النص 🔗
WEBVTT
00:00:51.289 --> 00:00:52.409
We would look at the metrics.
00:00:52.570 --> 00:00:56.729
Hey, this thing's got an 87% thumbs up and everybody hates it.
00:00:56.890 --> 00:00:59.689
So the intention was this is gonna be super helpful.
00:00:59.850 --> 00:01:02.729
And so we rolled it out and it just completely failed.
00:01:02.890 --> 00:01:05.289
What was it supposed to help employees do first?
00:01:05.530 --> 00:01:12.650
We have channels for our instructors when they have questions, when they run into problems, when they, you know, need to uh call out for some reason.
00:01:12.810 --> 00:01:20.009
And we put it into an AI reply box, and it just completely failed because of the accuracy.
00:01:20.170 --> 00:01:21.530
Now let's talk about the fun stuff.
00:01:21.689 --> 00:01:22.569
What's working?
00:01:22.890 --> 00:01:26.329
Now we audit every single schedule every single day.
00:01:26.490 --> 00:01:26.890
Wow.
00:01:27.049 --> 00:01:34.170
And the unlock there is that now the managers, their time is spent connecting with their instructors.
00:01:34.409 --> 00:01:40.490
Was this a failure of the AI or a failure in how you introduced it and rolled it out to the employees?
00:01:40.729 --> 00:01:43.530
And it's a hundred percent issue.
00:01:45.370 --> 00:01:48.409
Hey everybody, and welcome to another episode of Using AI at Work.
00:01:48.490 --> 00:01:50.490
My name is Chris Daigle, and I'm the host.
00:01:50.650 --> 00:01:57.770
I think one of the obvious opportunities with AI right now is finding new ways to improve how we run our business.
00:01:58.009 --> 00:02:02.250
Now, for some of us, that might mean automating something that we're already doing.
00:02:02.490 --> 00:02:15.530
But one of the things that I think is is missing from a lot of the conversations I'm having with executives is that it's they're not using it to discover problems or opportunities that maybe you didn't know existed with your existing business model.
00:02:15.930 --> 00:02:17.689
Today, our guest is Matt Harvey.
00:02:17.849 --> 00:02:21.930
Matt is the co-founder and CEO of Coastline Academy.
00:02:22.009 --> 00:02:26.009
You can find them at Coast.ai, great domain, by the way.
00:02:26.250 --> 00:02:38.569
And um, despite his tech background, uh I would say that Matt's probably had a lot more experimentation going on with different parts of his business with AI recently than he did over the career that he had in technology.
00:02:38.729 --> 00:02:44.169
Some of those experiments, like with all of us, have worked really well, and others, uh not so much.
00:02:44.409 --> 00:03:02.169
So um what we're gonna do in this conversation today is get into the real-world experience of putting AI to work in a service environment that's multi-state, multi-jurisdictional, and dealing with something as sensitive as uh, you know, putting a new driver behind the wheel.
00:03:02.409 --> 00:03:12.729
So, this is regardless of what your industry is, I think that the lessons learned from uh their initial start to their pivot to finally finding what worked is gonna be uh takeaways for everybody.
00:03:12.889 --> 00:03:21.209
So, Matt, before we get started, is there anything as far as like by way of your introduction that you think people need to know that would help them kind of understand the context of where we're gonna go today?
00:03:21.609 --> 00:03:21.929
Sure.
00:03:22.009 --> 00:03:22.250
Yeah.
00:03:22.409 --> 00:03:24.489
So uh co-founder and CTO.
00:03:24.729 --> 00:03:27.289
Um, so my other co-founder is a CEO.
00:03:27.530 --> 00:03:30.969
Um, so yeah, I mean, my background has always been in tech.
00:03:31.129 --> 00:03:35.209
It's I've been involved in AI since before we called it AI.
00:03:35.370 --> 00:03:39.929
In fact, our first product as a company was fully AI driven.
00:03:40.009 --> 00:03:41.530
It was a computer vision product.
00:03:41.689 --> 00:03:52.090
And so, like you mentioned with that pivot to a service company and providing driving lessons, um, we have had technology at the core the whole time.
00:03:52.409 --> 00:03:57.530
And yet we're delivering a service, like you said, a sensitive service.
00:03:57.609 --> 00:03:59.209
We're a people-first company.
00:03:59.370 --> 00:04:08.090
And so I'm really excited to talk about how you know we've utilized technology to better the human aspect of what we do as a business.
00:04:08.729 --> 00:04:15.929
Before we jump into my questions, I guess uh what would you want the listeners to take away from today's conversation?
00:04:16.009 --> 00:04:17.850
What do you think would be the big win for them?
00:04:18.569 --> 00:04:19.370
The big win.
00:04:19.530 --> 00:04:25.449
So I think about AI adoption in the workplace in two separate systems.
00:04:25.689 --> 00:04:35.370
There's the individuals who are using specific tools to make themselves more efficient, to enjoy work more, things like that.
00:04:35.530 --> 00:04:40.410
And then there are the system-level company AI initiatives.
00:04:40.730 --> 00:04:43.850
And I think those get conflated a lot.
00:04:44.090 --> 00:04:52.170
And so when I think about AI in the workplace, I often think about these things as there's two categories of AI use.
00:04:52.329 --> 00:04:57.290
And so hopefully we can dive into both of them and we can talk a little bit about how we use each of those.
00:04:57.449 --> 00:05:04.250
Um, but you know, to me, I think a big takeaway is we need to be thinking about AI at the company level, at a systems level.
00:05:04.329 --> 00:05:16.569
And we also need to be thinking about AI as what are your employees doing day to day in their own world where they're maybe using AI more bespoke, more in isolation, and they're both really valuable.
00:05:16.890 --> 00:05:20.250
So two different conversations essentially about AI and the organization.
00:05:20.730 --> 00:05:21.450
I think so.
00:05:21.770 --> 00:05:29.770
So for the listeners, they might be thinking driving
Load full transcript
school, but I want you to kind of share the scale and scope of what you guys are actually pulling off at Coast.
00:05:30.410 --> 00:05:34.410
Yeah, yeah, it's really, it's it's such a fascinating pivot, right?
00:05:34.569 --> 00:05:36.330
So you mentioned that in the intro.
00:05:36.490 --> 00:05:43.930
Um, you know, we did not leave my when I say we, my co-founder and I, we didn't leave our jobs to start a driving school.
00:05:44.010 --> 00:05:49.050
Um, you know, we both had, you know, great jobs in Silicon Valley and the, you know, the tech world.
00:05:49.290 --> 00:05:53.450
And we had this grand vision of improving driver safety.
00:05:53.530 --> 00:05:57.370
You know, our mission from the beginning has always been to eradicate car crashes.
00:05:57.530 --> 00:06:01.129
We thought we were going to do that by putting AI in everybody's vehicle.
00:06:01.290 --> 00:06:06.250
So you can think autopilot in an old Prius, that sort of thing.
00:06:06.410 --> 00:06:08.250
And so that's what we set out to do.
00:06:08.490 --> 00:06:18.330
Now, as a business, we made the realization back in 2016, 17, 18 when we were experimenting that that just wasn't going to be the right business.
00:06:18.569 --> 00:06:22.650
But we did really want to make an impact on driver safety in general.
00:06:22.810 --> 00:06:25.129
And so we went through a lot of iterations.
00:06:25.290 --> 00:06:26.650
Now, this is before AI.
00:06:26.810 --> 00:06:33.850
I can't even imagine how much more efficient it would have been back then to spin up smoke tests and things with AI.
00:06:34.010 --> 00:06:37.770
Uh, but thankfully, we were, you know, uh, I was a very fast engineer.
00:06:37.930 --> 00:06:49.450
We spun up many different experiments, we tried all sorts of things, and then we discovered that this driving education space was essentially untouched by software.
00:06:49.770 --> 00:06:59.930
And we were really surprised to find how sort of antiquated the systems and the whole experience of learning to drive was.
00:07:00.090 --> 00:07:02.250
And so we had this aha moment.
00:07:02.410 --> 00:07:05.450
Hey, we can apply our, you know, our tech chops.
00:07:05.689 --> 00:07:18.330
You know, this is when you know Airbnb had taken off, Uber had taken off, and we thought, hey, we could do that for the learning to drive space, which would be fun because we could apply all of the technology and the user experience knowledge that we had.
00:07:18.490 --> 00:07:21.930
Um, we actually used AI right at the very beginning, which we can talk about.
00:07:22.010 --> 00:07:28.569
Um, but then we could also make an impact on our mission and do something that you know we felt really good about when we woke up in the morning.
00:07:29.129 --> 00:07:32.330
Um to answer your question about scale, sorry, real quick.
00:07:32.490 --> 00:07:34.490
So we started in 2018.
00:07:34.650 --> 00:07:37.530
We're now the largest provider of driving lessons in the country.
00:07:37.770 --> 00:07:43.770
We operate in uh seven, uh eight states, depending on how you count it, uh, depending on what service is.
00:07:44.170 --> 00:07:52.410
Um we're we cover 95% of California, um, and we have just we've grown to you know be the largest driving school in the country.
00:07:53.129 --> 00:07:55.610
So what was it that you saw?
00:07:55.770 --> 00:08:12.490
I mean, obviously, I can only imagine the uh investment and infrastructure necessary to introduce uh AI augmentation into the driving experience for commercial and uh you know uh personal use vehicles and that sort of thing.
00:08:13.129 --> 00:08:20.810
What what was happening to where uh you guys decided, hey, this isn't the angle, but I see a green chute over here.
00:08:20.890 --> 00:08:22.569
Let's let's let's dig in there.
00:08:23.770 --> 00:08:24.170
Yeah.
00:08:24.410 --> 00:08:44.490
So um what's so interesting about this uh space as just an industry is that it's a service that almost everybody needs, and yet it was serviced by and still is by hundreds and across the country thousands of very small businesses.
00:08:45.049 --> 00:08:53.210
And there's some upside to that, which is if you live in a place with a good small business driving school, you're very happy, right?
00:08:53.529 --> 00:08:55.370
You some of these driving schools are amazing, right?
00:08:55.450 --> 00:08:56.730
These people have been doing it a long time.
00:08:56.889 --> 00:09:03.049
They their passion is there, their attention to safety is there, but that's in that one spot, right?
00:09:03.210 --> 00:09:07.289
And you can imagine in California alone, there's something like 600 driving schools.
00:09:07.450 --> 00:09:14.090
Um, there's a lot of those places where people live that they can't get access to this high quality education.
00:09:14.250 --> 00:09:21.610
And so what we learned along the way, though, is that it's really hard to scale this type of business to a significant size.
00:09:21.769 --> 00:09:29.929
And so when you do have a really high-quality driving school and they try to scale, they get up to five, six, maybe even 10 driving instructors.
00:09:30.089 --> 00:09:34.490
And then the logistics and the operational complexity just explodes.
00:09:34.730 --> 00:09:49.289
And that was really where we saw not only could we, you know, offer a good experience and bring software into it, we could bring software, automation, operations, and AI to the entire landscape, which would allow us to scale beyond what anybody else had done.
00:09:49.610 --> 00:10:08.969
So for the listener, maybe you don't run a multi-state driving academy, but um, I would I what I'm seeing here, this maps to trades, home services, anything where there's going to be the logistics of scheduling and uh driver times and routes and all the things required here.
00:10:09.049 --> 00:11:08.740
And it's only compounded in your environment with I mean getting people behind the V the wheel of a vehicle is um a little uh a risk, let's
say.
00:11:09.059 --> 00:11:19.139
So you guys have this business, you've got different locations, you've got obviously different packages for people, you've got scheduling, you've got state, local, federal requirements to follow.
00:11:19.459 --> 00:11:27.059
Um, it seems to me that like you get behind the wheel, maybe your grandfather or somebody taught you how to drive, easy peasy.
00:11:27.299 --> 00:11:28.579
But it's not that easy.
00:11:28.819 --> 00:11:39.620
So, how did you guys realize that customers might be having trouble kind of navigating all of the uh the process of, hey, I'm gonna teach, I'm gonna want my daughter to get her license.
00:11:39.699 --> 00:11:43.059
I need to find the right place, we need to schedule it and those sorts of things.
00:11:43.139 --> 00:11:49.219
Because you guys, some of what you do or a big portion of what you do isn't, hey, I'm gonna show up to the driving school and take off.
00:11:49.299 --> 00:11:53.459
You guys actually will arrange for home pickups and that sort of thing, right?
00:11:54.099 --> 00:11:55.459
Yeah, that's exactly right.
00:11:55.620 --> 00:12:02.259
And and that's where it really does, the the um the complexity just explodes because it's also the vehicles, right?
00:12:02.339 --> 00:12:08.980
So we have vehicles, park places, we have our instructors, you know, they may be at their own home, they may be in an office.
00:12:09.139 --> 00:12:27.539
So it it if you think about the traveling salesman problem, it's it's like that, except we don't have all the information ahead of time because we don't know yet, you know, when somebody comes to book, we're trying to give them an optimal experience, we're trying to give them optimal opt optionality while keeping our instructor schedules really efficient.
00:12:27.699 --> 00:12:30.659
And it's it's a really hard, you know, computer science problem.
00:12:30.819 --> 00:12:33.219
It's where AI shines, right?
00:12:33.699 --> 00:12:38.339
What's AI good at reading tons of data and finding optimization opportunities?
00:12:38.500 --> 00:12:55.139
Um, but the the other thing for us was, you know, as part of our DNA, and and this will really go into um, you know, I think as we're actually using AI at the company, it's part of our DNA to be obsessed with interviewing our stakeholders.
00:12:55.379 --> 00:13:01.299
And so before the pivot, we had been doing, you know, we set a goal when we decided to start the company.
00:13:01.539 --> 00:13:07.299
Every single week we're going to do 10 customer interviews at least each between me and my co-founder.
00:13:07.699 --> 00:13:22.339
And the amount of information and insights, things that, you know, you you can reflect as much as you want, but when you actually talk to the customers, and then later when we talk to our employees, the amount of insights we get is just it's just unbelievable.
00:13:22.500 --> 00:13:25.939
And that's actually why we found the driving school space to begin with.
00:13:26.099 --> 00:13:27.779
Neither of us have been thinking about it.
00:13:27.939 --> 00:13:32.179
But people would bring up these little nuggets, like, oh, my driving instructor never showed up.
00:13:32.339 --> 00:13:34.500
I had to pay cash, they were running errands.
00:13:34.579 --> 00:13:35.539
They're we're like, what?
00:13:35.779 --> 00:13:42.419
Like, how is this all happening in this service that's so critical and so important to the health of everybody?
00:13:42.659 --> 00:13:54.259
Um, and so that's constantly just soliciting feedback and getting a deeper understanding of where people's pain points are, that's really what opened our eyes to this whole thing.
00:13:54.659 --> 00:14:05.299
So I want to talk about that actually, because you know, we we've got listeners and that um maybe have never done that outside of you know talking to a customer throughout the course of their service delivery.
00:14:05.539 --> 00:14:08.099
What was this customer interview process?
00:14:08.259 --> 00:14:09.059
What were you looking for?
00:14:09.219 --> 00:14:09.939
What were you asking?
00:14:10.099 --> 00:14:15.860
What was the feedback from the customers when you asked them, hey, can we get some time on the phone to ask you these questions?
00:14:16.019 --> 00:14:19.779
Because I think that other listeners, if they're not doing this, totally get it.
00:14:19.939 --> 00:14:26.179
Huge insights because you're going to discover these problems you didn't realize were problems until you talked to your customers.
00:14:26.419 --> 00:14:28.419
So how did you guys structure that?
00:14:28.500 --> 00:14:31.539
What were the types of questions that what were you looking for out of it?
00:14:32.179 --> 00:14:33.219
Yeah, yeah, yeah.
00:14:33.379 --> 00:14:34.019
Great questions.
00:14:34.179 --> 00:14:45.219
So I I think the the a really important thing to understand about this process is that it's a lot of work and there's no shortcuts until after the fact.
00:14:45.379 --> 00:14:47.860
So at the time, you know, we had these huge transcripts.
00:14:47.939 --> 00:14:52.419
Now it's so you got, you know, you turn on granola, um, you have your AI summarize it.
00:14:52.500 --> 00:14:54.740
So that is such a big benefit now.
00:14:54.899 --> 00:15:03.939
But the way that we structured these things is um because we had our mission of eradicating car crashes, we knew that we wanted to talk to people about road safety.
00:15:04.179 --> 00:15:09.860
And so we just started by asking friends who had kids, hey, can I talk to your kid?
00:15:09.939 --> 00:15:13.459
Or hey, parent, you know, what what are you worried about?
00:15:13.620 --> 00:15:17.379
Um, oh, you're worried about them, you know, I don't know, texting and driving.
00:15:17.539 --> 00:15:18.579
Okay, well, let's talk about that.
00:15:18.659 --> 00:15:20.819
What, you know, what are you doing now to address that?
00:15:20.980 --> 00:15:26.419
What and so we actually, you know, we thought, hey, maybe
we could do something with um, you know, texting and driving.
00:15:26.500 --> 00:15:28.419
And, you know, we explored some things there.
00:15:28.659 --> 00:15:39.299
Other things were, you know, that came up were um, you know, I I just don't feel comfortable to like sitting in the car with my kid while they're driving because like I can't do anything and it's scary.
00:15:39.860 --> 00:15:41.459
Okay, well, tell me more about that.
00:15:41.620 --> 00:15:43.219
Tell me more about that, right?
00:15:43.620 --> 00:15:50.819
And um, and what we would do is after we would have these sessions, we would my co-founder and I would just compare notes.
00:15:50.980 --> 00:15:51.699
What did you learn?
00:15:51.779 --> 00:15:52.339
What did I learn?
00:15:52.419 --> 00:15:54.819
Hey, wasn't it interesting that Sally said X?
00:15:54.980 --> 00:15:55.779
Oh, that's interesting.
00:15:55.860 --> 00:15:58.019
Maybe that's a thread we can pull on a little bit more.
00:15:58.179 --> 00:16:04.579
And then sometimes we would go back to people we talked to talk to and say, Hey, you had mentioned this thing about your driving instructor.
00:16:04.740 --> 00:16:05.939
Like, tell me more about that.
00:16:06.259 --> 00:16:07.699
How did you book your lesson?
00:16:07.860 --> 00:16:10.659
You know, and and that's where it it all compounds.
00:16:10.740 --> 00:16:12.419
But I mean, we went to great lengths here.
00:16:12.500 --> 00:16:21.459
I mean, we we would sit in coffee shops with a little sign that said, you know, tell me about driving, and we'll give you a$10 gift card, you know, to like this this coffee shop.
00:16:21.539 --> 00:16:23.939
And so people come up and say, like, like, really?
00:16:24.099 --> 00:16:26.980
I just have a conversation with you, and you'll give me 10 bucks.
00:16:27.219 --> 00:16:28.500
Like, yeah, let's do it.
00:16:28.659 --> 00:16:30.659
And so that's where I say it's a lot of work.
00:16:30.819 --> 00:16:33.219
You got to just get out there and just talk to people.
00:16:33.939 --> 00:16:37.139
So, what would you have missed out on had you not taken that?
00:16:37.219 --> 00:16:44.019
Because I I I want to pull on this a little bit more because I don't see this happening a lot with uh the clients that we're working with.
00:16:44.179 --> 00:16:47.219
And I I know that there would be some major unlocks.
00:16:47.379 --> 00:16:50.899
They'd hear just like you did, like, oh wow.
00:16:51.459 --> 00:16:53.379
So, where did you come up with the questions?
00:16:53.459 --> 00:16:55.059
Was it just an organic conversation?
00:16:55.219 --> 00:16:58.179
Was there a structure to this inquiry?
00:16:58.339 --> 00:17:01.219
Um, and how did you kind of compile everything?
00:17:01.299 --> 00:17:05.139
Was this pre-gen AI, I guess, to support the analysis?
00:17:05.619 --> 00:17:08.179
It was, it was all pre-gen I gen AI.
00:17:08.259 --> 00:17:12.179
So we literally just had a Google Drive with every conversation.
00:17:12.420 --> 00:17:16.500
And so, um, and I honestly, like we didn't even have transcription at the time.
00:17:16.579 --> 00:17:26.980
So I remember doing Zoom call after Zoom call, you know, for people who would who would jump on uh or you know, on the phone, and I'd literally just be typing as fast as I could, you know, I'd type everything that they said.
00:17:27.219 --> 00:17:37.059
So, in terms of the list of questions and the structure, um, we started with each conversation started with a theme.
00:17:37.219 --> 00:17:40.019
So we would say, hey, I'm gonna talk to this person.
00:17:40.179 --> 00:17:44.339
They're 17 years old, they just got their driving, their driver's license.
00:17:44.579 --> 00:17:50.980
And for this conversation, I'm going to center it around, you know, what are they nervous about now?
00:17:51.299 --> 00:17:52.259
I'll just make this up.
00:17:52.339 --> 00:17:54.019
Uh, you know, make up an example.
00:17:54.099 --> 00:17:59.700
What another one would be um, you know, tell me um, you know, how did you study for your test?
00:17:59.860 --> 00:18:03.059
Okay, so I'm gonna frame all my questions about how they study for the test.
00:18:03.460 --> 00:18:09.539
And but it started super vague, super top of funnel, if you think about it in a sales term.
00:18:09.779 --> 00:18:11.700
Yeah, a bit spray and pray, right?
00:18:11.779 --> 00:18:14.660
Let's let's just ask a bunch of people stuff about driving.
00:18:14.819 --> 00:18:18.900
And then we would just pull on the threads and get deeper into the funnel, deeper in the funnel.
00:18:19.059 --> 00:18:23.220
And then by the end, we were asking very specific questions.
00:18:23.380 --> 00:18:24.900
How did you find your driving school?
00:18:24.980 --> 00:18:25.140
Right.
00:18:25.220 --> 00:18:27.620
When we when we realized that we wanted to do this.
00:18:27.779 --> 00:18:30.660
Um, you know, did did your instructor show up on time?
00:18:30.819 --> 00:18:32.259
What did you like about your instructor?
00:18:32.420 --> 00:18:35.140
What are the things that your instructor said that made you feel comfortable?
00:18:35.299 --> 00:18:37.779
What did they say that made you maybe more nervous?
00:18:37.940 --> 00:18:40.420
And so we got more and more and more detailed.
00:18:40.579 --> 00:18:52.019
But we even took it so far as when we finally did start the driving school, my co-founder and I became licensed driving instructors and we taught 100 lessons each.
00:18:52.259 --> 00:18:53.619
And same thing.
00:18:53.859 --> 00:18:55.779
Hey, so nice to meet you, you know.
00:18:55.859 --> 00:18:59.139
And then 20 minutes into the drive, you got to take a break, right?
00:18:59.299 --> 00:19:01.940
These 15 and a half year old can't drive for two hours straight.
00:19:02.099 --> 00:19:03.940
Hey, so how'd you hear about us anyway?
00:19:04.099 --> 00:19:05.379
Oh, your friend told you.
00:19:05.460 --> 00:19:06.500
What did they tell you?
00:19:06.659 --> 00:19:09.299
And so, and to this
day, we continue to do it.
00:19:09.619 --> 00:19:22.500
One of the conversations that we had, uh one of the things that we talked about in the the pre-interview a few weeks ago was this internal AI support bot that your employees weren't very um bullish about initially.
00:19:22.819 --> 00:19:24.180
You remember that conversation?
00:19:24.500 --> 00:19:24.899
Oh, yeah.
00:19:25.379 --> 00:19:25.619
Yeah.
00:19:26.019 --> 00:19:31.859
So that's not what most of us who haven't rolled those things out would expect.
00:19:32.019 --> 00:19:37.139
We would expect that the employee would get this tool and be extremely enthusiastic about, oh, this is easy to use.
00:19:37.299 --> 00:19:38.420
What was that experience?
00:19:38.579 --> 00:19:41.460
What was it supposed to help employees do first?
00:19:42.579 --> 00:19:47.139
Yeah, so we have a lot of driving instructors across the country, like we talked about.
00:19:47.460 --> 00:20:01.539
And because of that, there's a lot of things that just come up day to day in their in their jobs, uh especially as they're going out on the road and you know, meeting all these people and putting, you know, thousands of miles on cars.
00:20:01.779 --> 00:20:06.819
And so we have most of our communication as a company happens in Slack.
00:20:07.139 --> 00:20:15.700
And we have channels for our instructors when they have questions, when they run into problems, when they, you know, need to uh call out for some reason.
00:20:15.940 --> 00:20:26.339
And all of that information is in Slack, which is really nice because it gives us a single place to farm basically all of the harvest, all of the information over the years.
00:20:26.579 --> 00:20:40.659
And so we did that and we put it into an AI reply bot with the with the intention that most of the questions have one answer because they've happened so many times.
00:20:40.899 --> 00:20:44.339
Hundreds of instructors over many years, you get a lot of the same things.
00:20:44.500 --> 00:20:48.099
So we thought, hey, wouldn't this be nice if they just got an answer right away?
00:20:48.259 --> 00:20:52.019
They didn't have to wait for a support person, they didn't have to wait for their manager.
00:20:52.180 --> 00:20:55.619
So the intention was this is going to be super helpful.
00:20:55.859 --> 00:20:57.139
And so we rolled it out.
00:20:57.299 --> 00:20:59.460
Now, this was this was about a year ago.
00:20:59.700 --> 00:21:02.980
So, you know, you can make an argument that we're still early.
00:21:03.059 --> 00:21:04.980
It was definitely early a year ago.
00:21:05.139 --> 00:21:18.019
Um, and it just completely failed because the the accuracy, like the percentage of times that it was right was just below the acceptable threshold, right?
00:21:18.180 --> 00:21:22.339
So if you if you think about that type of service, it has to be right.
00:21:22.659 --> 00:21:22.899
Yeah.
00:21:23.139 --> 00:21:35.139
I mean, I'm just making this up 90% of the time, 95% of the time to be
useful because those five or 10% of the time that it's not right, the impact of that is significant.
00:21:35.539 --> 00:21:37.940
And so, and that's what ended up happening, right?
00:21:38.019 --> 00:21:39.139
We would look at the metrics.
00:21:39.299 --> 00:21:44.259
Hey, this thing's got an 87%, you know, thumbs up, and everybody hates it.
00:21:44.420 --> 00:21:48.740
You know, like how do you weigh these two data points?
00:21:48.980 --> 00:21:55.299
And, you know, then of course we dug deeper into it and had those conversations and said, you know, tell me more about that.
00:21:55.379 --> 00:21:56.899
Like, why was this not helpful?
00:21:57.059 --> 00:22:06.099
And that's where we saw that we were just that last five, 10% was making a huge negative impact on the human experience that people were having.
00:22:06.420 --> 00:22:20.180
So this tracks, I was just uh speaking at uh a Vistage Executive Summit in Charlotte last week, and they put up a survey that had all the the Vista, which is if you're not familiar with Vistage, it's like a local CEO peer groups, but it's a a global organization.
00:22:20.420 --> 00:22:29.700
And they they had this question in the main room, and it was like, you know, um uh what the question was, what concerns you most about scaling AI in your business?
00:22:29.940 --> 00:22:33.779
It popped up a QR code, the individuals could do it, then do it real quick on their phone.
00:22:33.859 --> 00:22:37.700
So we were seeing the you know the the bar graph change in real time.
00:22:37.940 --> 00:22:38.819
And they did this.
00:22:38.980 --> 00:22:40.259
I had spoken at one in St.
00:22:40.420 --> 00:22:49.059
Louis and one in San Francisco in the past month, and they did the same thing, and consistently the number one concern was trust in the outputs.
00:22:50.180 --> 00:23:10.259
So, what you're sharing here, it it for those listeners who haven't maybe rolled something like this out, you need to pay attention to this next, these next few questions we're gonna talk about because you it's likely that you're going to have your team and your, you know, the rest of the executive staff or whoever's participating in this have those same concerns.
00:23:10.339 --> 00:23:19.379
So I want to ask you, was this a failure of the AI or a failure in how you introduced it and rolled it out to the employees?
00:23:20.180 --> 00:23:21.059
Great question.
00:23:21.299 --> 00:23:24.259
So and it's a hundred percent a trust issue.
00:23:24.339 --> 00:23:26.259
I like nail on the head.
00:23:26.420 --> 00:23:34.019
Um, and that's what really hurt us about this is that, like, you know, we hurt the trust that we have established with our employees.
00:23:34.180 --> 00:23:39.220
So where the root of the issue was, um rollout I thought was good.
00:23:39.700 --> 00:23:50.019
In terms of our communication about it, I I don't want to, so I want to be careful not to say it was a failure of the AI, because that's a failure of us.
00:23:50.420
--> 00:24:11.779
And so, you know, I believe it was a failure in the way that we built the AI, in that it didn't have the enough information and it was not properly tuned to say, I don't know the answer to this, let me get you a real person.
00:24:12.899 --> 00:24:17.460
And when it comes to trust, I think that second part is critical.
00:24:17.700 --> 00:24:34.019
And you know, and when we talk later about some of the AI initiatives that we've done that have worked really well, one of the key um common threads is that there's a human in the loop validating and verifying the output of the AI.
00:24:34.740 --> 00:24:42.899
And in a world where the AI is overconfident and it's wrong, that's the recipe for disaster.
00:24:43.220 --> 00:24:43.859
Yeah.
00:24:44.259 --> 00:24:50.579
Now, this type of like AI mechanism is very uh, it's like low barrier to entry in your organization.
00:24:50.740 --> 00:24:51.700
It makes a lot of sense.
00:24:51.779 --> 00:24:58.500
It's kind of like general QA thing, whether it was HR policy or how do we, you know, what's our sales process or whatever.
00:24:58.659 --> 00:25:06.339
So this is something that I think um a lot of companies, when they get started, this type of thing is like one of the ideas that's on the whiteboard to do.
00:25:06.500 --> 00:25:14.899
So, what advice would you give them if they're wanting to create their version of this uh Slack-based knowledge bot that's AI powered or whatever?
00:25:15.059 --> 00:25:21.059
What would you tell them to do based on the lessons that you learned from your experience and kind of with where the models are today?
00:25:21.539 --> 00:25:28.099
Yeah, and so thankfully the models are much better now, um, which, you know, that's like a freebie, which is yeah, which is really nice.
00:25:28.259 --> 00:25:30.819
And and we should expect that to continue.
00:25:31.059 --> 00:25:40.019
So, you know, the the mantra is like design for where the models will be in six months, which is true, and you can't roll something out now that will only work in six months.
00:25:40.259 --> 00:25:50.659
But that aside, the way that I think about introducing this stuff is it comes very much from my engineering background, which is that there are different phases of rolling out the product.
00:25:50.980 --> 00:25:59.059
And to make it as simple as possible, there's the development, there's the QA, after release, there's the monitoring.
00:25:59.299 --> 00:26:12.019
And I think as us on the business side think about how we roll these things out, we really need to think about that mindset of yeah, of course, everybody knows you have to test, test, test before you roll it out.
00:26:12.259 --> 00:26:19.940
But the part that I think is missed often is then you have to monitor, monitor, monitor after it actually gets rolled out.
00:26:20.099 --> 00:26:25.460
And that monitoring can take the form of automated monitoring, whether it's, you know, thumbs up, thumbs down.
00:26:25.700 --> 00:26:29.299
But more important than ever, the human connection.
00:26:29.460 --> 00:26:38.019
You need to, after you roll these things out, you need to talk to the people who have experienced the other side of it and just get their candid feedback.
00:26:38.339 --> 00:26:39.220
And what does that look like?
00:26:39.299 --> 00:26:45.859
Just get like, let's say I was going to roll out something, uh, you know, HR policies for the organization, how much PTO time do I have off?
00:26:45.940 --> 00:26:46.579
What holidays?
00:26:46.740 --> 00:26:47.299
Whatever.
00:26:47.460 --> 00:26:53.700
Uh, does that look like a small pilot group of people that use it for uh a week, a month, whatever?
00:26:53.779 --> 00:26:58.899
And then in the process, I'm kind of having a weekly check-in with them and saying, hey, what's working?
00:26:59.059 --> 00:26:59.619
Blah, blah, blah.
00:26:59.779 --> 00:27:01.859
Or how how would you structure that?
00:27:02.180 --> 00:27:03.779
Yeah, that's that's exactly right.
00:27:03.859 --> 00:27:15.460
That that's one of the ways that we do it is um so we'll again thinking about this as an engineering rollout, we limit the scope to a certain set of people, which is what we do on the engineering side always, right?
00:27:15.859 --> 00:27:21.220
Uh you feature flag a new product or whatever, it only goes to a subset of people.
00:27:21.539 --> 00:27:23.139
That's a controlled group.
00:27:23.299 --> 00:27:28.500
You communicate to them beforehand, hey, you're gonna be in this group after you use this.
00:27:28.579 --> 00:27:31.299
I'm gonna reach out to you, I'm gonna get some feedback.
00:27:31.539 --> 00:27:40.980
The other thing that we do a lot that I would advise everybody to do is AIs are so good at reading data for you.
00:27:41.220 --> 00:27:49.700
And so, for instance, in this chat uh, you know, this chat product example, we of course have all the traces, all the conversations that have happened.
00:27:49.779 --> 00:27:52.980
You know, we strip them of PII and everything, but they're there.
00:27:53.299 --> 00:28:03.379
You take Astra or Fable right now and say, hey, go read the last 25 conversations and tell me what was good, what was bad, did people get frustrated?
00:28:04.099 --> 00:28:06.259
And and it's like a miracle, right?
00:28:06.339 --> 00:28:11.139
It comes out and it's like three people had this problem, and you say, Oh, who are those people?
00:28:11.299 --> 00:28:12.819
And then I reach out to them, right?
00:28:12.899 --> 00:28:15.059
And you know, like I said, use Slack for everything.
00:28:15.220 --> 00:28:16.099
I'll just reach out.
00:28:16.259 --> 00:28:22.740
Hey, uh, you know, Kimberly, I saw that you had this weird interaction with the chatbot.
00:28:22.980 --> 00:28:24.339
What were you trying to do?
00:28:24.500 --> 00:28:26.659
What do you think the right answer could have been?
00:28:26.819 --> 00:28:27.940
It's just a follow-up.
00:28:28.180 --> 00:28:29.619
So thank you for sharing
that.
00:28:29.859 --> 00:28:32.500
Um, now let's talk about the fun stuff.
00:28:33.299 --> 00:28:34.420
What's working?
00:28:34.579 --> 00:28:36.579
Because again, I keep going back to this.
00:28:36.740 --> 00:28:47.379
I don't know how many uh driving schools listen to the podcast, but I do know that we have a lot of um manufacturing and trades, but also like a lot of home services and things like that.
00:28:47.539 --> 00:29:00.659
And it seems like there's a you know, takeaway the widget, and there's a lot of the same issues that they're dealing with that are uh constraints causing frustration with customers and those sorts of things.
00:29:00.980 --> 00:29:11.139
So what are you what's working well with what you guys have done by introducing generative AI into what used to be kind of a disjointed and and clunky experience?
00:29:12.019 --> 00:29:12.339
Yeah.
00:29:12.500 --> 00:29:24.980
So one of the things that we have done, or really a theme that we look at, is um each person at the company they have a job to do.
00:29:25.460 --> 00:29:28.659
And their job is comprised of many tasks.
00:29:28.899 --> 00:29:37.220
And what we really want to focus on is not how do we automate a job or you know, take away a job.
00:29:37.379 --> 00:29:40.819
It's how do we automate or enhance a task?
00:29:41.299 --> 00:30:16.259
And so what we did at the company is we looked at what are the tasks that make up the jobs, which of those tasks are a waste of uh humans' time, are not done well by a human, or most importantly, perhaps, what are the tasks that we're supposed to be doing more of that we just aren't because they're impossible to be done by a human, the scale is too big, the it's too tedious, nobody wants to do it.
00:30:16.500 --> 00:30:19.460
And so we really love focusing on those.
00:30:19.700 --> 00:30:25.779
What are the opportunities that AI unlocks for us that we just wouldn't be doing otherwise?
00:30:26.259 --> 00:30:41.059
And so a specific example of that is because we have so many instructors out on the road, and because the complexity is so great with all the different vehicles and addresses and traffic in LA, you know, all the things.
00:30:41.299 --> 00:30:48.819
Um after a day is done for an instructor, that produces a time card.
00:30:49.059 --> 00:30:52.740
And we have hundreds of these time cards every day across the country.
00:30:52.899 --> 00:31:03.460
And you can imagine over the course of a year, that is a massive amount of data that no human can look through to find, you know, where's the experience bad for the instructor?
00:31:03.539 --> 00:31:04.740
Where is it bad for the student?
00:31:04.899 --> 00:31:08.259
Where is an instructor being inefficient without even knowing it?
00:31:08.339 --> 00:31:09.940
They because they just say, hey, this is my job.
00:31:10.019 --> 00:31:15.619
I have to drive across, you know, the city three times when we have an instructor on either side of the city who could have been doing those.
00:31:15.859 --> 00:31:31.379
And so that for us has been a huge unlock of applying AI to the idea of um, you know, schedule efficiency, schedule planning opportunities, um, you know, finding coaching moments for our instructors.
00:31:31.539 --> 00:31:36.579
And so our managers would do some of that because they need to.
00:31:36.899 --> 00:31:40.339
Now we audit every single schedule every single day.
00:31:40.500 --> 00:31:40.819
Wow.
00:31:41.059 --> 00:31:49.460
And the unlock there is that now the managers, their time is spent connecting with their instructors, right?
00:31:49.619 --> 00:31:55.460
Connecting with each other, it focusing on things that they are uniquely suitable to focus on.
00:31:55.700 --> 00:32:03.379
Because let's be honest, nobody wants to sit there clicking through a calendar, saying, like, oh, you know, was this lunch break a little bit too long?
00:32:03.539 --> 00:32:07.379
Like, talk about mindless work, but it's important work.
00:32:07.539 --> 00:32:11.059
And so that's where AI has really, really shone for us.
00:32:11.700 --> 00:32:13.299
So I like this idea.
00:32:13.379 --> 00:32:19.940
Uh, and I first got introduced to it maybe by the Department of Uh Labor Statistics or Bureau or whatever.
00:32:20.180 --> 00:32:26.740
This idea that a job is uh made up of projects which are made up of tasks, right?
00:32:27.220 --> 00:32:35.059
So how do you and when you ask somebody what they do, if they've been doing it long enough, they don't know how they do what they do.
00:32:35.139 --> 00:32:36.180
They just they just do it.
00:32:36.259 --> 00:32:39.460
It's become kind of like um unconscious competence, right?
00:32:39.779 --> 00:32:48.980
So how would you go about um helping somebody who's never really had to think about their job in that fashion before to start to identify those tasks?
00:32:49.059 --> 00:32:58.740
Is that something that AI, like I've I mean, if they're using the tools, I guess that AI can look at their calendar and their in, I mean their yeah, their email inbox and their meeting transcripts and understand this?
00:32:58.899 --> 00:33:05.379
Or is it something that you have the individuals, let's say I've I've got some a department, let's say we want to start in finance.
00:33:05.619 --> 00:33:23.220
Um like how how would a company go about and and maybe a micropilot helping the process owner start to break down their quote unquote job into these tasks into task level stuff that we could start to identify, oh, AI could help with this.
00:33:24.420 --> 00:33:33.940
It it really follows a similar theme to what we talked about earlier with interviewing customers and starting broad and then drilling down.
00:33:34.259 --> 00:33:43.700
We do the exact same thing with our managers, with you know, your example of finance, we did it with our finance people, our our accounting folks.
00:33:43.940 --> 00:33:47.779
And it's there's no shortcuts to this.
00:33:48.019 --> 00:33:55.539
It's sitting
down, talking to them and saying, just walk me through how you did month end close this month.
00:33:55.700 --> 00:33:57.700
Just just talk about it, right?
00:33:57.940 --> 00:34:05.059
And it doesn't have to be, hey, you know, what tasks do you, you know, write down your algorithm for closing the books, right?
00:34:05.460 --> 00:34:08.019
People are like, I don't know, I just close the books, I do this, I do that.
00:34:08.180 --> 00:34:10.099
Just tell me, tell me what you do.
00:34:10.260 --> 00:34:10.819
Oh, okay.
00:34:10.899 --> 00:34:16.579
You look through QuickBooks, or we don't use QuickBooks, but as an example, oh you yeah, you look through QuickBooks.
00:34:16.739 --> 00:34:19.059
Okay, what are some of the things you find?
00:34:19.220 --> 00:34:24.579
Like, oh, um, you know, every month you have to go look in the bank account for these like 17 checks.
00:34:24.819 --> 00:34:25.460
Interesting.
00:34:25.619 --> 00:34:26.019
Why?
00:34:26.180 --> 00:34:27.139
Why do you have to look through?
00:34:27.300 --> 00:34:28.819
You know, you can do the five Y thing.
00:34:28.899 --> 00:34:30.099
Why do you do that?
00:34:30.340 --> 00:34:32.019
Oh, because they're not in QuickBooks?
00:34:32.099 --> 00:34:32.579
Why not?
00:34:32.739 --> 00:34:34.819
Well, because it blah blah blah.
00:34:34.980 --> 00:34:42.659
And that's when I think you really unearth the gold is you know, to this fictional example of the checks.
00:34:42.899 --> 00:34:49.779
Hey, AI could just every day have a scheduled task that goes and looks to see how many tasks are misclassified.
00:34:49.940 --> 00:34:51.380
It's really good at that.
00:34:51.539 --> 00:34:54.980
Nobody wants to do that, it's the most boring work in the world.
00:34:55.220 --> 00:34:59.539
Okay, there's one tiny task that we've we've chipped away at.
00:34:59.780 --> 00:35:02.500
And I think that's that's an ongoing process, though, right?
00:35:02.659 --> 00:35:17.220
I think that's the key part is that yeah, you don't just set up uh, you know, uh, hey, let's look at how we how we can use AI this week and interview everybody and be done with it in the same way that we're still interviewing our customers and our instructors.
00:35:17.460 --> 00:35:19.220
This is an ongoing process.
00:35:19.460 --> 00:35:22.740
How is AI working on those checks?
00:35:22.820 --> 00:35:24.420
No, it actually created more work for you.
00:35:24.579 --> 00:35:26.180
Okay, well, let's roll that back.
00:35:26.260 --> 00:35:27.380
Let's think about the next thing.
00:35:27.700 --> 00:35:53.699
Considering that that we've got decision makers on this and people who are at all strata of the business, but what are some tools that, as the co-founder and CTO of Coastline Academy, that you found, and it might be a free or a$20 a month, doesn't have to be some earth, earth-shattering kind of um impact, but what are some of the tools that you find yourselves going going to regularly in your own kind of like knowledge work tool
stack that are AI enabled?
00:35:54.099 --> 00:35:54.340
Sure.
00:35:54.500 --> 00:36:00.820
So the one that caught on like wildfire on our management team is Whisperflow.
00:36:01.699 --> 00:36:10.900
And you know, I I know uh it's like I barely touch my keyboard anymore, and I can type really fast.
00:36:11.059 --> 00:36:17.219
Like I just to put it out there, like I'm you know, a top point one percent typer, not anymore.
00:36:17.460 --> 00:36:25.619
Um, and and whisperflow was interesting because I started using it really, really early, and it was immediately impactful.
00:36:25.860 --> 00:36:37.940
And to watch other people resist it because it's ridiculous, like you're telling me I'm not typing my emails anymore, like there's no way, and then they try it, it wildfire.
00:36:38.099 --> 00:36:42.179
I mean, now we're you know, we're on the team plan, we got the leaderboards, we got the whole deal.
00:36:43.139 --> 00:36:45.059
So that's a big one.
00:36:45.300 --> 00:37:06.500
Um, and then kind of a it's not controversial to say this, but what we have found to be really helpful is um when Claude released artifacts, that has been such a nice way for people to share the work that they're doing at that individual level.
00:37:06.659 --> 00:37:11.139
So, not so much at the company system level, but how are you using AI?
00:37:11.460 --> 00:37:41.779
And we've really productionized a lot of that now, where you know, our CEO, for instance, will do an analysis, and now part of his workflow is I'm going to produce the results, I'm gonna make sure that they're live updating, and I'm going to release them as an artifact and share them with the team so that we can see, hey, it basically has become a real-time BI tool, and the adoption of that has been really actually quite surprising.
00:37:41.940 --> 00:37:51.940
Uh, as a CTO, putting my CTO hat on, I'm a little resistant typically to free reign of data because it's a double-edged sword.
00:37:52.019 --> 00:37:59.300
You know, I I don't know how much we can get into this, but it's great that people have data access and it's terrible when they draw the completely wrong conclusions.
00:37:59.699 --> 00:38:16.900
Um, but I have seen that as especially once Fable and Astra came out, the number of poor assumptions as that has decreased, the value of essentially people owning their own BI tools has really gone up.
00:38:16.980 --> 00:38:18.900
And that's been really interesting to watch.
00:38:19.300 --> 00:38:19.699
Yeah.
00:38:19.860 --> 00:38:29.779
So for the listener, Whisperflow is a um a highly capable uh speech to text translator, as Matt was saying.
00:38:29.860 --> 00:38:33.779
He doesn't use his keyboard anymore, he talks to through Whisperflow.
00:38:34.099 --> 00:38:41.300
Um, and the scoreboards that it's showing you, like it's calculating like how much time you're saving by speaking versus typing and those sorts of things.
00:38:41.699 --> 00:38:43.460
Uh, but he kind of referenced it.
00:38:43.619 --> 00:38:50.820
It's a little weird if you've never done it at first, like you've never talked your thoughts out to AI.
00:38:51.059 --> 00:38:52.500
I've seen it happen a lot.
00:38:52.739 --> 00:38:56.420
So, a little bit of advice if you're new to this, first off, try it.
00:38:56.820 --> 00:39:06.659
And yes, it's gonna be a little weird, it might be a little uncomfortable, but keep trying it because at some point, um, I saw this from a programmer, uh, some post on X recently.
00:39:06.739 --> 00:39:09.300
It was like like the human is the constraint.
00:39:09.380 --> 00:39:14.900
The like AI can can cook, but they're waiting on me to input it.
00:39:15.059 --> 00:39:17.300
And it used to be how fast I could type.
00:39:17.380 --> 00:39:24.820
Now, Matt's got an advantage being a uh uh having all that experience coding and all that for all those years, but all of us can talk, right?
00:39:24.900 --> 00:39:26.820
So there's no impediment to you doing that.
00:39:26.980 --> 00:39:38.099
And my little bit of advice would be if it's weird looking at the screen and watching the the voice modulation as you you speak, turn away from the screen and just say, okay, what I really want to do is we've got this problem client.
00:39:38.179 --> 00:39:42.420
And here's the issue it problems kind of us, kind of them, but and just ramble.
00:39:42.579 --> 00:39:46.739
And you'd be surprised with the amount of context that you can give the models simply by rambling.
00:39:46.820 --> 00:39:50.019
It'll pull out the ahs, the ums, the gaps.
00:39:50.099 --> 00:39:52.739
Um, but if you're not using that, I totally agree.
00:39:52.820 --> 00:39:55.619
It's uh um it's a hack.
00:39:55.699 --> 00:40:01.539
We're we're becoming just the the meat proxy, I guess, is the the term that I've heard, right?
00:40:01.779 --> 00:40:02.900
Um yeah.
00:40:03.300 --> 00:40:04.500
So okay.
00:40:05.219 --> 00:40:10.579
Now for the listener, I love actually the what you just said, I think is actually really important.
00:40:11.300 --> 00:40:14.019
The turning away and just talking.
00:40:14.259 --> 00:40:19.059
Yeah, that's something that I I totally did not see coming when I started dictating.
00:40:19.219 --> 00:40:23.539
But there have been many times now where I will start to dictate a problem.
00:40:23.699 --> 00:40:28.259
Like, hey, you know, we have you know, this model is doing that, and I can't quite figure out why that.
00:40:28.340 --> 00:40:35.059
And then I go, oh, and I hit escape, I end the dictation, and I've solved the problem.
00:40:35.300 --> 00:40:45.460
We call it rubber duckying or rubber ducking in the in the engineering world, where you you know, you have a rubber duck on your desk and you explain to it the problem, and yeah, you talk to it, right?
00:40:45.539 --> 00:40:46.659
And then it helps.
00:40:46.900 --> 00:40:48.820
Dictating is the exact same thing.
00:40:48.980 --> 00:40:53.860
The number of problems that I've solved by trying to
explain is really remarkable.
00:40:54.179 --> 00:40:54.579
Yeah.
00:40:54.900 --> 00:40:57.059
Great tip for anybody out there.
00:40:57.219 --> 00:41:03.139
Um, I and I didn't even think about that, but as soon as you started talking, I was like, I know where this is going, and the answer is yes.
00:41:03.300 --> 00:41:13.619
Um, yeah, just verbalizing it and hearing yourself and being forced to externalize it rather than internalize it makes you choose the words.
00:41:13.779 --> 00:41:16.420
And how do I explain this concept or this issue?
00:41:16.579 --> 00:41:18.820
So um fantastic tip.
00:41:18.980 --> 00:41:29.699
Um, and then on the artifact side, um for anybody that's using Claude, that's uh standard now, and uh I guess what's the Chat GPT equivalent?
00:41:29.860 --> 00:41:30.659
Do they have a term?
00:41:31.059 --> 00:41:42.259
Yeah, they have sites, they have sites, but yeah, it the Claude has really, really taken the crown on that, yeah, on that uh dimension.
00:41:42.340 --> 00:41:49.300
I I'm I'm a huge codex user, huge chat GPT user, but artifacts for businesses is just it's unmatched.
00:41:49.699 --> 00:41:51.380
So, listener, I want you to catch that.
00:41:51.619 --> 00:41:53.539
He's not locked into one LLM.
00:41:53.779 --> 00:42:06.579
As you start to use these tools, it's okay to have a Claude account, a Gemini account, maybe a Grok account, and a Chat GPT account, because each of them kind of it becomes your home for a certain type of activity.
00:42:06.739 --> 00:42:08.099
So keep that in mind.
00:42:08.579 --> 00:42:23.779
Um so uh I a question I I guess that um has been coming up a lot because internally um we've been testing with agents, quote unquote agents, since uh early in the year with OpenClaw.
00:42:23.860 --> 00:42:27.860
It was cumbersome, it was too techie for me, and that sort of thing.
00:42:28.019 --> 00:42:38.099
However, there's been Muse released and Grok bot and dots, and now uh just yesterday Claude announced their version.
00:42:38.179 --> 00:42:39.380
Uh what are they calling it?
00:42:39.539 --> 00:42:40.099
Do you remember?
00:42:41.460 --> 00:42:45.139
Yeah, it's uh we just added it to our website.
00:42:45.460 --> 00:42:45.779
Okay.
00:42:47.539 --> 00:42:51.219
But essentially we have a we love your bot section, duet.
00:42:51.539 --> 00:42:52.019
Duet.
00:42:52.820 --> 00:42:59.699
So for the listener, if you haven't messed with these, um, they kind of all start as personal assistant agents, right?
00:43:00.019 --> 00:43:08.739
Um, but very quickly, if you're doing what I call if you're thinking in AI, you start to go, ooh, I wonder if it could, and it can.
00:43:08.900 --> 00:43:14.500
It starts to help with uh a broader scope of personal life, but then it also starts to help in the in the work environment.
00:43:14.659 --> 00:43:24.259
So I think that the the leap from an agent, me, I'm not technical, me using an agent to wow, this thing is indispensable, is is happening.
00:43:24.420 --> 00:43:29.539
Are you guys you you mentioned that you might already be using those at Coastline?
00:43:29.779 --> 00:43:35.460
Um, are you having any experiences internally with teams using these types of agents?
00:43:36.340 --> 00:43:41.619
Yeah, so we also experimented a lot with open claw.
00:43:42.099 --> 00:43:46.659
And despite my technical background, it was so clunky and combosive.
00:43:46.739 --> 00:43:51.219
But everything you just said, uh 100% accurate.
00:43:51.380 --> 00:43:56.980
Um, in fact, we just set one up uh about a month ago to try again, you know, when they released 2.0, yeah.
00:43:57.139 --> 00:44:02.579
I was thinking, okay, this could be our multiplayer uh you know agent for the workplace.
00:44:02.820 --> 00:44:05.860
And uh, you know, I actually just shut it down a few days ago.
00:44:06.099 --> 00:44:25.940
Um, however, the this whole emergence of dots and grokbot and muse, and uh I think that there's so much to it, yeah, but it's very subtle, and I would highly encourage anybody who has not experimented with it yet to just start using pick one, doesn't matter.
00:44:26.099 --> 00:44:28.980
Yeah, uh you know, like Muse is free.
00:44:29.059 --> 00:44:29.380
Yeah.
00:44:29.619 --> 00:44:33.539
Um so you know, that's a and it's really good.
00:44:33.699 --> 00:44:34.019
Yeah.
00:44:34.179 --> 00:44:38.420
Um, I use Grokpa, sorry, Grogbot and Dots primarily.
00:44:39.059 --> 00:44:43.059
Um, I I'll I'll actually I'll tell you that this just happened last night and it blew my mind.
00:44:43.219 --> 00:44:51.619
So I've got dots on, and I was working on a kind of a hairy bug on you know, on an engineering thing.
00:44:51.779 --> 00:44:53.860
And and then it was time to put the kids to bed.
00:44:54.019 --> 00:45:07.219
And so I, you know, stop working on the bug, and and then I get an alert on my phone, and it's my dot saying, Hey, I just noticed that you're working on this, but um, the feature flag is still turned on.
00:45:07.380 --> 00:45:09.219
So people are still hitting this bug.
00:45:09.300 --> 00:45:13.619
Do you want me to turn the feature flag off until you have a time to go?
00:45:13.860 --> 00:45:17.219
I didn't ask it to do it, I didn't even know that it was looking.
00:45:17.460 --> 00:45:17.860
Yeah.
00:45:18.099 --> 00:45:20.900
And then, and I just replied, Oh my gosh, yes.
00:45:21.300 --> 00:45:23.539
And it turned off the feature flag.
00:45:23.779 --> 00:45:28.019
Kids go to bed, I come back, I finish fixing the bug, and I turn it back on.
00:45:28.179 --> 00:45:39.139
And so, of course, that's an engineering example, but I think this proactive AI, the more it has an understanding of your business and your needs and your job.
00:45:39.699 --> 00:45:43.219
In your tasks, I think that there's going to be a lot of potential here.
00:45:43.539 --> 00:45:43.940
Yeah.
00:45:44.099 --> 00:45:53.059
And guys, hate hate to end on such a uh cliffhanger like that because that that conversation right there
needs to happen like on this episode or this podcast.
00:45:53.219 --> 00:46:01.139
That is a that that was a major leap to me, somebody who's been, you know, immersed in at least generative AI usage for three and a half years now.
00:46:01.380 --> 00:46:06.340
Like that to me was a big like, whoa, AI is getting good kind of thing.
00:46:06.500 --> 00:46:09.619
So we'll have some links to those in the show notes and that sort of thing.
00:46:09.779 --> 00:46:17.219
So you know, you guys are doing cool stuff in um an industry again that has a lot of similarities to other industries.
00:46:17.380 --> 00:46:20.099
Are you sharing insights anywhere externally?
00:46:20.259 --> 00:46:24.500
Are you uh producing thought leadership content or anything like that?
00:46:25.219 --> 00:46:34.259
Yeah, I mean, not a ton, but um, you know, what what we have realized is that we as a driving school are at the forefront of a lot of this stuff.
00:46:34.420 --> 00:46:34.659
Yeah.
00:46:34.900 --> 00:46:44.099
You know, the engineering side of me is always like, oh, you know, because we like you mentioned before, we're in this bubble and it feels like it's everybody's doing it and everybody knows it.
00:46:44.179 --> 00:46:52.420
And the more I talk to people outside of this bubble, most people still don't even use AI, let alone are they applying this stuff.
00:46:52.579 --> 00:46:56.099
So, you know, you mentioned Coast.ai earlier.
00:46:56.259 --> 00:47:05.619
That is our corporate website where um by the time this goes to press, there will be some pieces on there for people to digest.
00:47:05.699 --> 00:47:08.739
So if you did want to follow along, that would be the place to go.
00:47:09.059 --> 00:47:09.300
Yep.
00:47:09.380 --> 00:47:13.139
Okay, and we'll have that in the um and the show notes as well.
00:47:13.219 --> 00:47:28.259
Uh Matt, I always enjoy talking to somebody who's like in the trenches like me, and um it's it's hard to like most of your buddies you probably can't call and say, like, man, have you tried or or the the feature flag, you know, proactive suggestion, right?
00:47:28.420 --> 00:47:30.019
To somebody else, they're like, Oh, that's cute.
00:47:30.099 --> 00:47:31.539
And you're like, no, no, no, you don't understand.
00:47:31.619 --> 00:47:36.739
Like this thing is it's truly co-piloting what I'm doing now, which has been a big breakthrough.
00:47:36.900 --> 00:47:39.380
But thank you for sharing the the lessons learned.
00:47:39.539 --> 00:47:47.059
Um, it's exciting to hear when a pivot finally like hits, because um the alternative is is a little rough.
00:47:47.139 --> 00:48:08.659
So I'm glad to hear that Coastline Academy has taken off the way that it has, and that you guys are introducing your your your core purpose of you know eliminating uh automobile accidents into the driving education environment for people who um a little scary to see them behind the wheel uh in some cases.
00:48:08.739 --> 00:48:10.739
So I appreciate all the work that you guys are doing.
00:48:10.980 --> 00:48:17.860
And any closing remarks or anything like that that you think that uh the listener, based on what we've talked about today, should really uh focus on or pay attention to?
00:48:18.980 --> 00:48:21.699
Yeah, I mean, first of all, thanks so much for having me on here.
00:48:22.099 --> 00:48:22.659
Fan of the show.
00:48:22.820 --> 00:48:24.739
This has been great, great conversation.
00:48:24.980 --> 00:48:34.019
Um, yeah, and you know, it's one of your guests recently said nobody's late, or maybe that was you who said nobody's late to AI yet.
00:48:34.259 --> 00:48:36.900
We're all still figuring this out together.
00:48:37.139 --> 00:48:37.539
Yeah.
00:48:37.860 --> 00:48:46.420
And so I think just you know, if you're a business leader, a business business owner, an operations person, just know you haven't missed it yet.
00:48:46.980 --> 00:48:56.259
There's so much to be done, and the the future is, I would say, still yet to be invented around this new paradigm of AI.
00:48:56.900 --> 00:49:01.059
So the more you can use it, the more you can encourage other people to use it.
00:49:01.219 --> 00:49:05.940
We're not late, the party's just getting started, and there's a lot of fun to be had too.
00:49:06.259 --> 00:49:06.739
Love that.
00:49:06.900 --> 00:49:07.300
Love that.
00:49:07.380 --> 00:49:08.579
That's a great way to end the show.
00:49:08.739 --> 00:49:10.340
So thanks everybody for listening.
00:49:10.420 --> 00:49:20.739
Um, and you know, if you've gotten anything out of any of the episodes uh that you've caught of using AI at work, I'd really appreciate it if you would leave a comment or a review on whatever your podcast platform is.
00:49:20.820 --> 00:49:31.219
And if you know somebody who's also kind of uh trying to figure all this stuff out, they're not quite sure where to go, uh share an episode, share the link to the pod uh to the website, uh usingai at work.com.
00:49:31.380 --> 00:49:48.019
And if you haven't been to the site, that's the best place to go to get uh all the resources that we produce related to each episode, uh full listing of things, categorized, all that jazz to really uh become a great free resource for you to continue to have uh interesting insights for people like Matt who are in the thing.
00:49:48.099 --> 00:49:56.579
They're like in the trenches doing the stuff, which is not what you're gonna necessarily get from the the um the 22-year-old on YouTube who's uh vibe coding something.
00:49:56.659 --> 00:49:57.460
Not nothing against that.
00:49:57.539 --> 00:49:58.340
It's awesome to see.
00:49:58.500 --> 00:50:07.300
However, for most of the listeners, um you want a little more seasoning in your advice than um you might get on YouTube.
00:50:07.539 --> 00:50:08.900
So uh please share that with them.
00:50:08.980 --> 00:50:13.860
And again, we'll be back next week with another fantastic episode of Using AI at Work, and thanks to everybody for listening.
00:50:14.019 --> 00:50:16.500
Thanks for tuning in to Using AI at Work.
00:50:16.659 --> 00:50:20.739
Don't forget to subscribe for more conversations about how to use AI at work.
00:50:21.059 --> 00:50:27.539
And a special thank you to our sponsor, Chief AI Officer, for empowering businesses with AI education and training.
00:50:27.699 --> 00:50:34.579
Visit their website for a free AI readiness assessment and AI strategy guide to help you get started using AI at work.
00:50:34.820 --> 00:50:39.139
That's www.chiefaiofficer.com.
00:50:39.460 --> 00:50:51.300
Follow us on Twitter at the handle UsingAI at work and visit www.usingai at work.com for free resources to help you harness AI in your role.