:56you get. So it it is one of those areas

24:00where the control and the autonomy

24:03is is handed off so you have a lot of

24:07control when you're over on on

24:10you know, Copilot especially like in E7

24:13version of Copilot where you have a lot

24:15less control

24:17of the agent and but you have a lot less

24:20cost and a lot more power when you're

24:22over here in the Open Claw, Nemo Claw

24:24and Hermes layers.

24:26So

24:28but I did want to bring up one you know,

24:30bit of tech that we're also seeing.

24:33The interface. So the layer at which

24:37we're interacting with things.

24:40Had been there everybody goes just to

24:41chat GPT and then into that

24:44you know, interface to get agentic work

24:48done

24:49and generative you know,

24:51work done.

24:53And

24:54now what we're seeing is because of the

24:56need for other agentic and automation

24:59where you're triggering off your your

25:01name claw your your claw

25:03that was going to uh slack very heavily.

25:06I know we've done it in slack. Teams you

25:08can do it. It's a lot harder. Telegraf,

25:11nobody who's probably watching this is

25:13is tying Telegraf into your uh thing.

25:16That's something developers do. That's

25:18something small business may do. That's

25:20something who's who's a one-man shop

25:22vibe coder is going to do. When it we're

25:24talking about doing it for more of a

25:26business, especially health care

25:27business, you're going to tie it into

25:29something that's like a slack or or a

25:31teams.

25:32But, now coming on the market is uh from

25:35Jack Dorsey who founded Twitter

25:37uh originally before selling it.

25:39Um

25:40is a product called Buzz. And I've been

25:43looking at Buzz and it is a interesting

25:46uh from the ground up has a lot less

25:49functionality when it comes to what you

25:51can do with uh some of the the core

25:53functionality.

25:55But, it has all the observability, the

25:57permissions, the protocol. It's not

26:00bolted on. It's um

26:02it's it's all built in. Um

26:06you know, it's not one of those things

26:07that I don't see as leaving slack this

26:11year. But, it is one I want to keep out

26:13there because Buzz is a slack type agent

26:17with a lot of ability to tie it into our

26:20agentics. So, I can foresee using um

26:25throwing out the the normal chat GPT

26:27interface and going with a more

26:29collaborative interface where a team uh

26:32like a marketing team is using Buzz and

26:35using Buzz to interact with your agents

26:38and your automations to be able to fire

26:40things off. Such as let's say you're

26:42putting together the WTE newsletter for

26:45a given vertical. I can see the team

26:47going in there and having the Claude

26:51agent, let's say cuz I put it to the

26:53router uh to go over to Claude and pull

26:56back and and proof and create an email

26:59that's going to go out and then it's

27:00going to go to an automation layer to

27:03put it into our email system.

27:05But Buzz can be that interface that

27:07allows everybody to come together. We've

27:09been building one of our own interfaces.

27:13We called it Winston and it's tied into

27:15Slack. I could very much see that the

27:17interface for our what was going to be

27:19our Winston internal is now going to be

27:21probably Buzz or something of that

27:23nature in 2027. Don't know if it's going

27:25to take off, but it is one of those to

27:27keep an eye on that it could and if

27:29Slack doesn't

27:30get some development out there to fill

27:32those gaps, I could foresee somebody

27:35migrating off of Slack, maybe not in

27:372026, but in 2027 for sure.

27:40So,

27:42but that's kind of the landscape and

27:45where everything lines up and where

27:47everything I feel is going. The big

27:50question I have is what questions do you

27:52have? So, I talked a lot I probably

27:55covered here. I probably made one or two

27:57people's heads spin.

27:58We love getting the questions, so go

28:00ahead and let me know what questions you

28:02have. I'll be happy to answer them.