Episode 58
· 44:45
colin_1_02-05-2025_130726: Yeah.
So how is new job life?
cj_1_02-05-2025_160726: I forgot
how many meetings you're in DevRel.
There's just so many meetings.
yeah.
And especially when spinning up at a
big company, as part of my onboarding
doc, there was a list of like 30 names
go have one on ones with these 30
different people from different teams.
And for some of them, it had a sub bullet
that was like, and when you're in the
meeting with them, ask them for more
people to talk to about X, Y, and Z.
And so then you end up really
like talking to 50 or 60 people
just in the first several weeks.
And so between that, and then project
management meetings and team meetings
and all hands meetings and, org meetings.
At Kraftwerk, there were several
days where I had zero meetings.
yeah, it's been an adjustment.
What would you guess is like the
percentage of time that you're
in meetings versus not right now?
colin_1_02-05-2025_130726: Yeah,
as soon as you said that, it's that
balance of being in the meetings
about the work versus doing the work.
And at some point the
work has to get done.
or I guess a lot of talking about doing
the work instead of doing the work.
I'm pretty good about
protecting my calendar.
Would probably say 40 50 percent
of my time is in meetings, and
that's probably too much still.
cj_1_02-05-2025_160726:
That's better than what I'm at
right now.
So
colin_1_02-05-2025_130726: do
you guys use Google Calendar?
Have you seen the little stats
thing that tells you, like, how
much time you're in meetings?
There's
If you look on the left hand
side, there's like time insights.
So actually I can just tell you right
now, it's actually probably less than 40.
So I average 15.
7 hours of meetings a week,
is what this is telling me.
cj_1_02-05-2025_160726: Okay.
colin_1_02-05-2025_130726: And what I
love is it's like one on ones, three and a
half hours, three plus guests, ten hours.
And then, it's, according to this, it's
saying, Like 30 hours that I'm out.
So I guess it's only 25 percent
of my time is in meetings.
That's pretty good.
cj_1_02-05-2025_160726: Wow.
Okay.
So now that I'm looking at
this, I am way overestimating.
This says I'm only in
meetings 10 hours a week.
colin_1_02-05-2025_130726: think
that it's, do you think it's correct?
cj_1_02-05-2025_160726: I gotta go
count it up because that definitely
colin_1_02-05-2025_130726: Yeah.
I like this because I actually
haven't dug into this.
It shows recurring versus one time.
And then I think that will help
me to identify is there something
in here I can get rid of?
And then I like this, like my one on ones.
I think are all necessary and
I want those and then you can
also track your focus time.
So I think you have to put on your
calendar what focus time is and then it'll
calculate that as well, which will be nice
because I do blocks, which I wondering
if some of my blocks are being counted
as meetings because I'll do writing
time, but it's not set as focus time.
But I bet that'll Make it even less.
So I guess I'm not in as many meetings,
but we're like ramping up for GDC.
So there's a lot of meetings around a
lot of work and the thing is GDC is going
to happen at a set point in time and the
work has to be done So i'm going to start
being a little ruthless about do the
meetings about the work need to happen
cj_1_02-05-2025_160726: In prep
for GDC, how many kind of like
different things are you involved in?
Is it like multiple workshops,
multiple talks, expo hall, like
what's the scope that's expected?
colin_1_02-05-2025_130726: we just
decided we were spread too thin and we
got rid of a talk at an external event
So now we're only doing stuff at GDC.
but yeah, I'm doing a talk at the booth.
So those kind of, I'm going to give the
same talk to like our online audience.
So I can reuse that.
but my big push is I need to have
content done before GDC, like the
docs basically for some new stuff.
So that's what I did last year.
it's amazing that even though
we know when GDC is every year.
It always feels like this, February
is always the crunch because
things aren't done being built yet.
all of that happens and then the docs can
be done soon ish, but there's meetings
about them and reviews and some people
have different opinions about how they
should be built and stuff like that,
cj_1_02-05-2025_160726: We have a
big company conference end of June.
And this is going to be the, one of the
first times that they have like developer.
Specific content.
And so we are just now trying to
figure out what does that look like?
Is it going to be workshops?
Is it going to be talks?
Is it going to be like, what's the
most value we can add in like the two
days we're going to have with folks.
And it's intimidating, like the
amount of work that needs to
get done between now and then.
And,
colin_1_02-05-2025_130726:
How much time do you have?
cj_1_02-05-2025_160726: I
guess like, four months.
colin_1_02-05-2025_130726: Okay.
cj_1_02-05-2025_160726: And maybe two
workshops and expert panel and like
Expo Hall office hours type situation.
I don't know.
We'll see.
We will see.
colin_1_02-05-2025_130726: Yeah, more
and more, I think of like, how do I
create something there that then can
be reused to so that it's not one and
done, it can be used in another event.
So the next event is not as intense.
I say that in like GDC, everything's
being done from scratch again, because
it's a completely different product too.
So it's just, It's tricky.
But, yeah, even some of the workshops
that we do online, like they don't end up
living on YouTube or something afterwards.
So if you missed the live event,
you missed it and we could
do better at that, I think.
And
cj_1_02-05-2025_160726: that I was asking
for was recordings for The workshops with,
and it's everything is a brand new muscle.
It's Oh, we're not sure
if we can record it.
If we can, where's it
going to be repurposed.
We've recorded stuff in the
past and it wasn't used.
So like, how do we set that up?
And so I'm like, there's so many things.
That are happening in like the very
early seed stages, being like the first
dev advocate and in many years here.
And so basically standing up the entire
DevRel function while at the same
time, yeah, trying to prepare for a big
event, trying to get content situated,
but also planting the seeds with
the right people and the right teams
about, Hey, we should set up a YouTube
channel and Hey, we should have first
party SDKs and Hey, we should like.
Move to this docs platform and Hey,
we should do like this X, Y, and Z.
And it's just, yeah, lots and lots of zero
to one things happening right now that are
exciting and I'm pumped about the work.
but yeah, definitely a lot to get
done in a short amount of time.
colin_1_02-05-2025_130726: It's
almost like the how to win friends and
influence people type stuff, right?
Cause you can't do all
of the things yourself.
But, you gotta plant the seeds.
There's a word.
in community managing called tumbling and
tumbling or like having a tumbler is like
a person who goes around and like tries
to get the people who are you're keeping
the conversation going so you want like
the quiet people to speak up you want some
of the People are too noisy to be like,
okay, we hear you, but like you don't
get to dominate everything all the time.
It's a little bit about trying to
keep the community alive and going.
And without it, communities tend to
fall apart or you get the one really
loud thing that it's like, everyone's
assuming that we're going to do the
thing that the loud person's doing.
And then there's a lot of people
who have really good ideas, whether
in the community or in the company.
And those things may not
rise to the top just because
they're not the loudest thing.
and then when those people don't
feel heard, then they leave.
And that might be again, leaving the
community or leaving the company.
So it is, it's a weird balance.
but it's also Oh, I'm going to roll up my
sleeves and go start the channel myself.
But then it's not sustainable because
someone, you need backup, you need, a
good flow of content and stuff like that.
And, as we know, doing this podcast,
sometimes it's good to take some breaks.
Sometimes you're like, yeah,
we're just not feeling it today.
cj_1_02-05-2025_160726: In a lot of
these one on ones is just trying to seek
out who are those people who are maybe
already like itching to make content
or itching to engage with the community
and in certain ways, or maybe they
have contacts with devs in different
kind of like personas or something.
One of the, one of the
really surprising things.
Joining this company, having only really
done dev advocacy at Stripe was that
the flavor of it here is so different.
The dev is not the decision maker
when it comes to buying our software.
The dev is often like a third
party systems integrator, or
they are like ISV partners.
Or they are line of business devs
or they're like business analysts
who are using no code tools.
but they're not like, hardcore tech
CEO, CTO that is at, Bay area startup
that's launching some AI thing.
Like it
is very much the opposite of that.
It is like dev at a logistics company
who's been there 17 years writing.
net and Yeah.
So number one, the needs of the audience
and the needs of the developers are
just so different because they're
not clamoring for the latest and
greatest bleeding edge technology.
There's definitely some that are
in that scope, but many others are
at these giant enterprise companies
doing, long, long running projects,
long running implementations.
I'm also just very much in like the.
Listening and understanding and
trying to get to know the community
phase too.
But yeah, it's way different for sure.
colin_1_02-05-2025_130726:
Is there such a thing?
I think of Stripe and Postman and all
these, are there, even Salesforce, right?
is pretty enterprise.
are there, obvious developer champions
of Samsara, like in the community, or
is it that, they're just heads down
doing the work in their companies and
they're using Samsara, but they're not
necessarily like out there going to
Dreamforce or whatever the equivalent is.
cj_1_02-05-2025_160726: I think there's
like obvious champions internally.
and we, it's a very different model.
Like Stripe would never write code
for customers necessarily, or do
implementations for customers.
We had suggestions for third party experts
and contractors and consultants in the
community, but we do have a pretty heavy
Implementation process where we like
want to have a successful onboarding.
And so we have teams of people
who like work really hard to
make sure that's successful.
And so the boundaries and the expectations
of the teams are just so different.
So
colin_1_02-05-2025_130726: just think
like some, I know our devs identify
as discord devs, And you have Shopify
devs and Salesforce developers.
And I guess that's the hope and the
dream where it's the Kleenex thing,
where your name becomes synonymous
with the thing you're doing.
interesting.
cj_1_02-05-2025_160726:
think we'll get there.
I think we'll get there.
We have a long ways to go, but
I think we can make it happen.
There's so many cool products,
that you can build on top of the
platform and we're releasing more
and more that are really making it
possible to build some really cool.
Companies on top of, there's this
thing called the AT 11, which
is a, industrial grade air tag.
And it basically lets you
get like GPS locations off of
this air tag through an API.
And the physical device is rock solid.
Like you can.
Hit this thing with a golf club,
and, you can bolt it on equipment
and so I'm just super pumped for all
the different interesting use cases
that people will build with that.
As we expand to more.
Physical operations and different tools
and different devices that support
different types of physical operations.
I think there's a lot of cool
stuff that companies can be
born on top of the platform,
colin_1_02-05-2025_130726: yeah, a
certain vintage of winery that puts
these on their barrels or, whatever.
cj_1_02-05-2025_160726: Yes,
colin_1_02-05-2025_130726: from,
from grape all the way to the bottle.
cj_1_02-05-2025_160726: Yeah.
Or even cannabis track, you have
to go like seed to sale or whatever
it's called or whatever.
They have some sort of tracking thing.
that's an example, use case that comes to
mind, but also even just craftwork, right?
Like we, One of the things that we
had a lot of trouble with was just
understanding, like who has the 24 foot
ladder or like, where is the vacuum
Festool, ceiling sucker machine thing
or whatever, or like, where's that 10,
000 sprayer that we use for cabinetry.
And so being able to just bolt one
of these on there and know like
at all times where your expensive
tools are, it's pretty, pretty cool.
colin_1_02-05-2025_130726: Interesting.
the flavors of Deverell
thing back to that.
I feel that a lot because it's
interesting looking from the outside in.
I think a lot of people think it looks
the same, but I've been doing a lot of
interviews and stuff too as we've been
hiring new dev advocates and it's very
interesting to see because we'd known
it's different if you're under marketing
and it's different if you're under sales,
different if you're under engineering.
What are the goals of the company?
Are we trying to convert like through
solutions engineering, which now
you're teaming up with sales and
solutions engineers to build something.
They're probably using your docs.
Are you self serve platform
for like developers?
And we've seen how some of those
have gone from like good to bad.
Twitter is a good example of, I
don't know, like 2015 is when they
shot themselves in the foot or they
just deleted everyone's API keys.
And we're like, yeah, no more API.
And We all know where Twitter is
now, but, but that was such a good
little developer community where
people were like encouraged to build
their own Twitter clients and tweet
bots and all these little things,
Twitter ific that came out back then.
It felt like the golden age of
like developer friendly tools.
And Stripe was part of that in
the early days, SendGrid, Twilio,
all those types of companies.
And, I think like companies that are a
little bit bigger enterprise y, they look
at those and they're like, we want that.
They've all been doing it for 15 years.
you do have a lot to stand up.
but, like we said earlier, you
can't do all of it yourself.
And, you want to jump ahead, but
you got to lay brick by brick
and get the foundation going.
cj_1_02-05-2025_160726: Totally.
Wanted to ask you about SDKs.
We have an open API spec, and it is
cobbled together from two different
internal API builders that use go.
and then we like use some command
line thing to upgrade it from
like open API, B2 to open API, B3.
There's two things that
I really want to happen.
Like one is to have first party SDKs
and two is to have really solid docs.
we are on readme and I tried
upgrading to the readme refactored,
like the new like version of it.
Cause I want to use get.
For authoring and it just, it fell apart.
A bunch of things fell over and broke
and I had to go on support and beg
them to revert my upgrade request.
So curious to hear your advice and like
feedback about docs and, SDK generation.
colin_1_02-05-2025_130726: Yeah.
that's a whole episode.
Um, yeah.
I think that starting
with the spec is great.
I think, I would not be surprised if,
does README have a SDK Gen offering yet?
Not yet.
I would not be surprised if they do
eventually, so if you can figure out
what went wrong with the new system, it's
probably also tied to, you Like, if you
try to use your spec in some of the tools
out there, it'll fail for a few reasons if
you're like not quite following the spec,
or the tool's interpretation of the spec.
So there's a really cool, generator
where you can do your own SDKs.
And they are like as bare bones as
you can In our ecosystem, most of
our stuff is, like, all of our SDKs
are mostly built by the community.
And we want to eventually
release one, but, we don't yet.
And it's do we do a, do we do a thin one?
Do we do, because we don't, want to do the
full batteries included, like all these
like builders and like super powered,
features we want to be able to offer.
right now, the way that we see the
spec is like, if it's in the spec,
then we should also document it
and you should also support it.
And that's like more of that's
just the law of the API right now.
But I've fed it into some
of the generating tools.
And I've created a CLI, it's not Great.
Cause like our spec is missing
descriptions and like field
descriptions as well as just
like end point descriptions.
we played with some tools like read
doc Lee and some of these other ones
where you can inject stuff and upgrade.
Like you said, like we generate our spec
from the API and then I have marked down
files for all the descriptions and it.
Merges those all together, and
then now I have a super spec.
there's a tool I'll find and share that
is like a spec validator from Zooplo.
That's really good, because then it'll
point out like, like a Google Lighthouse
of this is how good or bad your spec is,
and so you can identify those things.
The closer you get to being like
all green on stuff will help with.
There's also a hundred now.
It feels like a hundred.
There's Fern, there's Speakeasy,
there's Stainless, there's all
these other tools out there.
sometimes it's it's the build
versus buy decision all over again.
cj_1_02-05-2025_160726: Yeah, I, went
onto those, like the, those last three
that you mentioned, Fern, Speakeasy
and Stainless, and they all have these
sort of spec editor things where you
can upload the spec to it and then
it'll spit out a bunch of errors.
And there's probably like 15 or so
validation errors that are happening.
So we, we definitely have
some cleanup work to do there.
all the users that I've met
so far, many of them are just
rolling their own HTTP clients.
And a lot of times it's in sharp, C sharp
and they're just using built in the built
in HTTP client to make whatever API call.
So yeah, excited to give them a proper
tool that has all the like IntelliSense.
And one of the nice things that
we, in our position, we do actually
have like the descriptions for the
fields and the descriptions for the
API endpoints, which is awesome.
And I'm grateful we, we do, but
there's things like examples and.
In terms of like input request body
examples and output, we do have like
responses for 200s and sometimes
error responses, but not always.
So there's some work to be done there.
colin_1_02-05-2025_130726: I think
some people like to generate the spec
and then say it's done and throw it
over the fence But like the thing that
the examples are a great one because
it's do you where do you put the code?
for the examples the requests examples or
response examples, and then do you test
them and I think that you mentioned like
the standalone like just doing an SDK.
It's like a really thin client is it
Like, I think you and I are both used
to, the syntactical sugar of Ruby.
Is it the Ruby way, or is it just,
this ham fisted, sure, it makes an
HTTP request, but it's not fun to use.
it makes, brings no joy
whatsoever to the experience.
And not every engineer or, tools
person is going to be optimizing for
developer experience or surprise and
delight but like honestly even like
the original heroku like there's just
even the animation when something is
happening it's just like that's extra
work that makes you feel good about
it using it so you know even the when
you generate a new app and it has the
funny name like those little things
are noticeable the thing that i noticed
when I use our API keys versus Stripe's
is, and this is probably on purpose.
If you double click on an API key
from Stripe, it copies the entire
thing because it uses underscores.
If your API keys have dashes and
other symbols in them, double clicking
on it will only grab part of it.
so it's like such a weird little example,
but I know that's probably why it works.
Is generated like that.
is that a secret
cj_1_02-05-2025_160726:
I don't know for sure
that was there like
way before I got there,
so maybe.
colin_1_02-05-2025_130726: I mean I
see some of these things and it's like
weird This is a sign of getting older or
something but like same their co working
space we have Ikea coffee cups and we
have other coffee cups The ikea ones are
superior for the most mundane swedish
reason that you would never imagine The
bottom I don't this is hard to describe
on radio, but the cup, there's like
a ring on the bottom when it's in the
dishwasher, that ring on every cup is full
of water when you take the dishes out.
All the IKEA ones are like an X, with,
and there's no water on any of them
because all the water drains out.
And someone thought about that, and
they're like, we have to do this.
And that's the stuff that I think we
should be bringing that level of detail.
To the things that we build, but sometimes
you got to fight for, that little
bit of extra, experience and delight.
cj_1_02-05-2025_160726: prioritizing
those tiny little details is especially
important when the developer is like the
chief persona that your entire company is.
Set out to serve
and when they're not, it's and
they're an afterthought, it's okay,
yeah, we're going to build all the
new features for the dashboard.
Maybe it'll make it in the API.
And if it makes it in the API,
it'll probably be like 80%.
And then
colin_1_02-05-2025_130726: Yeah.
I guess with discord, you aren't
evaluating another API, right?
If you want to build on discord,
you have to use our API.
So you have to use whatever
experience we give you.
I don't love that.
That's like a reason to not.
Make it great, if you have to use it.
Same with, Sims 4, it might be that,
someone above you is choosing it and you
have to implement it or, some partner
is using it so you also have to use it.
I think with some of our game development
stuff, I think it's like producers and
studios that are picking it and then
the game developer has to go read the
docs and, would they have picked it?
No, but if you want to ship
on discord you have to, right?
So same if you were to go do instant
game on facebook, like there is only
one way there isn't like three ways
to pick from so yeah, that adds to the
different flavors of devrel for sure
cj_1_02-05-2025_160726: One of the
other things I was curious about too.
So Logan, my son has been playing
around with cursor and he spun
up a game and pie game and.
I like showed him a couple
little things about here's how
you use GitHub or whatever.
And he's like pushing commits to GitHub
under my name and using cursor like
a boss and just Oh, it didn't work.
It broke.
fix this error, make it full
screen, make it multiplayer,
make it do this, make it do that.
Put these sprites using chat
GPT to generate different
screen backgrounds and stuff.
And he's at a point where I
think he wants to publish it.
And I was like, Oh, for sure.
You should try to publish it on discord.
Cause I, I know that like the, there's
an insanely huge community of users
and because it's two player, we
try to get it up and running there.
And then I did like a tiny amount of
research and it seems like getting
pie game into the browser is quite a
bit of work just because you have to
like, Use, I forget, like web, WebGL or
something, or compile,
colin_1_02-05-2025_130726: Like wasm.
Yeah.
cj_1_02-05-2025_160726: Wow.
Down to WASM.
Exactly.
Yeah.
And so I was like, I don't
know if you want to do that.
maybe we should just rewrite the
thing in JavaScript and HTML,
but,
colin_1_02-05-2025_130726: And
there's web based game frameworks,
like Phaser and some others too, so
Phaser might be a thing to look at.
especially because Phaser has an
actual starter project too for Discord.
cj_1_02-05-2025_160726: Nice.
Okay.
colin_1_02-05-2025_130726: So yeah, you
ran into the classic thing of I have
a game that runs on a computer, right?
So our activities are really
meant for web based stuff.
but that's a good example of we
haven't written a lot of guides yet
because we haven't had the time, right?
Do we work with phaser and phaser
writes the guide and then we write the
unity guide or something like that.
So those are the things that
we're still figuring out.
But, yeah, you were
mentioning the AI stuff.
I think one, what age does
Logan get his own SSH keys?
cj_1_02-05-2025_160726: Oh, he,
that's like for his birthday.
He's I want some, I want an
open AI API key.
colin_1_02-05-2025_130726: You're like,
I don't want to be committing his dad, I
need, I don't want to, I don't want a car,
I just want, I need my own github login,
I need to be doing my own commits, because
he's got to start that commit graph early.
cj_1_02-05-2025_160726: He has,
no, he has his own get hub.
He has his own get
commits.
but he cursor when he's using
cursor, he's using it on my machine
I have a paid cursor version and he
used cursor on his own machine and
like immediately ran out of tokens.
So he like hit the rate limit.
He's dad, let me use yours.
Cause I already ran out
of tokens or whatever.
I was like, Oh man, this is crazy.
colin_1_02-05-2025_130726: we'll
send this episode to Cursor
and say we need a family plan.
They'll be like, this is the
first we're hearing of it.
cj_1_02-05-2025_160726: Yeah.
It's yeah.
colin_1_02-05-2025_130726: How
much of that, so how many AIs
are you paying for right now?
cj_1_02-05-2025_160726: Oh,
this is a great question.
colin_1_02-05-2025_130726:
Which is a weird question to
be asking at the year of 2025.
cj_1_02-05-2025_160726: I pay for cursor.
I pay for Descript.
I don't know if that counts,
colin_1_02-05-2025_130726:
I think it counts.
cj_1_02-05-2025_160726: I just canceled
11 labs because I was paying every month
because I had this aspiration that I was
going to go and build something with it.
And I was like, Oh yeah,
this is the month for sure.
I saw the, every time I
saw it hit my credit card.
I was like, this is the month I'm
going to build a thing on 11 labs.
And since I, Built story
fuel, which was like a podcast
story kind of generator thing.
And that was a long
time ago, pre GPT four.
And then I hadn't touched it since then.
And so I've just been paying
like, okay, so canceled that.
What else?
we're paying for one chat, GPT membership.
and I have chat, GPT, like
enterprise through work,
colin_1_02-05-2025_130726: When you
pay for chat GPT, are you paying
for the 20 a month or like the 200?
cj_1_02-05-2025_160726: Yeah.
Yeah.
colin_1_02-05-2025_130726: Gemini is free?
Question mark?
I don't actually use, I've
never really used Gemini.
cj_1_02-05-2025_160726: Yeah.
Okay.
So those are all the tools that
I'm paying for like tool wise.
And then I have usage based accounts
with both open AI and Anthropic where
I have API keys that I'm using like
from my website or from like little,
I set up like a script that runs.
Yeah.
That reviews my PRs on GitHub and sends me
a Slack message with here's some content
that you could tweet about, like something
that you did that was interesting
this week or something like that.
colin_1_02-05-2025_130726: and then you
pay for cursor, but then are you also
paying for the model underneath it?
Or are you just using
whatever comes with cursor?
cj_1_02-05-2025_160726: I think
that it's using cloud sonnet
colin_1_02-05-2025_130726: Okay.
cj_1_02-05-2025_160726: and
it's not using my API key.
it's using theirs.
And I think part of that is they have
their own, I don't know how it works,
but I think they have their own models.
That also do like indexing
and some other stuff maybe, to
help make the rag bits better.
But, I'm paying for it and I
have an account through work.
So I, that's like a double
dip, But what about you?
What,
colin_1_02-05-2025_130726: I think
what's going to be interesting is
what happens when everyone decides
that they're paying for too many AIs.
And there's like this consolidation.
Cause, at work, we have copilot.
That's paid for in github.
I guess I have access to the free copilot.
I don't really use it so yeah, I
pay for d script I pay for Chat gpt.
I pay for cursor and I think
that's it right now at work We're
testing both claude and chat gpt
to pit them against each other.
So like They will be paying
for one of those at some point.
and then I think we're all
getting cursor at work too.
So that's going to be like extra, right?
this podcast won't go out until all
those things are settled, hopefully, but,
cj_1_02-05-2025_160726: yeah.
colin_1_02-05-2025_130726:
but we're all paying weight.
I don't think everyone is doing this.
I think this is engineers
that are doing this.
We all want to like, see
what the difference is.
Which one do you reach for first?
cj_1_02-05-2025_160726: I definitely
have been training myself more and
more when I'm in cursor to try to
stay in context, stay inside a cursor.
If I'm not like, if I'm doing
stuff for work, I use chat GPT.
the enterprise version is pretty good.
I've started organizing it into
projects and in each project, I'll
give it a system prompt that's related
to that helps keep it pretty tight.
And the enterprise one does like
fancy stuff where it won't leak your
work.
Whatever details.
So I use that one a ton
,
colin_1_02-05-2025_130726: I think that's
why all the enterprises are looking into
it because it's like we know you all
are doing this on your own please stop
and stop leaking stuff, which is smart.
So yeah, like at our
work, it's enterprise, too
cj_1_02-05-2025_160726: totally.
I feel like the, the chat GPT AP or
like the open AI APIs are, I don't know.
I don't want to say like the
easiest to work with cause
they're all like roughly the same.
Maybe it's just like the
one I've, I learned first.
And so I'm like most comfortable with it.
But, I did download, DeepSeek and run,
like running it locally with Olama
and, that every time I open up Olama,
I think part of the deal is that I need
to run like a, some sort of local web
UI or something, because when I run
it in the terminal, it just doesn't
feel as readable maybe, or like as
consumable as it is in the browser.
And Even though the content has
been pretty darn good, I just
feel like the, what I get in the
browser is just easier to parse
for me, like mentally or something.
have you run any of those local models
colin_1_02-05-2025_130726: I have it now
cj_1_02-05-2025_160726: Okay.
it's two commands.
It's like basically like brew
install and then do a thing,
or I don't even know if it's brew install.
It's Oh, llama.
You install the Olama app.
And then in the terminal you write
Olama run and the name of the model.
And it just pops up in a terminal.
That give you a prompt to start typing.
colin_1_02-05-2025_130726: I think
what I like about Cursor is that, you
can swap out what model it is, right?
I don't know if it works with Allama and
stuff, but, it feels like these are all
going to be interchangeable, even if you
go to Azure or, GitHub has this thing
now where you can compare two models.
Have you seen this?
So you can run the same prompt against two
models and it'll show you the difference.
we'll put a link to that, but it,
like you can swap them all out
in GitHub or Azure or AWS, right?
Like it, it doesn't really matter.
And so it's like going to become
like electricity and then it's
the layer on top of that or the
custom training for like you said,
like certain rag for programming.
is it, what does the game
development one look like?
And I still, I think no matter
what I do, I still find myself
not being worried about, I don't
think this is going to replace us.
It just makes us do things either
faster, better, stronger, right?
cj_1_02-05-2025_160726: It's been
interesting watching Logan build
basically this whole thing by himself
and saying Oh, I didn't actually code it.
I don't know how to code.
I'm like, you you prompted your
way through 500 lines of Python.
And like several classes, like
files, like tons of files.
It's a full get repo is
like a bunch of stuff.
I'm like, that's exactly what
professional programmers are doing.
Like you, the way you just did it
is the way that 90 percent of like
professional devs are doing it now.
so yeah, I don't know.
it's interesting.
One of the things I'm curious
about too, is whether or not
SDKs are even going to be.
Thing anymore, right?
are people going to do this
bespoke handcrafted implementation
or are they just going to ask
an LLM to spit it out for them?
I still think the SDK makes
it look better and has lots of
benefits in terms of retries and
auto pagination and like error
handling, things like that, that And
colin_1_02-05-2025_130726: We should
do this as a little project together.
We should try to prompt an SDK.
Because, if you prompt the
retries, okay, handle retries.
Because if you were normally
writing a prompt, we run into this.
I'll often say, build
me a quick Discord bot.
And it wants to default to Discord.
js.
So it wants to use an SDK because
it's the most documented SDK.
Our docs are, like, pure HTTP And
so I'll tell it intentionally, use
Ruby, use JavaScript, whatever.
Do not use Discord.
js.
Do it, if it's so vented,
just do straight WebSockets.
Don't use anything fancy.
and it'll eventually get there,
but then I have to tell it, what,
make sure you send the heartbeat.
Make sure you retry.
Make sure you catch errors.
it would almost be interesting to
say, What does a prompt look like?
And you could give it any API spec.
And maybe that is what some
of these companies are doing.
I don't know.
But, some of this is not LLM needed.
It's more just literal
function, generation and stuff.
But, I wonder how close it would get.
cj_1_02-05-2025_160726: When you're,
Documenting WebSockets, you surely
you have the format of the message in
the docs and then is the LLM picking
that up and it's able to know what,
colin_1_02-05-2025_130726: I
would assume it depends on what
it's been trained on, right?
So that's if I just go off of pure
chat GPT, like sometimes it knows
what to pull from and it's not been
intentionally trained on just RDoX.
So it might see some example of someone
doing something completely different
somewhere else and just that's where
I don't feel like we're there yet.
and I'm okay with that.
that's not, I don't want
something to build me something
without my intervention at all.
cj_1_02-05-2025_160726: yeah,
colin_1_02-05-2025_130726: if I don't
fully believe that there's like this
whole thing about any app that you could
possibly think of, you can now just
prompt it or like it just generates it.
It's like we could have built all
the apps that we could possibly
have wanted and we haven't because
some of them are not wanted, right?
They're not needed.
There is no market for them.
So sure, maybe.
The perfect to do list app for CJ could
be prompted, and you're the only user
of it, and it's hosted who knows where.
you don't even have to care.
have you actually, I am now paying
for another one, which is v0.
Have you used this at all?
cj_1_02-05-2025_160726: I have.
Yeah,
colin_1_02-05-2025_130726: Yeah.
that one feels more like a prompting of
a product because, a repo is created.
server is credits hosted and then
it's I was doing something with
GitHub and a UI popped up and
said, what's your GitHub keys?
And I put 'em in there and they get
added to the environment variables.
I was like, this feels so cool.
And then it's pulling in Chad CN
and, tailwind and all that stuff.
But like, does it matter that
it did it or it told me to put
this in this folder somewhere?
it's the same thing.
cj_1_02-05-2025_160726: yeah.
I'm also curious to go back to
the WebSocket documentation.
Are you, generating and, Artifact that
describes the messages that go over
web sockets, like the async API spec or
colin_1_02-05-2025_130726:
There is no API spec for that.
API.
Yeah, because it's basically an RPC API.
And so we don't have an
open API spec for that,
cj_1_02-05-2025_160726: Okay.
Yeah.
We have a Kafka.
We have Kafka streaming that's
going to come out very soon.
And one of the things that
I've been curious about is
like, how do we generate How
colin_1_02-05-2025_130726: like a spec.
cj_1_02-05-2025_160726: do we generate,
yeah, SDKs or yeah, use the open API
spec to generate messages, or like the
shapes of the message types or whatever
that you can do serialize into so that
you have like proper types and stuff.
And it's my understanding
that a sync API or.
colin_1_02-05-2025_130726: yeah.
cj_1_02-05-2025_160726: or just like
putting it in your open API spec as like
the different objects, almost like web
socket payloads, but they're for messages.
colin_1_02-05-2025_130726:
Yeah, I haven't explored that.
But I imagine that the problem with
open API is I think it's meant to
be REST based, but I don't see why
you couldn't have here's our REST
API spec and here's our Kafka one.
cj_1_02-05-2025_160726: Yeah.
And then the other, like when you're
talking about web sockets, like there's
other stuff that's not just the payload
that you need, like the channel or
whatever that you're going over and yeah,
whether or not you have a heartbeat in,
colin_1_02-05-2025_130726: Yeah,
and like the fact that you, when
do you send an identify and then
a heartbeat to keep it going and
there's just so much detail in there.
I'm, those have been around for
so long that there's gotta be.
I just haven't touched that before.
I think of even SOAP, right?
It's these things exist.
gRPC, documentation,
all that kind of stuff.
cj_1_02-05-2025_160726: So is web sockets
the, that's like the primary communication
channel when you're building an activity.
colin_1_02-05-2025_130726: No, so we
call it the gateway, and it is, a way
for bots to listen to events in a server.
When your bot is in a million servers,
you're listening to the events,
whichever events you're scoped to.
So maybe it's, when a person comes
online, all the way down to someone
typing, which can be way too many events,
so you have to be careful of, like,
how many events you're listening to.
Maybe we don't even let you
listen to typing, because
typing is a little intense.
But we listen to it, because we
need to give you the indicator
of someone's typing, right?
Yeah, so we have that.
But if you go look at our gateway
docs, you'll see like it's very
hand curated, just payloads.
And if it's wrong,
someone does a PR to it.
We don't have a spec for that.
And they don't match our
rest endpoint payloads.
So they're a little different.
cj_1_02-05-2025_160726: One of the things
I'd love to do, it's to build a thing
that will go out, you can give it a
discord channel, you can give it a Slack
channel, you can give it a docs site.
You can give it a couple of like
social, handles or something.
And then it just builds like a giant
listener for you that can start to
understand sentiment and start to surface
here's common themes and common, yeah,
common issues that devs are running into.
Do whatever happened to orbit?
what is it?
Is it, do they take it all down
colin_1_02-05-2025_130726:
I think it's gone.
Yeah.
Postman acquired the team.
So they're building like developer
communities tools into Postman.
cj_1_02-05-2025_160726: orbit dot love?
Oh, it's still there.
Uh,
colin_1_02-05-2025_130726: I
don't know if you can log in, but
cj_1_02-05-2025_160726:
have a login button.
Oh no.
colin_1_02-05-2025_130726:
pour one out for Orbit.
cj_1_02-05-2025_160726: yeah.
colin_1_02-05-2025_130726: CJ
was one of our early users.
that even brings back to that idea of
the tool, like you were talking, the
top of the show we were talking about.
Read me.
do you I think, or what's the other tool?
Speakeasy, Fern, Stainless.
They're all getting into this
like SDK generation as the first
thing, and then they're like, Oh,
I guess we could also offer docs.
But do you want your tool to do everything
like halfway, or do you want like a
really good docs provider, a really good
SDK provider, I'm assuming that it's
like an expansion thing for them, right?
They're like, we're already doing
this, so let's also give you this.
cj_1_02-05-2025_160726: I think
that it is products that are.
Supported by an open API spec.
It's what are all the cool things we can
build downstream from an open API spec?
And so the, I think the
most obvious is SDKs.
And the second most obvious is an
API ref and like guides and docs
and all the other stuff around
the API ref is just expansion.
And then I think.
Like command line interface and then
developer metrics and, a playground
and
colin_1_02-05-2025_130726: hmm.
cj_1_02-05-2025_160726:
sandbox and like all this other
stuff like comes downstream.
So I think it's a lot of the things that
you would expect from, like when you're
in any dashboard and you click on the
developers tab, can these companies build
colin_1_02-05-2025_130726: Do all of it.
cj_1_02-05-2025_160726: products?
Yeah.
To solve all those problems,
like logs, webhook events,
colin_1_02-05-2025_130726: still
want to build that as a Ruby gem.
I think the challenge that I run into
is it's like Shopify is like, Oh,
we're going to give you a website, but
it's like the worst website, right?
And it's we're going to do your
e commerce, but then we're also
going to be your website builder.
Or, some of the website
builders got into e commerce
and it's not the best e commerce
experience, but it's really great.
And so I always get a struggle
with it where it's okay, you can
make the reference all day long.
I believe you can generate that.
But when you get into the guides, then
it's like the worst editing experience.
It's like half of a CMS, maybe it's
version controlled, stuff like that.
So I hope, I would love for one
company to be able to ship that.
And then you end up with all
of them competing against each
other on, stuff like that.
So I don't know.
I haven't looked at this anytime I go to
stainless or any of their websites, I get
a bunch of emails cause I think they're
somehow using some sort of tracking,
cj_1_02-05-2025_160726: in order for
us to build our own first party SDKs
and do a good job at it and build
and maintain a solid docs product,
it would require several engineers
and several tech writers and several
people focusing on that full time.
At Stripe, there was a whole team that
just built the documentation and a
whole team that just built SDKs and
manage that SDK generation pipeline.
And it was, three, four or five full
time engineers that worked on it.
And so if a company can really master.
The SDK bit.
And if the company can really master the
docs bit, then I think it makes tons of
sense to pay even like a full salary,
like a full time dev salary of one or two
engineers in order to not have to bring
that in house and to be able to meet.
External devs where they are.
The thing about these companies
too, is they'll continue adding
languages that you might not have
otherwise had the bandwidth to add.
So that's right.
For a long time, I wanted to add support
for rust as an official language.
We never got there, but the other
seven were very highly supported,
but like even thinking about or
considering adding another language
was just like out of the question.
So
colin_1_02-05-2025_130726: and any new
feature that OpenAPI supports in the
OpenAPI spec or anything like goes across
all seven languages, all seven SDKs,
it's not if you did these by hand, you'd
have to go, re implement all of them.
That feature in all of
them, which would be a pain.
So yeah.
And then, I've seen it too.
are you really wanting to reinvent
the wheel on how SDKs are generated
and create a pipeline and learn
all the things the hard way.
And then someone leaves the team
and you got to find someone else.
And it's a lot.
cj_1_02-05-2025_160726: totally.
Yeah.
And yeah, as technology
evolves in terms of.
The core of these SDKs.
do you support async with Python and
what kind of typing do you support?
And can you swap out the HTTP client so
that you can do like connection pooling
and like telemetry and like all this
other stuff, like keeping up with all
of those things in so many languages
and in different ecosystems is tough,
so yeah, just paying someone
else to take care of it seems.
colin_1_02-05-2025_130726: seems
obvious.
seems like a check mark in
the buy column for sure.
cj_1_02-05-2025_160726: yeah.
Yeah.
colin_1_02-05-2025_130726: I've totally
forgot to even share with you, but I
guess it's like we can wrap up here.
I'm gonna be speaking at
Write the Docs in May.
Yeah.
cj_1_02-05-2025_160726: No way.
Congratulations.
That's
colin_1_02-05-2025_130726: I don't
think the schedule has been announced
yet, but, so yeah, I'm doing a talk
on having your docs be open source.
so
taking PRs
and just being in the open, not
every company gets to choose
whether or not they're in the open.
but we'll talk about what we learned,
things that we wish we could do
better, things we want to do better.
but a lot of the benefits of bringing
the community with you in your docs
and you're not asking for free help.
You're part, you're building it together.
So yeah, excited to do that.
Yeah.
Some more docs things.
cj_1_02-05-2025_160726: So you at
the last one, didn't you volunteer?
Nice.
Okay.
colin_1_02-05-2025_130726: Yeah, I forgot.
I think it was, mostly technical.
I love how they were talking about this.
if you go to conferences, it's
always best to be involved somehow.
And, you can be in the audience,
you can get to see the talks, but,
if you're volunteering or speaking,
you're gonna just have a better
experience, because The conference
is putting you on a stage, literally.
And it, multiplies whatever, if
your company's sending you, cool.
If you're sending yourself, you really
want to make it worth your while.
It doesn't make sense if these videos
are going to be online for you to
go and then, not talk to anybody.
I think, yeah, I think having Volunteered
probably helped because I was able to
meet the organizers and I also when
I volunteered I was helping all the
speakers get ready and getting it mic'd
up So I got to meet all the speakers
which was also a win So yeah I think
highly recommend if you're trying to
get into it to look for a conference
you can spot like If you can't sponsor
one volunteer for one don't necessarily
recommend starting one from scratch if
you've not been but I've been there too
cj_1_02-05-2025_160726: Nice.
Are you also volunteering
again at this one
colin_1_02-05-2025_130726: No.
I'll be focused on just making
sure the talk is really good.
cj_1_02-05-2025_160726: that after, GDC?
colin_1_02-05-2025_130726:
Yes, thank goodness.
cj_1_02-05-2025_160726: nice.
Yeah.
Yeah.
That
colin_1_02-05-2025_130726:
Cause, Yeah, you mentioned
your event is in four months.
Our event is in one month.
So I have a month to get ready.
GDC is next month, and then,
May will be right the ducks.
cj_1_02-05-2025_160726: Got it.
Yeah.
You are in crunch time
right now then for sure.
colin_1_02-05-2025_130726: For sure.
But yeah, glad we could shake off
some of the dust on the podcast.
I hope people had a good, little season,
season three, break before season three.
This is technically the
third year of the podcast, so
cj_1_02-05-2025_160726: We're back.
colin_1_02-05-2025_130726: back.
February 5th.
I think, three years will be in August.
cj_1_02-05-2025_160726: Wow.
Thanks for listening.
Head over to build and learn.
dev for all the resources and things
we talked about in the show notes
and we'll catch you next time.
colin_1_02-05-2025_130726: See ya.
Listen to Build and Learn using one of many popular podcasting apps or directories.