Subscribe
Share
Share
Embed
Two founders talk about how to build software businesses that are meant to last. Each episode includes a deep dive into a different topic related to starting, growing, and sustaining a healthy business.
Tyler (00:01.542)
Hey Rick, what's up?
Rick (00:02.776)
What's up, Tyler? I'm not getting the numbers again. I must be I I just never know when we're gonna start. So this is it always catches me off guard.
Tyler (00:08.859)
Yeah. Okay, I'll take lead from now on. yeah, well we we are recording.
Rick (00:12.814)
Okay. How are you?
Tyler (00:18.075)
Good. You pointed out before I hit record that I'm up in my third floor. I have three different venues you sometimes see me in. One is at work in the office. My normal home office is in the basement. I'm in our like attic right now. that is because my in laws are here, and that is because in about two hours we are heading to the airport and going to France for two weeks, and they're watching Sydney while we're gone. Yeah. not quite two weeks, but something like that, yeah.
Rick (00:25.709)
Mm-hmm.
Rick (00:39.672)
Woohoo That's fun. Two weeks ten days?
Tyler (00:49.431)
it's Thursday to Monday, whatever that comes out to. Yeah. Yep. It's we're going with my whole family. It's my mom's seventieth birthday celebration, so there will be eight of us there total.
Rick (00:51.596)
That is incredible. Wow. Wow. Good for you guys. That's great. good.
Rick (01:03.288)
How are y'all feeling about being apart from Sydney that long? It's probably the longest you've been ap apart.
Tyler (01:07.791)
Yeah, the previously we went to Telluride for three or four days. Shelley's starting to freak out a little bit, yeah. But I I think it'll be good for everybody involved, including Sydney, probably. And she's not old enough yet to she will not be sad about it, I don't think. next year we have to go to not have to, we're going to Slovenia for a family wedding and not bringing her. Then she'll be two and a half. She'll be sad about that.
Rick (01:16.599)
Yeah.
Rick (01:21.72)
Yeah.
Rick (01:36.832)
Yeah, she'll know what's going on there. But maybe she'll like, This is great, grandma, grandpa You never know. Yeah, I mean I w what you're describing sounds amazing. I we have the I think we've we've the most we've done since we've had children is like one or two nights. Yeah.
Tyler (01:38.865)
Yep. Yep. Yep. Just just spoil enough.
Tyler (01:52.057)
yeah. Yeah, it's it's so weird. you know, I would I knew like intellectually prior to having kids, like certain aspects of it's hard, but I you could have never convinced me how liberating it is to wake up and just not have a baby there. Just like I can make noise without bothering anyone. I don't have to check with Shelly on like, is it okay if I shower now or do you need to sh like, do I need to be downstairs right now? Yeah.
Rick (02:05.39)
Yeah.
Rick (02:14.252)
Yeah, there's just so much that revolves around it. It's it's so funny. Yeah.
Tyler (02:19.6)
Yeah, so going to France, very excited, except that so they just had this huge heat wave, like record heat wave. And it's gone now, I think, but it's coming back. So we're we're gonna be there in like hundred degree heat with no air conditioning. So well, I'm a little nervous, but
Rick (02:35.788)
But it'll be fun. I they're the World Cup favorites right now, it seems like, so it should be pretty fun.
Tyler (02:39.398)
I'm a little nervous if it there's a chance I think don't quote me on this, I think there's a chance that the US plays France while we're there. And I don't know French culture enough to know like would we be welcome at a sports bar watching that game or not.
Rick (02:54.038)
My experience is that at least are you are you gonna be in Paris or are you gonna be all over France?
Tyler (02:59.272)
we're we're gonna be in Paris for most of it, and then the second location, but I I'm not sure when the actual game I was just looking at the bracket right before we hit record where like the the the next game is against Belgium. no no, I take
Rick (03:02.51)
Yeah.
Rick (03:14.88)
Yep. If they But yeah, if they win that, then they play the winner I think of like France and Spain or something like that.
Tyler (03:23.556)
Yeah no nope nope nope I'm wrong. I don't think there's any way they play France while we're there. Okay, well that avoids that problem. Anyway, how are you doing?
Rick (03:24.214)
I haven't looked at the bracket.
Rick (03:31.52)
Yeah. I'm good. I'm we're going up to Park City tonight for for we'll see, tonight through Sunday, for the fourth of July. so we're staying in in a in a resort up there. That'll be fun. there's a heated pool, kids love to swim. we'll get out of out of Dodge for a little it'll be great.
Tyler (03:50.534)
Yeah, nice. J just for you said long weekend, like long weekend. Cool.
Rick (03:54.21)
Yeah, long weekend. We'll go up tonight and come back Sunday. See some fireworks. Park City does good fireworks.
Tyler (03:58.951)
See some fireworks. Is this like an outdoors like hiking, mountain biking, or just kinda hanging out at the resort?
Rick (04:05.127)
just a just a summertime ski resort.
Tyler (04:09.05)
Yeah. I I loved I I prefer Park City in the summer to the winter, personally.
Rick (04:14.518)
I won't go that far, but I will say that it is quite nice. the the only the only thing that I'm just trying to prepare Sable for is it can get a little chilly in Park City in July, like if this wind's blowing and the temperature, especially in the mornings and the evenings. And so I'm just trying to prepare for it's sweet, not it's not swimming like it's swimming in Salt Lake City. It's it's there's it's nippy.
Tyler (04:32.444)
Uh-huh.
Tyler (04:38.524)
Yeah, yeah. cool. How about how about work stuff?
Rick (04:44.946)
I'm I I'm coming out of a pretty hard stretch of just like putting my head down and reacting to a bunch of stuff. we have our windfall fiscal year runs from February first to January thirty first. So this is the last month of the second quarter and the first half of the year. And so there's a lot to get done before the quarter is over or the half one is over.
But it's been really good. I I I've I f I've I I worked really hard the past two months to make sure that I have line of sight. And I feel like I can take some take a break this weekend. It was nice.
Tyler (05:24.134)
Yeah. Man, talking to you makes me really appreciate a bootstrapped company where you don't have to care about quarters.
Rick (05:28.654)
Yeah. Yep, I know, I know.
Tyler (05:33.404)
But I wish for you one day when you are running your own business full time, ignore the word quarter.
Rick (05:39.956)
Yeah, yeah, exactly. I'll probably self figure out how to self pressure myself though.
Tyler (05:44.486)
Well, I mean, I think it's mostly a good way to pressure other people, right? Like you hire employees and how do you create a sense of urgency? You say July first's important. I can't tell you why, but i you got Yeah.
Rick (05:53.056)
It's it deadlines. well, tell me about like I I saw your first agenda item. I think it's hap. which I would love to know sort of how I can help with that.
Tyler (06:02.811)
Yeah.
Tyler (06:06.394)
Well, we I don't even think we've said so last episode I said we're going to Hap twenty six, that's Halthapalooza twenty six, and you said, maybe I'll come and I was like, Yeah, yeah fucking right, Rick. Stop overpromising. And now we both have tickets to go to Hap. So we're we'll we'll see each other in in Minneapolis, in about a month.
Rick (06:13.43)
Mm.
Rick (06:22.018)
Yeah.
Rick (06:25.952)
What do do I get to wear some less annoying swag at the conference?
Tyler (06:30.076)
That's good question. I don't so I have not been to a conference in about a decade. Last when so teleport yourself to a decade ago. It's twenty sixteen. Silicon Valley culture has not turned evil yet. People don't hate Elon Musk yet. Wearing a startup t shirt is very well regarded by most people. Is that still true? Like what what does a tech startup, not the word startup, what does a tech company wear to a conference?
Rick (06:56.054)
I mean what I what I depends on the conference. So some conferences have address code. but I w and and I think it also is industry dependent. this is a health insurance is this health insurance or is it health care?
Tyler (07:04.316)
Uh-huh.
It's health insurance and specifically, so my understanding of the landscape of health insurance conferences is there's kind of two types. One is it's like the health insurance carriers, big agencies, FMOs, policy makers, lobbyists, kind of talking about health and sh health care and health insurance like as an industry. And then the other kind is like these are a bunch of actual health insurance agents talking about how to sell health insurance.
We're going to the latter. This is like a bunch of independent, mostly solo health insurance agents. it's hosted by an FMO called Leclair Gro LeClaire Group. And it's it's not entirely, but it's mostly their downline. And sorry for the listener, I'm using all these acronyms. FMOs, again, are field marketing organization. It's basically like the the middleman between the insurance agent and the carrier. so yeah, it's about six hundred
Rick (07:55.16)
Field marketing organization.
Rick (08:02.147)
Yep.
Tyler (08:04.56)
I think there will be about five hundred agents there and they are primarily independent Medicare and ACA primarily.
Rick (08:13.238)
best analogy for what an FMO is is if you've ever bought Cutco Knives, there are these field marketing organizations that are private practices that r are like the central hubs of distribution for cut for cutco. And they basically recruit high school students to sell all their parents and their parents' friends in their neighborhoods. that is an FMO. yeah. I w I was the kid who got recruited.
Tyler (08:33.628)
Yeah, you used to be one of these, right? Not not the FMO, but the the kid that got recruited.
Tyler (08:41.708)
so yeah, it so this will be I think this will be pretty casual, pretty informal. This is not like a bunch of lobbyists in suits or exec there will be very few executives there, right? It's just kind of like normal people who run a part time or full time independent health insurance thing. So I'm guessing casual. I bet most people would be wearing like polos and khakis type outfits.
Rick (08:41.792)
And annoyed everyone in my neighborhood.
Rick (09:05.898)
Yeah, my my my preference in situations like this is if you're selling to the the primary attendee, you're a vendor, it's to take advantage of the branding opportunity and wear swag. I think you can wear swag in a nice way. There's collared shirt swags, there's t shirt swags, you can put a a blazer over a t shirt. you can wear sla nice slacks with a t shirt, tuck it in. I don't know. I know that's something probably Tyler, you're you're against, but you can tuck your shirt in. but no, I would wear I would wear swag.
Tyler (09:29.808)
Not tucking in.
Tyler (09:33.629)
Do you remember my little rebellion against you when we worked together with my dress code? You so you kind of everyone got laid off at the company, and you and I are kind of informally in charge, but really you are in charge. And you say everyone has to start wearing business casual to work every day. and I was just like, no, absolutely. We were roommates at the time. I was just like, no, you're not making me do that. and then we kind of butted heads about it a bit, and then
Rick (09:37.591)
No.
Tyler (10:01.552)
We had some big client come in to visit the office and I was the only one not and I think that day everyone was wearing suits and I was just like, no, not doing it. And they came in and you were showing them around and you were like, This is Tyler's is one of our programmers. And they what you tell me, told me at the time is then they came by and they were like, Wow, he must be good if you let him dress like that. Apparently they thought we had more technical credibility because of that.
Rick (10:24.814)
Yeah. Yeah, it was like the it had the opposite impact I thought it would have. But it's great. Like, okay, like lean into that. no, but I think it's a branding thing. I think I think you gotta take advantage of it. You wanna be walking around and them to see the less annoying brand and ask questions about it and talk about it.
Tyler (10:39.578)
Yeah. I I think that's right. And part of our sponsorship, we get a five minute they call it like a talk, but it's basically a a five minute pitch during lunch the first day, which I like. I've gone to conferences before and we're just like one of a hundred vendors. Maybe people walk by our booth, maybe not. But at this one, like every single person it's a much smaller conference, but every single person will hear our name multiple times. So then when they see it, it'll mean something as opposed to just, that that person has a shirt that's
Has some startup logo on it. so yeah, we probably need to get something printed. because I have like old swag, but nothing, nothing current. We have brochures getting made now. We just got our, you know, like the big backdrop and the standing banner and a tablecloth. So just got all those made. Let me tell you, Claude Design came in pretty clutch for all that. getting a bunch of just generic crappy swag like chip bat chip.
Rick (11:16.59)
Do you have brochures?
Rick (11:32.43)
It's awesome.
Tyler (11:37.169)
b bag clips and pens and you know crap that no one wants, but you're expected to have yeah. but yeah, I'll if we get if we get shirts and stuff, you want one? Okay. Although you should be there. I'm I'm assuming you're there as yourself, right? Okay. Then they're gonna ask you questions about and you're gonna be like, I have no idea.
Rick (11:41.88)
But but they'll take They'll grab
Rick (11:49.464)
Sure. I'd love to wear I'd rep I'd like to rep no, I'd like to rep less annoying CRM.
Rick (12:01.107)
I think I can like what what question is a s insurance agent gonna ask me about a CRM that I'm not gonna be able to answer? Like I w I can an insurance agent comes up and goes, Can you do custom objects on contacts? It's like you don't know what that means. No
Tyler (12:04.653)
yeah. I don't know. I don't know.
Tyler (12:14.544)
They might. They might. Well, that's true. It's not gonna look good when then you're like, by the way, I use pipe drive. But that's okay. great. Did I know that? What do you hate about pipe drive?
Rick (12:20.33)
Yeah, yeah. But I hate it.
Rick (12:30.124)
I've been trying to get off pipe drive for years.
Tyler (12:32.868)
well you want to switch to Hubspot. You hate it for the We are e we're in the we're more like pipe drive that like w the thing you hate about pipe drive you would really hate about less annoying CRM. cool, but yeah, so yeah, you and I are going. We're going a day early. I'm very excited about this. So we are mate meeting up Saturday, spending the day to day together downtown Minneapolis, which I lived for there for a summer doing an internship at General Mills. It'll be fun to see see the place again. And then
Rick (12:38.624)
Yeah, yeah, exact Yeah, yeah. That's fair.
Tyler (13:01.899)
Sunday we're driving out to the suburbs for this conference. yeah.
Rick (13:04.952)
That's all I've actually never been in the summer to Minneapolis. I've heard it's beautiful. I've been I've been in the middle of winter and I like people don't walk outside. so I'm very I'm I'm very excited too. What what what do you think we're gonna do Saturday, Sunday? Like just hang out?
Tyler (13:08.321)
it's great.
Tyler (13:12.166)
Mm-hmm.
Tyler (13:18.746)
I booked so hang out. yeah, we both get in around noon. Most cities and Minneapolis is like a big downtown. It's not a huge metro area, but it's a big, dense downtown. most cities like that would have a bunch of competing nice restaurants fighting for the number one spot. If you Google what's the nicest restaurant in Minneapolis, everyone's just like, there's a single answer to that. It's this one. I forget what it's called. We're going to that that night.
Rick (13:42.222)
Nice.
Tyler (13:46.485)
And otherwise I don't have anything planned, so we'll just, you know bounce around.
Rick (13:50.934)
Sweet. Drink drink some what do you call what are those heavy beers? Take sevens?
Tyler (13:56.839)
Tank sevens what what made us forget that previous night that we were hanging out. so yeah, I'm really looking forward to it. We're also I'm not I don't have a lot to say about HIPAA, but like all our banners and everything for the conference are really playing into the HIPAA compliant thing. So this is we were talking about the end of quarter pressure that a a company can put on people. This is the first time in a long time we've had like a real deadline. Like we have to be HIPAA compliant by whatever it is, July twenty sixth, because if we're not, all of those banners are lying.
Rick (14:00.49)
Yes, exactly.
Rick (14:27.427)
That's awesome. So you're doing your own version of quarters. it's an event.
Tyler (14:31.236)
Yep. Yep. so I'm about to leave for two weeks. I've assigned a bunch of stuff to the team to get done, hopefully during that time. And then when I come back I need to I I'm spending the next I've got a a week and a half or something after that that I need to like get the BAA drafted. Get I've got a lot of work to do when I get back from France.
Rick (14:51.052)
Very interesting. I'm I'm so excited about this. I I I don't have any I don't have any expectations other than to get to spend time with you and hopefully learn a little bit about the industry, the agent insurance agent industry, and also help less annoying CRM break through this target vertical. Yeah. So if there's anything I can prepare, let me know.
Tyler (15:07.932)
Breakthrough. All right. Well, I appreciate that. yeah, it'll be fun. So Okay. Yeah, I'm not even in that mode right now, but I will I will do that. Thank you. I I think we're 'cause we're bringing three people. Like it's me, Eunice, and Ruth are all going. So we're we're gonna be rolling pretty deep there.
Rick (15:24.908)
Nice. Nice. I've never met Eunice, I don't think.
Tyler (15:30.34)
Yeah, probably probably not.
Rick (15:31.244)
Maybe at your maybe at your party, but like not dead in not in a work a work setting. Ruth, I've met. I believe.
Tyler (15:35.28)
Yeah, yeah.
Tyler (15:39.484)
Okay. I don't remember. But yeah, cool. all right. do you have other stuff to talk about or should I keep going?
Rick (15:47.97)
Well I gotta I told you I'd update you on Guru. basically put that on ice. Claude has really gotten b have you played with the new Claude Slack integration?
Tyler (15:50.854)
Yeah.
Tyler (15:58.041)
Not yet. So so just as a re to make sure I'm re guru is the tool you were using to try and like pull all this information from the company together and then like ask AI a question and it can answer it.
Rick (16:06.892)
Yeah, so one of the challenges that I think a lot of companies face is, especially if you're smaller, g generally you're in this like no man's land at some at some place where my knowledge lives in people's heads. And so Claude, and there may have been attempts to document, but like it's sparse, it's outdated, it's really hard to keep update. So Claude really struggles with governing itself on like what is and allowing you to govern it as a as a manager of it, on what the con what context is good and what context is bad.
Tyler (16:19.804)
huh.
Rick (16:36.704)
across an organization. I think it's very, very good at my instance of Claude. Like I'm going to give you the context, but what I don't know, it doesn't really know. and so what Guru is doing, it's get guru.com, I think, or I.ai, is they're basically saying, listen, this is a huge problem. Claude's not good at it. and you you need a a middle layer, basically a context layer that
Tyler (16:47.132)
Mm-hmm.
Rick (17:03.104)
you can ingest all of your various sources of context, connect claude to, connect all your other tools to, and or like basically monitor what context is being used for what and when it's and when it's being used correctly, give it positive feedback. And when it's being used incorrectly, give it negative feedback. And then use that to continuously improve the organizational context you have. Anyway, that's the gist of guru. We we decided to put it on hold primarily because we our organizational context just we don't feel like it's
Quite where we want it to be. And then second is Claude's gotten a lot better. They in they updated their Slack integration. Have you played with this at all?
Tyler (17:39.662)
I read about it but I haven't used it now.
Rick (17:41.356)
it's incredible. Like you can add Slack to a channel and it's way smarter now. Like it can ac actually answer questions with all of its connected tools. You can build and it's and anyone can a ask it questions whether they have a Claude license or not. So it's almost like getting access to Claude without having to give people seats. So like the
Tyler (17:57.962)
I hadn't thought about that. Everyone at Less Annoying already has a clot account, so I that is a problem I had not considered, but yeah, that makes a lot of sense.
Rick (18:04.632)
So but yeah, you can add to a channel and be like, Claude, can you answer this question for me?
Tyler (18:08.218)
Yeah. Now, having said that, seat-based pricing for Claude is probably going away pretty soon, so that pricing benefit will probably not matter, but yeah. But you think the middle I feel like in general, the AI models, even as much as everyone's talking about them getting better and better, and maybe Fable I haven't used Fable enough to say, but prior to Fable, like Opus four point eight, I c I I feel like they haven't actually gotten that much better this year. But the the agent layer, the harness, whatever you want to call it.
Rick (18:13.966)
It's usage based. Yeah. Yes. It's
Tyler (18:36.784)
For clawed code, for cowork, and it sounds like you're saying that this Slack integration has some harness or something that is doing a lot of work beyond just what the model can do.
Rick (18:44.376)
And correct. And the reason that matters is that Slack channels have co their own context. and so if if you and I I I actually do believe we naturally as humans create Slack channels with context in mind, like and it just happens naturally. Now there's dead Slack channels all over Slack, but like the Slack channels that you want Slack cla clawed in have context. And so it actually works. So anyway, that's the long story is like
Tyler (19:00.476)
Mm-hmm.
Rick (19:12.012)
I'm kind of like putting put punting that and and and trying to use Claude and other tools like Slackbutt.
Tyler (19:17.692)
Cool. I should maybe on this topic give an update. Cause so yeah, last episode we talked a lot about how do you create all the knowledge and then especially that's easy. You make a bunch of documentation, it ingests it, and then how do you like make it better? I did after that conversation, because I always feel like such an imposter. We have these conversations about AI and neither of us really know what we're talking about. But I did kind of try to learn a little bit more about that. So I thought maybe I'd just like give a quick recap of my understanding of this now.
Rick (19:44.076)
Understanding of what specifically?
Tyler (19:45.737)
Just like how that works. So so, okay. The the the key thing I feel like I didn't understand is Claude has a one million token context window or whatever. It's very hard right. So it's very hard to reason about what that means. That's the size of I'm probably so caveat, everything I'm about to say, I think is more correct than it was last episode and still probably very wrong. So if you're a listener that knows more about this than me and you're rolling your eyes, apologies. But so one token is something like
Rick (19:54.894)
What does that mean?
Yeah.
Rick (20:09.439)
Ha ha
Tyler (20:13.84)
three or four characters, I think, like letters, but it d it just depends on how
Rick (20:16.279)
tokens represent units of of text.
Tyler (20:20.644)
Yeah. I think it's tokens, not letters, because like it depends on the encoding and all like and it's not not LLMs aren't always text. Sometimes they're doing images, right? But it's like a p a a piece of data and like the billing, we're both on subscription billing, but like the thing they're pushing us towards is like every token of input and every token of output is gonna cost it's like five dollars per million tokens or whatever, right? the context window is basically saying this is the amount of like tokens that the
LLM can kind of consider at one time when generating a response to you. And when you're going back and forth with an LLM, the UI makes it seem like, okay, I say hi, it says hello back to me. And then I say, what's your name? And then it makes up a name and sends that back to me. And each one of these is a separate message. That's not how it works. And I already knew this part. The first message is hi. The next message is hello. The next quest message is hi, hello, what is your name?
Every message you send to it, you're sending the entire conversation you've had previously. So the number of tokens used by each of these requests goes up something like exponentially. Like it doesn't feel that way 'cause you're like, I just asked a one sent Yeah. Yeah.
Rick (21:31.988)
Sorry, the longer you chat, can I just like s state this in dummy terms like I am? the the longer you carry on the same chat in the same context window, the the more expensive every response becomes. Or you every response that you send in becomes.
Tyler (21:45.191)
Yes. Yes. And it's because of the input tokens, not the output tokens. so basically so the context w so the the way the UI makes it seem is that the the LLM is like alive and remembering this conversation. And when you send a new message, it's like I already have all the context of the previous message. It doesn't. It's like dying every time. And when you send a new message, you're sending
Rick (21:51.0)
Makes sense.
Rick (22:07.118)
Mm-hmm.
Tyler (22:12.912)
the full history of the whole conversation because it has no memory of what the conversation would be otherwise. At least that's my under again, maybe I'm getting things wrong here, but I think that's basically correct. so yes, that that's point one. As the conversation gets longer and longer, it gets way more expensive in terms of token usage, even if you keep your messages short.
Rick (22:31.544)
This is super helpful. what does this mean? Like like what what does this unlock for you in terms of how to use Claude or like like I guess like okay. What should I change? Okay.
Tyler (22:40.922)
Yeah. So I I I have a different point later cause this is stuff that I learned like a while ago, but with this, I think the takeaway is there's a thing called context management. So the context is basically saying how big of a message can you send to Claude before it's just like that's too big, I can't handle it anymore. and the answer is one million tokens. That's its context window. There's more to it where like people hypothesize that it gets dumber the bigger the more context you're sending it, but like
Rick (22:46.968)
Okay.
Tyler (23:08.612)
Once you get above a million tokens, it has to what does it call it? Compress it or that's not what they call it, whatever. They it basically says, I'm gonna take this whole conversation and I'm going to have the LLM summarize it into a shorter conversation. And so now every time you send something back, it's only sending the shorter version because I can't remember the whole thing at once. Well, hallucinations can happen anyway, but that's when it forgets your previous stuff. Like, yeah, early in the conversation, you're like, I want to be clear, every time
Rick (23:26.22)
And that's where a hallucination can happen.
Well that but that's where like they could if I get to the wrong stuff.
Tyler (23:38.501)
You edit Notion. I do this to it all the time. Every time you edit Notion, do line by line edits. Don't clobber the whole document, because I might have made other changes. And then you talk for like 30 minutes and then it clobbers the whole document. It's like, dude, what happened to that? And it's like, well, we compressed the context. I forgot that thing you told me, you know. and then like the the real power users of this aren't letting Claude compress it. They are doing it themselves. So they're saying, I'm at 200,000 tokens. This is about where I think it gets stupid.
Rick (23:40.227)
Yeah.
Rick (23:50.686)
Okay, yeah. Yeah, yeah, that's helpful. Thank you.
Rick (24:02.414)
Mm.
Tyler (24:07.706)
So now I'm gonna write a document that summarizes where we're at and start a new session. that's what like the power clawed code users are doing. Or they're probably doing way more than that, but
Rick (24:17.686)
And and and and this is getting into skills to to separate this out from the context window.
Tyler (24:24.198)
Do you mean like claude skills? A claude skill's just a prompt.
Rick (24:25.612)
Yes.
Rick (24:29.506)
But it doesn't have context in it.
Tyler (24:32.464)
I don't think the skill I think the skill is basically just a pre safed prompt, more or less. I yeah, who knows what's going on behind the scenes, but that's how it looks. It looks like it's just a markdown file.
Rick (24:43.02)
Yeah. Yes, college skills contain context because they use progressive loading stuff so they don't blow up your entire conference with unnecessary background information. So apparently it's it's like it's like on demand context, I guess is what yeah. Yes. But like does this mean does this mean like if you want to do a the way what I'm starting I'm I'm still very much in like a human in the loop AI workflow type thing. I'm not doing a ton of autonomous stuff outside of outbound
Tyler (24:53.26)
Okay, I don't know what that means. Yeah, yeah. It does it only pulls the skill in if you run it, yeah. so okay.
Rick (25:12.734)
email. and that's usually managed by a separate tool and they're do they're managing all this token usage on their end. I'm I'm more just saying I'm paying you a subscription. You figure this out. for Claude, if you're doing your own sort of multi-step workflows, it does this mean you need multiple different sort of touch points where you're pa passing context from one, I don't know, sit I don't I don't I'm calling it situation, but like it's not a chat. It's a
Tyler (25:19.579)
Mm-hmm.
Tyler (25:41.789)
I mean it's kind of a chat. We can call it a chat. Yeah, so I think like the like one takeaway here is like I I've been using Claude Cowork a lot versus code because I'm working on HIPAA where the the output is not code. yeah, you I probably have 30 different HIPAA conversations because I'm like I've I've maxed out the context window or I'm whatever. I'm I feel like I've I've done more than I should have in this conversation. And then I have it write a markdown file, and then I start a new chat and say, Hey, here's the markdown file.
Rick (25:43.192)
Context window.
Tyler (26:11.226)
Load that in to context and then let's let's keep going. Which again it it's basically the type of compression it would do on its own, but somewhat. I mean, I'm still leaning heavily on the AI for doing the actual summary, but yeah.
Rick (26:18.434)
But you're controlling it.
Rick (26:23.66)
Yeah, but but you're validating it. Like you're you're you're creating a human in the loop validation of the the right context being compressed and that's that's smart. That's super smart.
Tyler (26:28.602)
Yeah. Or or like here's another example. You're having a conversation and you're like, I just have a different question that doesn't need all this context. Go into just like normal clawed chat and be like, Hey, does HIPAA actually require this? It doesn't need to know all of the details of everything else I wrote, you know? Yeah. Yeah.
Rick (26:45.036)
yeah, yeah. And you save tokens by doing that. I had a thought. How how many skills have you created?
Tyler (26:54.79)
Personally none. our team has created them and I've like looked at them but and used them.
Rick (26:59.658)
okay, so first of all, like I I just want to take a step back. Anyone who's listening to this podcast who knows more than us is probably like rolling their eyes and so frustrated. But anyone who knows less than us is probably like this is so helpful. It's what this is a very polarizing conversation because but it's very this conversation is helpful to me because I am running a clawed co a co-work not a claw co-work, but a claw session for new users at Windfall later today. And I've got to figure out the best way to go get someone from zero to one on Claude.
Tyler (27:07.824)
Yeah.
Tyler (27:12.348)
Yeah.
Tyler (27:23.846)
Mm.
Rick (27:29.658)
in an hour. Now, now Claude at our in our organization is connected to everything. Like it's connected to Confluence, Atla Atlassian Confluence, Atlassian, R Rovo and Jira. It's connected to Slack. It's connected to Drive. It's connected to Salesforce, Gong, all of our call record. Like it is very powerful. but like where would you start?
Tyler (27:49.978)
Yeah.
Tyler (27:55.833)
I don't know. I mean I dun yeah, so when you say it's connected to that, like they just get automatically like they have to sign like they have to sign in for the first time and then they have to connect it to this stuff?
Rick (27:55.918)
Yeah.
Rick (28:07.734)
Yes. So so the organiz I'm an admin. You're probably an admin. At the admin level, you set up what connectors are approved and you sort of pre-configure them. Some of them require custom MCP servers and like like for example, Salesforce and Gong are in that boat. and others are sort of pre-packaged connectors by the tool. generally you have to connect them and re- like kind of OAuth in as an admin to unlock them or pre-configure them as an admin and then
Tyler (28:10.78)
Mm-hmm.
Tyler (28:22.63)
Mm-hmm.
Rick (28:37.174)
Each user has to install the connector into their own clawed instance and OAuth in. So like Yeah. So and then you know, some people don't realize that Alassian Robo is confluence and you're like, wow, you you are you are missing a very important piece of this puzzle. So
Tyler (28:39.676)
Jeez. We're still living in the Stone Ages, man.
Tyler (28:49.478)
Right.
Tyler (28:54.044)
I mean, I might go even further, like there's probably not enough time to really prep this perfectly, but in my experience, just looking at the people at lessoning CRM, there's a big gap in like even understanding just thinking, is this a problem AI can help with? I think the typical person just thinks, I know how Google works. I know when I would go Google something, I'll use AI instead, and I'll maybe get a more specific answer, which is like
Just scratching the surface of the things you can use it for. That's to me the biggest unlock I've had is just being like and I I don't I wish I had a good example in the moment. but like here are some some cases where someone who's not like thinking like an AI user wouldn't think to use it, but like like don't just give it a query, get an answer, and end of conversation, you know?
Rick (29:42.776)
So the way I'm thinking, tell me if this works from your perspective and then we can move off of this topic. I just want to make sure like I'd love to get your input on I have an hour with let's just call it 10 people who have never used Claude before. They've been on Gemini, so they have AI use, you know, sort of property use cases and they have workflows built on Gemini gyms. but they're very they're they're they're not anywhere as near as powerful as Claude can be with the connectivities that we have. Yes.
Tyler (29:50.554)
Yeah.
Tyler (30:06.652)
Sorry, can I interrupt and ask one question? What what tier of claw do they have? Like what kind of but like are they on the subscription pricing or API pricing? Twenty dollars, hundred dollars?
Rick (30:10.808)
To work.
Rick (30:15.469)
Subscription.
I think it's a mix. but none of them are they just got access. Let's say they they haven't had it the most of probably have not even prompted.
Tyler (30:20.015)
Okay.
Okay. I I say that 'cause I say that 'cause you go through the twenty dollar one so fast. So for twenty dollar people you've got to teach about context management, I think.
Rick (30:32.886)
Yeah, okay. I see your point. I see your point. Okay. Here's my here's my thought on that. Like, I don't think I would want to bring that into the first hour. I I would rather them hit the context limit and then do the session on that. so right now, like I think there's two outcomes for the session. One is everyone has Claude installed on their desktop and properly configured for purposes of leveraging the connected ecosystem of context that we have and tooling.
Tyler (30:42.204)
Mm-hmm.
Yeah, okay.
Rick (30:59.458)
They have they understand how to bring Slack into Claude into a channel in Slack and leverage it in context there. And they have like any, they have a path or like maybe in the session, we've actually moved some of their Gemini-based workflows to Claude, either as skills or as just like you know, a a a project of some kind. And then and then the next things are okay, let's actually do additional sessions where we take more.
Tyler (31:16.027)
Yeah.
Tyler (31:19.644)
Yeah.
Rick (31:27.752)
concrete workflows that didn't exist in Gemini and pull them into cloud. Like there's there's there's i instances in which people are logging into one system on a browser, extracting information from that system and going into another system and putting information in that system. That's something cloud cowork can
Tyler (31:43.023)
Mm-hmm. Yeah, I feel a little out of my element here because like this is such a a big company thing. I know in the grand scheme of things, Windfall's a small company, but this is just so different from how I've ever used it. Like that this like it is a corporate thing with all these connections. I I just I don't think I have anything to offer here, to be honest. Sorry. But I will say this.
Rick (32:06.306)
All good.
Tyler (32:08.772)
I think a useful thing to say, especially at a company like yours where it's high growth, probably lots of really ambitious people working there, is just saying like, what I'm about to show you we didn't have a month ago, and probably is not going to be how we use it a month from now. And there's tremendous career growth opportunity to lead rather than follow here. So you can follow what I'm about to show you, but the real opportunity is to figure this shit out for yourself and tell me how to use it. Yeah.
Rick (32:32.898)
And push us, push us forward. Like and like ask for more. reach the limits of what this can do. Don't wait for someone else to show you. Yeah, that's helpful. No, but it's a it's an overarching theme. Like we I I it goes back to like I w I just want these people to use I would like like you you like teaching about context, but you're like, No, I want them to blow the context out. Like that's that would be success. cool.
Tyler (32:42.18)
Yeah. But sorry, that's kind of wishy washy. Okay. C
Tyler (32:53.636)
Yeah, yeah, that that makes sense. That makes sense. okay, I want to go back to why I started talking about this, the what that led us to context windows and token management and all that, was basically say, how does a tool work that wants to ingest all this information, like a knowledge base, and then answer questions based on it? Because a million tokens is not it's a lot for a conversation. Like you can have a pretty long conversation with a million tokens, but it's not enough to ingest
our entire wiki, or like certainly not even close to enough to ingest all of the the support emails we've ever had. so I learned a little bit more about how that works. And maybe this is obvious, but like basically there's like you said, there's some middle layer that it ingests all of this data and then it it basically indexes it in some format. And I think the most common way to do this is a vector database. I've never used a vector database, but my understanding is
Whereas a normal database has it's like, get row number 72 and give me the data in that. A vector database has some sort of more like it's like word association. Like if you Google insurance agent and it might match with the term health broker. Or, you know, like it it's not a letter for letter match, it's some kind of context. I I I don't understand how it works, but that's what a vector database can do, I guess. So you take all of this information, throw it in a vector database.
And then you when you say to AI, Hey, I need I I have a question, here's the question, it then queries the vector database, gets all the related information, sucks that into the context of the LLM, and then answers your question. But the LLM does not understand like all the stuff in the vector database. It's just getting this tiny little sliver, which is what makes this so hard, is that tiny little sliver is not enough.
Rick (34:45.836)
And I'm just wondering like who is constructing the vector database relationship so that it's getting the right stuff? Like is that is that some human or is that another tool? Like I guess I'm sh I'm struggling with like how how the vector database is properly configured.
Tyler (34:55.813)
Yeah.
I think it mostly is magic and automatic. I'm sure like if you're there there is someone on earth who would know how to turn the knobs properly, but I think a typical all these AI startups that are so first of all, they're all using Opus, like they're using an anthropic or an open AI or one of a small number of these open weight models. Like they're all using the same models, right? There's nothing proprietary or special about the models at all. And then they're building their own
Rick (35:01.783)
Okay.
Tyler (35:28.336)
harness agent, middle layer, whatever. And they're almost all using again, there's like a handful of these vector databases, but let's to simplify and say they're all using the same vector database. So it's the code in between the LLM and the vector database that is what the startup is building. I think they have limited ability to c this is one of the the reasons AI companies keep failing us, I think, is you can't actually control the LLM and you're not, unless you're really, really advanced, you're not actually building your own vector database or whatever else you might
whatever data model you might use to store this. So the the actual surface area that they can control is very, very limited.
I think. This is again I I feel like I'm still just learning this stuff, but
Rick (36:11.51)
If you're listening to this and you're like, I want to start a business today, like what what is your current advice? Like like you you mentioned, like, hey, I hope one day you're running a bootstrap company and your life is better, Rick. that's basically what you said at the beginning of this podcast. And I'm like, like what what do you do? Like, what is the best bootstrap business idea today?
Tyler (36:27.772)
Fair enough.
Tyler (36:34.594)
Ugh. It's it's start an AI company and sell it before everyone realizes it's bullshit.
Rick (36:39.726)
but like no I'm serious though. Like you would you is it the same rules that apply? You just go into a an i an industry that's niche and high yeah.
Tyler (36:41.5)
Yeah.
Tyler (36:52.858)
Maybe I think my answer here is productize service. have some human element that i it's not pure software, because pure software, I mean I I'm I don't buy into all of the like it's over, we don't need human coders anymore. We still need human coders, but that used to be a big moat. Like when less annoying CM started, it's like, okay, yes, there's marketing, there's sales, there's design and all this, but also I can in two thousand nine, not many people could build a CRM and I could.
Which gave me a big advantage. One of my big advantages is gone. It was already hard without that advantage. It's even harder. Or sorry, it was it was hard with the advantage. I'm not saying it can't work. I'm just saying it's even harder. I'd say bring some domain knowledge for something outside of tech, add a productized service, something where AI can't just do it. A human needs to do it, but then superpower yourself with software. That which is what Leg Up Health is doing. That feels like the path to me.
Rick (37:48.416)
Mm. And I'm curious, like, would that human that whatever the human is doing, would they would you say that they are like do you have to have that human talent in house or could you outsource it to a IBM like a business process outsource company?
Tyler (38:06.958)
I I've you probably know more about this than me. My my instinct is I've s I b back in twenty twelve when I was trying to like be a startup thought leader, a thing I said a lot that I believe wholeheartedly is like you should not start a business where one of the core competencies is not like a founder should be an expert at every core competency. And so if you're saying
Rick (38:25.218)
But what if the core competency here is the productization of the labor, not the labor itself?
Tyler (38:30.874)
Yeah. I mean you could argue that's what I'm doing, sort of. Like I don't know shit about CRMs. I just I know about design and and coding, but I think we're an outlier in that regard. I think normally when you see this is a CRM for health insurance agents, one of the founders was a health insurance agent. I look at what you and J D do in Leg Up Health. If it was me and let's say JD, but he's not an employee, he's like a contractor or something. Like he's not a founder level employee.
Rick (38:33.079)
Yeah.
Rick (38:37.134)
Mm.
Rick (38:48.097)
Yeah.
Tyler (39:01.958)
That business would suck, I think.
Rick (39:03.468)
Yeah. Yep. I agree. Just just want to call out that you did have domain expertise before you started lessoning CRM and serving insurance agents and building a CRM for insurance agents.
Tyler (39:14.095)
yeah, yeah. But I sorry, doma domain expertise primarily would mean sales for me, I guess. Like
Rick (39:19.335)
I see, I see. Like being the agent.
Tyler (39:21.956)
Yeah, we talk to people on the phone all the time. And like, or like what would a clo a typical kind of tofu mofu bofu funnel, like content funnel look like? The top of the funnel content is like, here's how So a perfect example to look at, go look at the content for close dot com. They're another bootstrap CRM. Stelly FD, I think his name is, is their founder. He was a sales guy. And all of the they just obviously get what sales is in a way that I don't. He's always talking about stuff and using terms, and I'm like, I didn't know that was a term.
And he's like, Don't you hate it when blah, blah, blah. And I'm like, I didn't know that anyone ever experienced that. And like, they have a hundred employees and we have seventeen. Like they're they're better than us 'cause they actually know what salespeople need.
Rick (40:02.894)
I don't know if they're better that well.
Tyler (40:03.868)
I mean it's true.
Sorry, so they're more successful than us. I we're better for a certain type of person whose brain works the way my brain works, but like I'm doing what you're not supposed to do, which is it's pure intu well, sorry, now I talk to customers, but like in the early days it was pure intuition. I was just like, here's how I think a pipeline should work. And most people don't think that that's how a pipeline should work, but like one percent think think the same way I do and they like our product.
Rick (40:30.766)
It's awesome.
Tyler (40:32.334)
Anyway, sorry, I'm I'm on a soapbox, but yeah, what do you think? Like if you had to start a somewhat techie company, like what where's the opportunity out there?
Rick (40:42.21)
Well that's what I don't know. I just it it seems like I I feel like I don't know. If I I'm asking because I really actually don't know. I think leg up health is an aspect of this where there's like re regulation. I it high regulate like regul industries that are regulated is probably something I would look at, but that also brings pain and suffering that you are introducing into your life. yeah.
Tyler (40:59.1)
Mm, that's a good point.
Tyler (41:06.428)
Mm-hmm.
that's good. Right? We we've said before, every every business has to do something hard. It's like what hard thing do you want to do? And if you're if it's a regulated industry, immediately you've got a hard thing that no one else wants to do.
Rick (41:14.334)
Mm-hmm. Yeah.
Rick (41:20.588)
Yep. And I but yeah. And like take for example, I I just had a realization a couple days ago. Take our our free leg up starter health starter product, which is basically a spreadsheet for managing employees and built into a UI and giving them some amount of money per month, and they have a place they can shop. I could probably cloud code that this weekend. So could every emp company that we serve. They could build their own version of this. So it's like
Tyler (41:43.322)
Yeah. Yeah.
Rick (41:49.79)
Okay, what do we build? well it's the service around it's it's actually yeah it's the service around it.
Tyler (41:52.549)
Well, but what leg up help yeah, exactly. It's that JD is the one clicking on the buttons, not the customer. Which kind of pla yeah, so there's there's two s there's two types of productized service, it's maybe worth saying. One is like we get so I use bench I think bench dock or or no, I don't use bench anymore. They've flopped. I use pilot for bookkeeping.
But let's use Bench. They're actually a better example. They had their own proprietary software for like bookkeeping. So we didn't use QuickBooks. we logged into Bench's software. But then they had like bookkeepers behind the scenes doing stuff. So that's one version of it where like we're using software, but then the software is being enhanced by an army of people on the other side. But the other type is I'm I'm interacting directly with the human, and then the human has a bunch of tools that they use. Like let's say you're a marketing agency and you're like, y you know.
Rick (42:28.333)
Mm-hmm.
Tyler (42:45.914)
Five years ago we would have had to manually do this all by hand and we would have had to charge you ten times as much. But now we'll do the exact same work. You'll interface with us in the exact same way, but we're ten times as productive as productive 'cause of all these tools we've got. And you just pay less. there's probably like a spectrum between these two, but I think that second one is an especially interesting version. It's not like I've I've always said this about like up health. The the customers don't want to log in and manage anything. Like they just want to text J D and and tell what to do, I think.
Rick (43:13.43)
No, the pilot analogy is really interesting. And the yeah, this is very interesting. Thank you. I don't think I've seen that before, that analogy between what we're doing and outsourced bookkeeping. It's very similar, very similar, like very, very similar.
Tyler (43:21.936)
Yeah.
Tyler (43:32.944)
The other analogy we've used in the past, I forget what the company was called, but it was it was like a security company that it's like ring doorbell cameras, except there's a human on the other end, right? Sentinel, yeah.
Rick (43:39.478)
Yeah, yeah, yeah. Sent Sentinel. Sentinel, yeah. It's yeah, it's it's these are good examples. I need to come back to this. I think these are all really good examples.
Tyler (43:50.279)
What I wonder is like how far does it go? Like we the the common industry that's often said is AI proof is being a plumber. A plumber with custom software behind them, is that a better plumbing business or does that not even matter?
Maybe only a little.
Rick (44:06.508)
I mean, only to the extent that it gives them some sort of advantage. maybe it's, you know, g get at the end of the day, if a plumber is quick to answer my call and fix the problem, I don't really care how they do it, you know.
Tyler (44:18.522)
Yeah. But it are could could you make a plumber more efficient with like a bunch of custom software and and the answer's probably not. Like they they have to drive to the house and to pull out a wrench and probably software can't make that more efficient. But bookkeeping absolutely can be, etcetera.
Rick (44:34.412)
Yeah. I agree. I I buy that.
Tyler (44:38.384)
Okay. anyway, what started all this was just me saying having we like to think that AI has in its head all this knowledge and it doesn't. Every AI conversation, it only has a tiny sliver of knowledge. Getting it the right sliver so that it can answer the question is extremely challenging. And I d I do not personally believe that is a solved problem. I think certain companies are doing it better than others. I told you we use Resolve twenty four seven, which I think is doing it relatively well, but like
To me, this is still a a lot of people are like LLMs will hit a natural seal ceiling where they just can't get better. I'm not saying that that's definitely case, but like I think the whole like ingest everything and then give me the right answer to my question is much, much harder and further out of reach than people acknowledge.
Rick (45:25.206)
Yep. And that is fundamentally what just going back to tying it to our original conversation with my update is guru that is what guru is claiming to be able to do. and we I we don't I I would say that my my assessment is I don't feel confident that they can do it. And it would require so much effort on my part to do it that it's not worth the investment.
Tyler (45:33.116)
Mm-hmm.
Tyler (45:40.038)
Yeah, yeah.
Tyler (45:45.543)
Sounds right to me. Okay. we're doing a Claude security audit right now. We've been doing this for a little bit, but let me add a little backstory here. So most software companies have this where basically random outside they call themselves security researchers, find bugs, not not bugs, but like security vulnerabilities, and email them to the company, and then the company pays them a bounty. So like we we typically pay
Rick (45:46.562)
Yeah. What else is on your list?
Tyler (46:16.326)
Fifty to two hundred dollars if someone sends us like a minor thing, we'll pay. I think the highest we've ever paid is maybe five thousand dollars, I think. so first of all, we've been getting more high quality reports lately. Like normally these emails are just these these reports are just sp spam, basically. It's someone ran some automated tool that found something that's not even a Vuln. They didn't bother to check it, they just emailed us. pretty rare to get a good one. We've been getting good ones lately.
Almost certainly because AI, like these these people are all using the same AI tool. I don't know what tool it is. again, same LLM behind the scenes. I'm sure it's Claude something, but is it just using Claude or is it some like layer on top of that? I don't know. Anyway, in addition to that, a month or two ago we started pulled one of our developers off and just said, Can you build a internal tool that basically
Has Claude over and over and over and over again scan our code from different angles, like start at this entry point, start with this mindset, right? Trying to give it all these different prompts to just find every vulnerability it can possibly find. A, it's been going great. We found well I don't want to make it sound we found so much like wow, our code was so insecure before. But you know, there's all these little edge cases, and it finds them. And B, it finds the same things that these high-quality reporters are.
Rick (47:33.24)
Yeah, yeah.
Tyler (47:40.935)
Sending in. which is an interesting part of this because like there's only a small part of what I'm saying, but it seems to me that in the future there's no reason like if a security researcher ever finds anything, like how? They have the same AI we have, except we have access to our code and they hopefully don't. So like it should be asymmetric. We should just be much, much better at finding things than they are.
Rick (48:05.918)
Every time they send you something should you should say, We know. We're we already know. Yeah. We already know and we and you're you're you're too late.
Tyler (48:08.826)
Yeah, yeah, exactly. I mean it should be fixed, hopefully. Cause yeah, they're Yeah, because they're doing external they're they're just like looking at endpoints and then trying things. They can't look at the code and be like, I can see there's a problem here. They just have to like trial and error. This is called a pen test typically. so yes, A, I'm hoping like while I appreciate these reports from people, I th that feels like a a dying industry, especially if they're relying on the same AI
tools that everyone else has. but B, like, I don't part of me is scared, like attackers are gonna have this and, you know, the attackers might get access to Mythos before we get it, and then they have an asymmetric advantage temporarily. But like I think I'm more optimistic than I am worried that just something like like I've it's always been the case that there's no such thing as like hack proof software. But I think we could get a lot closer to that with AI, where it's yeah, we just have in a relentless full time
expert constantly scanning and looking for new holes and stuff. Like, can we actually just be unhackable? And I know the hubris I'm saying here we're about to get hacked tomorrow 'cause I just said that, but like right. Ex I am not challenging anyone. I want to be clear. But I don't know. I feel good about it, I think.
Rick (49:17.23)
you challenge me.
Rick (49:26.2)
That's awesome. That's a positive story about AI.
Tyler (49:28.272)
And also they Yeah, I know. And they just put Fable out, so we need we get one week of Fable, so I'm like I'm messaging our the dev that's doing it like run all the scans before they take Fable away from us.
Rick (49:41.25)
Tell me about core design.
Tyler (49:41.393)
That's all I have to say. Cloud design, yeah. Have you used it yet? so first of all, a thing that is confusing about Claude is the naming. And I don't know what they could do differently, but I've said even people on our team who use claude code, I still say sometimes like Claude versus Claude Code, and they it seems like they don't get that claude code is different from Claude. and anyway, cloud design is also different from Claude. It's it's a different thing. You I think it's
I think it's like cloud.ai slash design or something. I don't know. It's a different URL. Go to Cloud Design. But it's really specifically meant I think we talked about this a while back. It's really meant for creating design artifacts rather than text artifacts at the end of of a conversation. I can just share some thoughts about it, but like really, really good at simple stuff.
Rick (50:31.97)
What what are the for non designers are are like let's just say like I I would imagine people come to you and say, I need something related to design. Like I do that for leg up, for example. What are the if you're gonna give a list of the top three things a non designer should be using claw design for, what are those three things?
Tyler (50:41.328)
Mm-hmm. Mm-hmm.
Tyler (50:50.564)
Yeah, so if you go into it here, I'm gonna pull it up real quick, it it kinda prompts you for like what are you trying to make? And the options are prototype, slides, document, wireframe, animation. So let's start with slides there. I think making yeah.
Rick (51:02.35)
So I'm pardon me for a second. I'm in the desktop app. Where do I go to get into design?
Tyler (51:08.412)
It's not in the desktop app. Nope. Claude.ai slash design. Go there in a web browser.
Rick (51:10.102)
It's not.
Rick (51:14.446)
Claude dot AI slash design. Thank you. All right, I did not know it was not that's probably why I'm not seeing seeing it.
Tyler (51:21.306)
Yep. Yeah. They have started like advertising it a little more. but again, you the thing is you can design stuff in normal claude, so it's so confusing, but it's a whole different flow. I think the the typical corporate user that is not like a designer, I bet designing PowerPoint slides is probably the most obvious use case for this.
Rick (51:44.34)
Okay. But what if you use Google Slides?
Tyler (51:45.839)
And I or or that. I I think it it can export to different things. I'm I mean, worst case scenario, you export to PowerPoint and then import that into Google. But a a a workflow I've found to be really useful. So okay. An annoying thing about both cloud cloud code, planning mode, and cloud design is they're just like, I want to get to the end result as quickly as possible. And it drives me crazy. You you go into cloud code, you go into planning mode, you're like, let's talk. It doesn't want to talk, it wants to create the plan immediately.
What I've found useful is you go into like Claude, just normal chat or cowork or whatever, and you put the whole plan together for a design, and then you say, give me the prompt for this, and then you paste that into Claude Design. So I wouldn't use it for like the brainstorming part of it, but let's say you're doing a presentation. Plan the whole presentation with Claude in chat, and then at the end of it, say, Can you give me a prompt that I can give to Claude Design to design the whole to to create this whole thing?
Rick (52:32.502)
In a chat.
Rick (52:40.982)
And could this be everything from documents to websites to like what's where where is like is it like I understand this I think it's very clear like presentations, but like what about flow charts? Like or is this more design design or is it also like logic, like you UX versus UI?
Tyler (52:45.755)
Yeah.
Tyler (52:56.112)
Yeah. I've only used it for like mock-ups for for well, either UI design or the the like graphic design I did for like designing our banners for this conference, but design design. The the the big difference that I can tell from between this and normal Claude, I'm sure there's a different harness middle layer, but the interface is like a tiny little chat bot on the side or chat box on the side. And then there's kind of like a Figma canvas looking thing.
Rick (53:02.902)
Like wireframing.
Rick (53:25.421)
Yeah.
Tyler (53:25.74)
but you can't edit it. Or I I read an announcement that they said you can. I still can't. so it's not it's probably not good for wireframing, because like I or not wireframing for flowcharts, because with flowcharts I feel like you're constantly dragging back and forth. With this, every time you want to change something, you have to go to the chat box and be like, Can you change this thing for me?
Rick (53:44.738)
Wow, this is incredible. so the prompt I just did was I I've never used this before. So I went in and I said, I want to improve legup health.com. And it's like, okay, that's very broad. What do you want to re do you want to do a visual refresh? Do you want to improve conversion? Do you want to rework the hero specifically? Rethink how it works and options, vi full visual identity refresh. I can check as many as I wanted these. And then it's like, what's not what's not working on the current site? It's like basically doing a full intake of like a like a like a designer would do. Incredible.
Tyler (54:05.788)
Mm-hmm.
Tyler (54:13.04)
Yeah. No, it it's very good. And and the output is very good if what you want is is generic. I think it's
Rick (54:18.402)
Well, can it redesign my logo?
Tyler (54:21.732)
I've tried logo design with it and it didn't go well for me, but Leg Up Health logo is just the word Leg Up Health, so it could probably do better than that. Yeah. I haven't tried it with that. I tried it with like the less annoying serum, which is like a check mark with some outlines and stuff. It's trying to like build an S so there there are things it gets close but not close enough. And so one of my takeaways from this, it is not good at making a pixel perfect mock-up of
Rick (54:38.126)
Mm-hmm.
Tyler (54:50.074)
like the full lesson wing CRM app. Like I I try they have this the whole they they understand what like a design system is, which we have in Figma. We have a design system. I imported the whole thing into Cloud Design. And then now I can say like so for example, the BAA signing page for Hippo, which is like the the BAA is the contract that each customer has to sign with us. So we have to make a page that shows the BAA and has a button to sign it. I was just like make that using our design system.
And it made something that looked like approximately right, but it's still far enough off that it's it's not i would not use it for pixel perfect like UI design, personally. Great for marketing websites. Cause like there's so much it's so much simpler, first of all. Marketing websites are normally not like multiple columns and pop outs and all this stuff happening. But then also, marketing sites are all so similar, unless you unless you're like the one percent that wants something really unique, it can make a very generic looking marketing site that looks
super polished and professional. Yeah.
Rick (55:47.736)
Amazing. That's very cool. Thank you for introducing this. I had no idea.
Tyler (55:52.283)
Yeah. did I have anything any other notes here that Yeah, I don't know. And then the the other thing I'll mention is it's ri you can build actual prototypes in it. So I found it more useful for behavior design than actual like d like pixel design. So when I'm designing a feature that needs to look a certain way, I do that in Figma. But Figma's really bad at designing the behavior. Like what happens if I mouse over this thing? Or if if
I check this box and then I click this button. Versus if I click the button without checking the box, what should happen next? Cloud Design is actually coding up a real HTML JavaScript website for you. And it'll give you all that behavior. So like I, for example, I just a thing we are just we we need to work on for HIPAA. Sorry, this is like a little in the weeds, but HIPAA basically says you can't pass protected health information into s tools that are not HIPAA compliant. Demo desk, the tool you and I both use for
Customer calls, like a a a Zoom type thing. We don't have a BAA with them, not HIPAA compliant. So we can't do customer calls with HIPAA customers in demo desk. So we're updating our flow where when someone books a call with us, instead of sending them straight to Calendly, we're sending them to a page on our site that's gonna say, log in first, we're gonna look at your account, and based on who you are and whether HIPAA's turned on, we're gonna redirect you to different calendarly links. One will take you to a Google Meet, which is HIPAA compliant, the other will take you to Demo Desk. Does that make sense?
Rick (56:53.89)
Mm-hmm.
Tyler (57:20.112)
So I went into cloud design and walked through all this logic of like, well, if they're already logged in, show them this. If they're not logged in, show them that. If they say yes, they have an account versus no, they don't have an account. There it's a pretty simple UI, but there's like all these kind of logical steps of if they say this but not that, then take them here. The entire thing is working in cloud design. And then I just handed that off to the developer. So rather than being like, I have to explain all this to you, I'm just like click around, look at it. It it it it works the way it works.
Rick (57:48.767)
Do you share that how do you share that design with them? Like is yeah, what's the experience there?
Tyler (57:52.815)
Yeah. So they all have cloud accounts as well. So I can just give them the link and then they they can click around and use it. But even better, cloud design, cloud code can pull down the cloud design thing. So they should be able I've tested this out, but I'm not a coder as much anymore. They should be able to go into cloud code and say, here's the URL to the cloud design file. that already has all the logic in it. So no one needs to tell Cloud Code what the logic is. Just like pull down this file and make it happen.
Rick (57:55.63)
Mm-hmm.
Tyler (58:21.072)
And I'm sure like little things will be wrong here and there that we still need a human involved, but I think like ninety five percent of it'll just like one shot be done, probably.
Rick (58:31.662)
It's amazing. Yay. Okay, I'm gonna go play with this.
Tyler (58:32.624)
Yeah. All right, cool. I'll talk to you later.
Rick (58:38.062)
All right, see ya.