Transcript

Grab A Shovel · Founder Mode

0:00
For one of the long term things, I actually wrote this prompt
0:02
from JJ where I was like, hey, go look at all my memories and
0:05
all the things you think are ground truth and produce a
0:08
document of all the things that you think are are ground truth.
0:12
And then compare that against only things that I have told you
0:17
and dude, 70% was like they're like, oh wow.
0:22
Like I thought of this thing and then I reference did this other
0:24
thing and that's why. And it was like just a ton of
0:28
like and I think hallucinated the wrong word 'cause it it was
0:31
synthetic thing where like it responded to something and then
0:34
we talked for a while and in its thinking it had recorded
0:38
something is true that was only true to its own sort of like
0:41
belief. Hey, welcome back to Founder
0:58
Mode. Yeah, we're here again.
0:59
Just you and me, Kev. I like it when we do.
1:01
These solo episodes I might because I feel like it's the
1:04
unscripted thing always throws off the like well.
1:06
You're you're you're a risk when you're unscripted.
1:08
Definitely. But I think the other, yeah, go
1:11
ahead. No, I, I think it's good.
1:12
I think we talk, we talk enough outside of the podcast that we,
1:16
we have these conversations like, oh man, we should talk
1:18
about that on the podcast. That would be good.
1:20
And then by the time the next one comes, it was round.
1:21
It's it's gone, Yeah. Ephemeral.
1:24
What's the What's on your mind right now?
1:25
What are you thinking about? You know, it's interesting, I
1:28
tried this new AI e-mail thing and it it's, it's got a very
1:33
opinionated use case. And what they actually do is
1:36
they take all the emails out of your inbox and then they, like
1:39
send them later to you. They change when they arrive and
1:42
so it's forcing you to prioritize who you respond to.
1:45
Yeah, so there's a company back in the day called same Box that
1:48
it kind of turned your e-mail into a white list and they took
1:50
all the emails out of your inbox.
1:52
But now with AI you can teach this thing.
1:54
I'll just give them the prom. It's like it's the companies
1:57
called every. I kind of like this company,
1:58
like their every dot to if you've seen these guys.
2:01
Yeah, yeah, they have a bunch of little, little.
2:03
And so I've tried their so there I turned on their Cora to a
2:07
couple of my e-mail accounts, so CRA and so just to see.
2:11
And like in one of my e-mail accounts, I only get newsletters
2:14
and now I get no e-mail because it takes all of them.
2:17
And then you just don't read them.
2:18
No, you get an e-mail once or twice a day.
2:20
With the summary. With a summary of what's in
2:22
there and then you have to click and go read it now where it got
2:25
caught. And then we actually had a guy
2:27
that we're having a guest to my buddy Scott.
2:29
Like I right, his newsletter was there and I was like, Oh, that's
2:32
a good one. I forwarded it to you and then
2:34
you replied, let's have him on the pod that got shipped into
2:38
the like summary brief. So I was like, wait a minute,
2:40
like that's weird. That's an e-mail.
2:42
Like how could it be two days? And like Jason hasn't responded
2:45
And I went back and it said, Oh, Jason from Winston Francois
2:49
discusses this. So I was manually labeling that
2:51
stuff and having admins check it.
2:53
And that was my old model. And that works pretty well.
2:56
It literally gives me a things at the top that I can focus on,
2:58
but I always had the rest of the stuff there where I could kind
3:00
of browse through it at time. But this notion of like it going
3:04
out of my inbox is a little bit opinionated right now.
3:07
And so I'm going to see like again, it's I'm only trying to
3:10
call 2 of my 6 emails, as you say.
3:12
But yeah, we'll see. Yeah.
3:13
You know, I, I think, I think the idea here, this is where AI
3:17
is supposed to shine, right? The idea that you have, you
3:19
know, hundreds of emails you might get a day across 6 inboxes
3:23
and you probably don't need to look at most of them.
3:25
And a summary is particularly useful.
3:27
Like AI is very good at summarization.
3:29
I think the thing that we were talking about before the show
3:31
that I, I feel like this is a good segue from, is kind of good
3:35
AI, like AIAI is, is good and helping US versus bad AI.
3:40
And that newsletter you forwarded me kind of touches on
3:42
this topic of like the Turing test and, and, you know, are we
3:45
getting, are we kind of getting sucked into the machine a little
3:48
bit? I, I used the word sycophant
3:50
when I was sexting you the other day.
3:51
You're like, I can't spell that word, so I don't use it.
3:53
You know, I, I, I wonder, like how much are we trusting the
3:56
machines? And are they?
3:57
Are they helping us or are they hurting us?
4:00
Sometimes, yeah and I think I mean back to the e-mail thing
4:03
again, like the great examples like so I've been I vibe coded a
4:07
mobile app over the last 24 hours.
4:10
So first time I built like an iOS, Android web, kind of, you
4:13
know, mobile app. And that thing generated, you
4:17
know, 100 plus commits and 3040 PRS and all those emails went to
4:22
my inbox and that thing summarized it.
4:24
And it, it had this really long summary of like these 15 topics.
4:27
And I was like, what's this thing?
4:28
And it was like, it was like it. Was trying to make sense of it.
4:31
Trying to make sense of it, but also worse, it was reading all
4:34
of the PR change request details and trying to like understand
4:37
how we got there. And a lot of the stuff like I
4:39
never saw because the, the, the vibe coding.
4:42
I don't know how it made this mistake and added the testing
4:44
and fixed it. And now you have another AI
4:46
reading that and trying to educate me on what it learned.
4:49
And I'm like, like pretty, pretty, pretty over the edge.
4:51
Yeah, and not, not necessarily helpful.
4:54
I said this recently that like I was starting to find Chachi PT
4:58
had gotten really bad for me like and and less useful.
5:01
And I read an article just in the last couple days and, and
5:04
you probably already knew this, but I didn't realize that the
5:06
way that the chat sessions work, each chat, it goes back and it
5:10
stacks every single prompt. And so when you have sent one
5:14
message, there's one message sent.
5:16
And then when you send two, it's the first two and then you send
5:18
three, it's all three and so on and so forth.
5:20
And so I was having these tabs, there were 800 MB of RAM on my
5:24
Chrome tab because I had 100 back and forth messages between
5:30
me and chat and it became unusable.
5:31
Yeah, and what's interesting, this is funny.
5:33
So this is like this whole notion of like if you played
5:35
with these thing, they they do compaction and they try to
5:38
summarize and how they manage memory.
5:40
And my Co founder, you know, JJ, he actually manages his memory
5:44
directly. And so he doesn't allow like a
5:47
lot of these chat things is like, oh, it does the thing and
5:49
then it summarizes it and creates like a kind of a semi
5:52
hidden memory or like sort of knowledge based.
5:54
He goes and touches the memory file.
5:56
He sets it up so you're like, it's only allowed to add things
5:59
to memory that he approves. So he almost like change lists
6:02
this thing. But what's interesting is that
6:05
like the results that he gets over a long time are way more
6:09
consistent. And so like, I, I think there is
6:12
this notion of like he's LLM sort of like thinking that
6:16
they're smart, but a lot of times they're reading their own
6:19
cock, their own thoughts that were like suggestions.
6:21
And so for one of the long term things I actually wrote this
6:24
prompt from JJ was like, Hey, go look at all my memories and all
6:28
the things you think are ground truth and produce a document of
6:33
all the things that you think are are ground truth.
6:35
And then compare that against only things that I have told you
6:40
and dude, 70% was like they're like, oh wow.
6:45
Like I thought of this thing and then I reference did this other
6:47
thing and that's why. And it was like just a ton of
6:51
like and I think hallucinated the wrong word because it was
6:54
synthetic thing where like it responded to something and then
6:57
we talked for a while and in its thinking it had recorded
7:01
something is true that was only true to its own sort of like
7:04
belief. We're like, chap, like you have
7:06
5 locations and it was like, oh, no, we had talked about having
7:11
one location, but we were thinking that there might be a,
7:13
a time to open some more. And they said what would be the
7:15
number where we'd have to change the strategy.
7:17
And it was like, oh, once you're at 5 locations, you have to
7:19
change the strategy. But we never agreed that that
7:21
was happening with, but somehow in the chain when I went back
7:24
and said ground truth, it's like, oh, you have one location,
7:26
but multiple times I reference you as having 5.
7:29
And so they're like a bunch of chats after the fact have been
7:32
referencing this kind of like implicitly wrong data.
7:35
And so I think there is this again, it's an interesting
7:38
prompt. So it's a go do all the things
7:40
that you know about me didn't compare that against things that
7:43
I've actually told you versus things that you think you know
7:46
about me. And it was like, surprisingly
7:48
wrong. Yeah.
7:48
So I think that I think we should, we should try to break
7:51
down the frameworks that that we both use.
7:53
So like just starting with the AI piece, like good AI versus
7:57
bad AI, right, like good use cases and good and and right
8:00
model purpose built for right thing that you're trying to do,
8:03
right. You have a pretty, you have a
8:05
pretty smart framework for how you do some of these agentic
8:08
stuff where you where you think about how you interact with the,
8:11
the model versus the model going off and spinning up agents and
8:15
using those agents and doing, you know, doing specific tasks
8:17
with the different model. Like talk through, you know,
8:20
what you think about this stuff in the context of like founder
8:22
mode and building companies and augmenting human staff with AI
8:26
and agentic staff. Like what's your what's your
8:28
approach to doing that, Kevin? Yeah, I think to me, and it's
8:31
funny, I was just talking to like one of my teammates before
8:34
this call and before this recording or whatever.
8:36
And we were talking about like, you know, how do you approach a
8:39
workflow where there's no AI today?
8:41
And then how do you like, what's the most intelligent way to
8:43
apply AI to that? And like, how do you think about
8:46
what you should, you know, how do you separate the problem?
8:49
And I think there's one piece of that that's like, how would you
8:52
train a new employee to do that job, right.
8:55
And I think what tools would they need?
8:57
What training would they need? What context would they need?
8:59
What access would they need? And I think this idea that the
9:04
content, it's like, well, they need to log into this system to
9:07
get data. They need to log into this
9:08
system to get the task queue and they need to log into this
9:11
system. So like, let's just take like
9:15
like a very simple one. Like let's just say you're a
9:17
finance person and you need to produce a weekly 12 week cash
9:22
flow and two systems of record, right?
9:25
You have let's say QuickBooks and you have like Salesforce,
9:28
right, or HubSpot or whatever you use for sales.
9:30
And so you're like, okay, well, the first thing you need to do
9:33
is just have access to those two systems and dump the data and be
9:36
like, hey, what do we think we're going to sell over the
9:38
next 12 weeks? And what do we think of those
9:40
sales will convert to cash on what time?
9:42
And then what stuff do we already have sold that's going
9:45
to collect cash on? And then what do we think the
9:46
expenses are? Like pretty simple.
9:48
But what you first figure out is like, well, the human that's
9:51
doing that is constantly micro correcting both systems.
9:54
It's like, oh, the sales pipelines kind of wrong.
9:56
And the QuickBooks thing is got these weird errors that we don't
9:59
track correctly. And the Angel for me.
10:02
Corrections. Yes.
10:03
And so how do you know about those corrections?
10:05
So for me, it's like, well, first let's just fix this, fix
10:07
the system of record. And so the first thing I did in
10:10
those two things was wrote an automation that just scanned
10:12
those two systems of record and been like, oh, if there's ever
10:14
an anomaly, let's send that to the person in charge of it.
10:17
And SO1 keeps the bookkeeper. And the other case is like the
10:20
sales Rep, it's like, Hey, this deal is not in the right place
10:22
or it says it's AQ one deal, but the close dates in Q2, OK, fix
10:27
it. And then it's like, OK, generate
10:29
the report. And now have, you know, sort of
10:32
the, the sort of finance team look at it and be like, OK,
10:34
what's good and bad? What will we change?
10:35
What will we highlight? And so to me, like it's that
10:37
kind of layered step and then you start running it, you know,
10:40
kind of in a shadow mode where it's like, it's not doing the
10:43
job, but it's sending it. And then pretty soon you're
10:45
like, damn, we're not making any corrections.
10:46
It's kind of working. And then you can kind of like
10:48
productize that flow. And so I think to me, it's,
10:51
again, pretty much anything can be broken down into that way.
10:55
I think the part where it's tricky is like, and I'd be
10:58
curious on your side is like, because I've done some of this,
11:00
I'm like generating blog posts or generating articles or
11:03
content. As you look at this stuff and
11:05
you think about like that's a very practical thing that your
11:07
team does all the time. Like how do you guys approach
11:10
it? And like, how do you think about
11:11
like the important pieces there? I'll tell you, two years ago we
11:14
were making AI content on behalf of clients very differently,
11:17
right? We was Chachi BT and Mid Journey
11:19
were the two tools we used and we would generate articles and
11:23
then we would use human editors to review them.
11:25
We would do the internal linking, follow all the SEO best
11:28
practices. We would go to mid journey and
11:30
and kind of create handcraft prompts to make images that
11:34
looked appropriate for those those articles.
11:37
And that still was a lot of people hours.
11:38
Fast forward to today and I have an agent that's running 24/7
11:42
generating a content for for Winston Francois for our clients
11:46
that can generate dozens of articles every day and publishes
11:49
them every 30 minutes. And to your point, it's gotten
11:52
so good that I don't touch it and then it can just go.
11:54
But in order to set that system up, it's not just one agent
11:57
that's been given a prompt that goes off.
11:59
It's that there's the agent that comes up with the taxonomy in
12:02
the schema that is relevant to all the different topics that we
12:05
need to discuss that are important for keyword search and
12:08
zero click search on these AIAI kind of chat bots.
12:11
There's there's the writer agent that goes and generates the
12:14
content. There's the reviewer editor
12:15
agent that reviews that content from the writer and says, hey,
12:18
that wasn't very good. Let's go back and revise it.
12:20
There's the creative agent that's going and making an image
12:23
that pairs nicely with that content.
12:25
And then there there's the final like Q agent that make sure,
12:27
hey, does this article have an image?
12:28
Is it within the word count that we need to hit touch the topic
12:31
and match the taxonomy in the schema?
12:33
And then that kind of final publisher agent pushes it live
12:35
and then throws an error and says, Hey, this one didn't go
12:38
live for this reason. And so like now we've got a team
12:40
of 6 or 7 virtual, you know, AI agents that are doing that work.
12:44
And I think, you know, it's, it's about designing to your, to
12:48
the earlier point you made in this, this conversation, it's
12:50
like making sure that they have the right tools.
12:52
So they all have access to the Google Nano Banana for image
12:56
generation. They're using Claude for
12:57
writing. They're using the WordPress API,
13:00
the REST API to send back and forth the stuff and publish.
13:03
And then, you know, that's all coming back into Slack to talk
13:05
about it, right? Like giving them the right tools
13:07
and access and then setting up the team for success, which goes
13:10
back to what you said a moment ago, which is OK.
13:13
So those, that's how we kind of build systems and build kind of
13:16
virtual teams of AI agents. But when you're building a
13:19
company of humans still, which we all are still doing well, how
13:22
do we, how do we set that up for success?
13:24
Like I have a lot of thoughts from my early Google days
13:26
watching Sheryl Sandberg do the higher A's, higher A's versus
13:29
don't hire the bees because they hire the seasoned bees.
13:31
But like you've, you've basically been building startups
13:33
for almost 3 decades now. Kevin is your kind of framework
13:37
for building high performing teams.
13:40
You're looking for different talent to match with those
13:42
different organizations as you scale across them.
13:45
And to me, the the fun part is this kind of zero to 10 kind of
13:50
thing where it's like, hey, you're just getting started.
13:51
There's a lot of autonomy. You're not really even writing
13:54
things down and tracking. Like there's this really high
13:56
bandwidth discussion of like what needs to go be done?
13:59
And the kind of people you want there are like highly
14:02
trustworthy, but also like highly autonomous where they're
14:05
like, hey, I can just go own this area and kind of figure it
14:08
out and don't need a lot of like management, right?
14:11
They don't need a lot of things. But I think the other thing you
14:14
need to be able to be brutally, truthfully honest.
14:16
And I'm a pretty tough person, like with that, like honesty of
14:19
like, this is not good work or like this is you need to go
14:22
harder or faster or whatever. And I think, yeah, So I think
14:25
that creates an environment that works for some people and
14:29
definitely doesn't work for others.
14:30
But for the right people, it's like invigorating.
14:33
Like they're like, I just can't stop doing the thing because
14:36
it's so exciting. And they have like so much, much
14:39
autonomy that they really like. It's really endless in terms of
14:42
the impact that they can have and the creativity that they can
14:45
apply to how they do their job, when they do it, where they do
14:48
it, and that kind of thing. And so I think that part of the
14:49
startup hiring is like the best but.
14:52
Yeah, the phrase I hear you say all the all the time is just
14:55
like, you know, grab a shovel, which I, which I like a lot.
14:57
And I think, I think what that actually is, is the idea of
15:01
hiring for high agency. So people, and I use the phrase
15:05
external locus of control, which is like, there's, there's people
15:07
that think the world happens to them and that's the internal
15:09
locus of control or sorry, high internal locus of control.
15:12
External locus of control is like, oh, everything happens to
15:14
me. I have no control of my world.
15:15
Internal locus controls like I completely control my destiny.
15:18
I'm in charge of where it goes, right.
15:19
And I think what I've seen, because I've seen you manage
15:23
teams and people you know to sometimes to varying degrees of
15:26
success, but I think it's because it's that filter that
15:29
you're applying, which is like, does this person have the
15:31
ability to go off and do work and come back and deliver
15:35
without me having to spoon feed them instructions or things to
15:39
do? And there's different bars for
15:41
the administrative assistant person versus the like software
15:44
engineer versus your Co founder, right?
15:46
And I'm not saying that you're you're using the same framework
15:49
to evaluate all those people, but there's the consistent
15:51
thread through all of them, which is like, are you a high
15:52
agency person with an internal locus of control that believes
15:55
they can go do things and get them done on their own without
15:58
somebody having to tell you what to do next?
16:01
Yeah. And then you raise your hand
16:02
when you need help, right? And then you raise, help me out.
16:06
Yeah. And you're like, hey, I'm, I'm,
16:07
I'm asking for help. I, I do think that, yeah.
16:09
The shovel thing's an interesting one because I think
16:11
as you're starting companies or even in in bigger companies,
16:14
it's very common to get a lot of advice from people and they're
16:17
like, hey, able to tell you how they think it's to do.
16:19
This done this way. And I'm like, that's awesome,
16:21
man. Like if you're so right, then
16:24
just go do it. It's surprising how many people
16:26
don't want to pick up the shovel.
16:27
Like we always talk about like in interviews.
16:29
Are they willing to pick up the pin and go right on the board
16:31
and like draw like what they're doing and scope it out?
16:34
And I think the beautiful thing in startup, you don't, I'm like,
16:36
you don't need to go to the board, just go right to the
16:39
work. And then like, cuz, look, we
16:40
don't, I don't know the answer and we're gonna have to go
16:42
figure this out. But a lot of it's like, are you
16:44
just willing to get your hands dirty and like really get
16:46
involved? And it's just really easy for
16:48
people, especially ones that have had some success or have
16:51
worked at larger companies were like, Oh yeah, like I've kind of
16:53
earned the right to play this role or to like, to like deliver
16:57
advice and to sort of show you the way.
17:00
And I'm like, that's awesome. But like the best way to show me
17:02
the way is show me the results. And like that surprisingly
17:06
filters out a lot of people that that frustrates the shit out of
17:10
them because they're just like, damn, like I just want to tell
17:12
you how to do like, hey, I don't, I don't do that anymore.
17:15
And you're like, what do you mean do that anymore?
17:16
Like I don't get on cold calls or I don't yeah, I don't sit on
17:19
1st sales calls or like no. My job is one of those phrases
17:23
that's the kiss of death. Yeah, I don't, I don't fix bugs
17:26
anymore. Like I just work on these
17:28
architecture stuff like I'm not a coder like, you know, and
17:30
you're like, wait a minute, like you know, I don't just shouldn't
17:32
be. I don't you just be like, Hey,
17:33
what's the necessary? Thing what do I need to do Yeah
17:36
yeah, it's a self selection thing, right?
17:37
People are self selecting out as, as I've heard you say before
17:40
and and I think in true founder mode spirit, the founders that
17:44
roll up their sleeves and are working alongside by their team
17:47
and fixing bugs and getting on sales calls and talking to
17:50
customers and doing whatever, like putting out whatever the
17:53
big fire is, tend to be the ones that both earn the most respect
17:58
from their team and also kind of encourage and and end up
18:02
cultivating an environment of people that all operate that
18:05
way. They just want to go get stuff
18:07
done and win. I think you say this a lot too,
18:09
is like, I just want to win. Yeah.
18:10
Winning's really, really infectious.
18:12
Yeah. It feels really good.
18:14
And like, I mean, winning at all causes can be dangerous, but I
18:17
think there is this sort of like figuring out what winning looks
18:20
like and then everybody kind of being on that same page, rowing
18:24
in the same direction, all that kind of stuff.
18:25
Like you pick your favorite analogy, like has been sort of a
18:28
fun thing to go do. And it's like, how do we make
18:31
sure that the first two set us up to be successful to make it
18:34
even faster to onboard the next 8, right.
18:37
And and then creating that in an excite where that's exciting for
18:40
people and not like, oh, wow, you have to do this other thing.
18:43
We're going to do this eight times.
18:44
It's like, no, let's like, again, hack the process to make
18:47
this work where it's like kind of fun.
18:50
And then because of that, now we can go do these other things
18:52
that we couldn't do before because the team went from, you
18:54
know, five people to 12 people or whatever the number ends up
18:57
being. And so I think that stuff all,
18:59
you know, when, when, when taken together, it's like, does this
19:02
person fit that culture or not? And that's probably why you and
19:05
JJ work so well together is because, like, you guys come at
19:07
it from a very different place. Yeah, no, there's definitely
19:09
like feelings aren't probably high on my like little chart
19:12
when you like we used to do these three 60s at big companies
19:15
and stuff and like emotion, empathy, feelings like they're
19:19
like like would get zeros on him or something and you're like.
19:22
Are you OK, Kevin? Yeah.
19:23
And you're like, are you OK? And you're like, well, no, I
19:25
like, but I just like the other things like competitive winning,
19:28
like, you know, they pick these other like, you know, sort of
19:30
words that show up at the in the top five out of the, you know,
19:32
your little skills skills matrix.
19:34
And so I think, again, there's different kinds of cultures and
19:37
quite frankly, there's different people.
19:38
Like some people need more of that maybe calm energy or sort
19:42
of like reflective sort of thinking of like, Hey, how do we
19:46
how did we get to where we are? And how do we think through
19:48
that? The shift in terms of like, you
19:51
know, well, they're like, hey, we need a culture.
19:52
Like the team's down, we need a culture talk.
19:55
And like, you know, there's the hey, we should go do an off site
19:58
or we should go do a like team building event or.
20:02
We go into a room and be like, let's go draw a chart of what
20:05
winning looks like and redefine how why we're here and how do we
20:07
win. And I always motivated with that
20:10
sort of approach versus the like, hey, let's go have a group
20:13
family picnic. Although those are cool too.
20:14
Like the good thing is the big companies do lots of that stuff.
20:17
So there's plenty of time for group family picnics.
20:19
Yeah, and yeah, I just really like the my mindsets built
20:24
around like the other side of it.
20:25
I think for the startup itself, JJ and I find that balance
20:29
sometimes just literally debating that in front of the
20:31
team. Like, I'll be like, dude, we got
20:32
to go reach this thing. This is what the customer wants
20:34
is how we're going to go achieve it.
20:35
And he's like, oh, we should, you know, wait and see what this
20:38
person thinks or that person thinks.
20:39
And then I'll be like, no, I don't think so.
20:40
We're just going to go do it. And if we, you know, and it's
20:43
fun because, you know, we've known each other long enough now
20:45
that we can kind of complete each other's sentence and we
20:47
know why the other person's saying it.
20:49
And one of our teammates, he's always like, yeah, he's like,
20:51
he's like the energy that in the anxiety that I bring.
20:54
And so energy is probably the nice way of describing it.
20:57
The anxiety is probably the most truthful way of like, I feel
20:59
like we're moving too slow. I feel like we didn't achieve
21:01
this thing many times. Like kind of can snap the team
21:04
out of like, hey, we're just in the drudgery of this thing and
21:07
we're not seeing like the golden apple that we're going to go
21:09
collect once we reach that point.
21:11
And yeah, so I think the the shorter version of it is like
21:15
there's a balance. But to me, it's just way easier
21:17
to describe it in very direct kind of capitalistic, like how
21:20
do you win terms and like think of that as the mountain that
21:23
we're conquering together as a team.
21:24
Yeah, I, I mean, I think it's right.
21:26
And so, you know, when we zoom out and think about like good,
21:29
good team member, bad team member, good AI, bad AII think
21:34
it has a lot more to do with like what are the criteria that
21:38
you're filtering for and what is the thing you're trying to
21:40
achieve? And maybe, you know, if you, if
21:42
you want a very cushy feel good environment, then you'll you'll,
21:48
you'll, you'll select for that type of person that I don't
21:51
think they'll be as hard driving as what you're describing as
21:54
your preferred operator. And then the results may not
21:57
match. I remember being at Google, you
21:59
know, 22 years old, right out of college and just thinking like,
22:02
what is this? This is insane.
22:04
This is like an extension of college.
22:05
Like all my meals are paid for every day.
22:08
And I felt like people, and this is, you know, 20 years ago, but
22:10
like, people were still working pretty hard.
22:12
Like it wasn't, it wasn't like just like, oh, go get my free
22:15
meals. Like you see a lot of the social
22:16
media videos now somebody works at Meta or Google or Microsoft.
22:19
They're like, Oh, I show up at 9:30 and I get my breakfast and
22:22
then I like get my coffee and then I sit at my desk for an
22:24
hour and then I go get my lunch. And like, I don't, you know, I
22:27
didn't feel back then. I felt like, oh man, this is so
22:29
great. I'm going to work really hard.
22:30
I'm going to be in this office for 12 hours.
22:32
And and but I think that was that's me and not the place
22:36
probably. Yeah, No, I think it is in
22:38
itself selects, yeah, the right people into the right place.
22:41
I think the good, bad AI, good human versus or bad employee
22:44
kind of thing. I think about it is like, is the
22:47
time I'm spending with this team member giving me leverage or
22:50
output where I'm like feeling like that's a good, like a
22:53
positive relationship? Like the more time I spend with
22:55
this team member or employee, am I getting more output for that?
22:59
Or is that an extension of my thinking?
23:01
And then I think AI is similar, right?
23:02
Is the time that I'm spending talking or chatting with this
23:05
agent? Yeah, you sometimes are probably
23:08
going to be at. Hourly rate, yeah.
23:10
Is it, it's, it's like, what else could you spending your
23:12
time on? What's the opportunity cost of
23:14
investing this person? I just remember when Loom first
23:16
came out and like what a game changer that was, right?
23:18
It's so amazing and like what I want to do now.
23:19
And I think Grok is the first model that does this.
23:22
I don't know if the other ones are doing it yet where it'll
23:24
watch the videos. Like I want to start making
23:26
looms for my AI agent in the same way I do them for people on
23:29
my team and to see if that can deliver the same kind of
23:32
results. Because like that's a super
23:34
leverage thing to do is like, OK, I'll spend 5 minutes, I'll
23:36
make this loom. I'll send it off to the agent or
23:38
the team member and then they're going to go do that thing 1000
23:42
times because it's a repetitive, you know, we do it every day or
23:44
whatever. Yeah.
23:45
And I think the magic is like, you know, again.
23:49
And are they like, as you spend more time with the agent or the
23:52
human, is that compounding? Like, are the conversations
23:55
getting more impactful with less amount of words?
23:57
We're like, oh, we already get like what the mission is
23:59
together and what we're going to go do.
24:01
And so the things that you're asking of me and the things that
24:03
I'm asking of you is creating this loop where you're like,
24:06
yeah, we can spend, you know, the same 30 minutes together
24:09
once a week. And it feels like we're getting
24:11
a lot more done between each other.
24:13
And you're also finding a way to leverage like the time together.
24:16
And then similar with AI where it's like, man, I, I want to
24:18
just keep talking to this agent because it feels like it's
24:21
letting me do things and helping me get things done.
24:24
Where that. But the 20 minutes or the 30
24:26
minutes that I invested talking or chatting with this thing
24:29
produced something that I wouldn't have been able to like
24:31
spend 30 minutes like on my own or with even another person.
24:35
Because a lot of the, at least the things that I've been using
24:38
the latest agents of 4 is like reaching into like large data
24:41
sets of like whether it's CRMS or tools or like the Internet
24:44
and then merging that with some like kind of local data and then
24:47
synthesizing over that and then producing it.
24:49
But I think the one thing that I, you know, the risk is that
24:52
it's so easy to do that and then have it produce a 30 page dock.
24:55
Yeah. And then now do I tax other
24:57
people by sending them a 30 page dock or do I distill that back
25:00
down and send them A1 pager of all of the things that I learned
25:03
that, you know, now we have impact with each other 'cause I
25:05
do think that is. The right question and you know
25:07
there's been many a day where your sleep score is suffers
25:10
because you stayed up late working with the agents.
25:12
More than I would have imagined, you know, then in the like
25:15
again, used before I measured my sleep score years ago, like
25:18
clearly stayed up late addicted to like, you know, working on
25:21
Google ads or trying to crack this coding thing or something.
25:23
And now it's insane the number of things that sort of, quite
25:26
frankly, just deeply capture my attention.
25:29
Kevin, that was a fun one. We covered a lot of ground AI,
25:31
e-mail tools, LLM memory problems, agentic workflows, and
25:35
then we flipped it into how you actually build teams that
25:37
operate the same way. Yeah.
25:39
And I think the through line is the same whether you're talking
25:41
about AI or a new hire. You know, it's like, how do you
25:44
compound the time you spend with them?
25:46
Or are they taxing the time you spend with them?
25:47
Yeah. That's the filter and honestly
25:49
the memory audit you describe telling, you know, chat or
25:52
Claude to separate what you actually told it from what it
25:54
thinks it knows. That's a move every founder
25:57
using these tools should actually try.
25:58
Yeah, 70% wrong, man. It was a scary number, like, and
26:01
if you're building on top of that, it's shaky.
26:03
And to me, it was pretty cool to have this dock.
26:05
And now that I've edited the dock, it's like Kevin's ground
26:08
truths. I think the next move is to sort
26:09
of turn that into like something more project specific versus
26:12
like all the random things I've talked to, chat and clawed about
26:15
over the last 30 days. All right, Kevin, what are the
26:17
founder mode top five takeaways? Yeah, I think you want to manage
26:20
your AIDS memory kind of like you'd manage employee
26:23
onboarding, design your agent teams rather than single agents,
26:26
right? You want to really give special
26:28
instructions, provide those guardrails and then hire for
26:31
high agency, you know, this internal locus of control.
26:34
And how do you really understand?
26:36
Like, you know, can these people go execute and get things done?
26:39
And then the time you're spending, whether it's with, you
26:41
know, the best employees or your best sort of AI tools should
26:44
compound, right? You shouldn't be doing a one to
26:46
one thing. You should see that sort of
26:48
improve over time. And then finally, my favorite,
26:50
you know, winning is culture, right?
26:51
Like there is a culture of sort of like achieving together.
26:54
And I think it's super infectious and really fun.
26:56
If you're a founder figuring out how to build with AI and humans
26:59
side by side, or you know someone who is, send them this
27:02
episode. And if you haven't already,
27:03
subscribe to Founder Mode. Leave us a review so more
27:06
founders and operators can find the show.
27:08
That's it for today. Because the best teammates,
27:10
human or AI, don't wait to be told what to do next.