April 28, 2025 • 46 mins
We all hear about AI every single day, but what are AI Agents and how are they used? Join us today as we talk with Rizel Scarlett from Block!
More about Rizel
X: https://x.com/blackgirlbytes
LinkedIn: http://linkedin.com/in/rizel-bobb-semple
Bluesky: https://bsky.app/profile/blackgirlbytes.bsky.social
YouTube: https://www.youtube.com/shorts/vHK9Xg_d6Sk
Block AI Agent: https://github.com/block/goose
Resilient Coders: https://www.resilientcoders.org/
LinkedIn
Follow us on
X: The Angular Plus Show
Bluesky: @theangularplusshow.bsky.social
The Angular Plus Show is a part of ng-conf. ng-conf is a multi-day Angular conference focused on delivering the highest quality training in the Angular JavaScript framework. Developers from across the globe converge every year to attend talks and workshops by the Angular team and community experts.
Join: http://www.ng-conf.org/
Attend: https://ti.to/ng-conf/2025
Follow: https://twitter.com/ngconf
https://www.linkedin.com/company/ng-conf
https://bsky.app/profile/ng-conf.bsky.social
https://www.facebook.com/ngconfofficial
Read: https://medium.com/ngconf
Watch: https://www.youtube.com/@ngconfonline
Edited by Patrick Hayes https://www.spoonfulofmedia.com/
Stock media provided by JUQBOXMUSIC/ Pond5
More about Rizel
X: https://x.com/blackgirlbytes
LinkedIn: http://linkedin.com/in/rizel-bobb-semple
Bluesky: https://bsky.app/profile/blackgirlbytes.bsky.social
YouTube: https://www.youtube.com/shorts/vHK9Xg_d6Sk
Block AI Agent: https://github.com/block/goose
Resilient Coders: https://www.resilientcoders.org/
Follow us on
X: The Angular Plus Show
Bluesky: @theangularplusshow.bsky.social
The Angular Plus Show is a part of ng-conf. ng-conf is a multi-day Angular conference focused on delivering the highest quality training in the Angular JavaScript framework. Developers from across the globe converge every year to attend talks and workshops by the Angular team and community experts.
Join: http://www.ng-conf.org/
Attend: https://ti.to/ng-conf/2025
Follow: https://twitter.com/ngconf
https://www.linkedin.com/company/ng-conf
https://bsky.app/profile/ng-conf.bsky.social
https://www.facebook.com/ngconfofficial
Read: https://medium.com/ngconf
Watch: https://www.youtube.com/@ngconfonline
Edited by Patrick Hayes https://www.spoonfulofmedia.com/
Stock media provided by JUQBOXMUSIC/ Pond5
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:08):
Welcome to the Angular plus Show. We're app developers of
all kinds share their insights and experiences. Let's get started.
Speaker 2 (00:21):
Hello and welcome to another episode of the Angular Plus
Show podcast. My name is Laura Newsome.
Speaker 3 (00:27):
I'll be one of your hosts today with me. Today
I have my co host Chao Chow. How's it going?
Speaker 4 (00:34):
Hey, he's doing pretty well. Nice spring weather.
Speaker 2 (00:39):
Yes, I've got my door open, so like, if you
hear birds, I doubt if you will, but if you do,
it's me because the birds are out and it's great.
Speaker 3 (00:48):
My other co host today is que Q. How's it going.
Speaker 5 (00:51):
It is going excellent. Thanks for asking, Laura. How are
you today?
Speaker 3 (00:54):
I am excellent as well. I'm doing well.
Speaker 2 (00:57):
And we have a guest today. Oh no, Risel, Oh no,
I did that right, Riselle Scarlet? Yeah right, okay, so
full the sculpture all afternoon. I've been saying Riseel, Riseell,
and so like, I'm so afraid.
Speaker 3 (01:12):
So it's Riselle.
Speaker 6 (01:14):
Don't know if you call me rise, I'm not gonna
I'm not gonna be like, oh my gosh.
Speaker 3 (01:19):
Yeah, that's how I feel about.
Speaker 2 (01:20):
Like. My name is l A r A and so
it's Lara but people sometimes call me Lara or Laura
and I'm like, I don't care, You're like things I'm
worry about. Soll welcome to the podcast. Uh do you
want to introduce yourself to the listeners?
Speaker 6 (01:38):
Yeah? So, like you said, my name is Roselle Scarlett.
I'm a developer advocate at Block. Specifically, I just became
a tech lead for the open source several team, so
Block is going all in non open source, which is
really exciting. Other things about me, I'm a new mom
and that's it I could think of off the top
(01:58):
of my head.
Speaker 3 (01:59):
Nice. Well, congratulations on the promotion. That's really exciting.
Speaker 2 (02:04):
Do you want to so you have the way that
you got into tech or you went through the Resilient
Counters program?
Speaker 7 (02:12):
Yeah?
Speaker 3 (02:13):
Do you want to tell us a little bit about that?
Speaker 2 (02:16):
Yeah?
Speaker 7 (02:16):
Definitely, so I did.
Speaker 6 (02:19):
So. Okay, when I wanted to get into tech, I
had initially been studying like psychology in college and I realized, Okay,
this is not going to work for me financially, because
I started asking around and people like, yeah, you need
like a master's to like start getting a real like
job in that, And I'm like, oh no, I don't
know if I have master's money right now, so let
(02:40):
me like step back and figure out. And then I
also ended up having to drop out of college. So
I like googled how to make a lot of money
and I love it. Like text things stuff started coming up,
and I was like, Okay, maybe I can like study
computer science. But I still couldn't go to college at
the time, so I I started looking up like free
(03:01):
ways to learn to code. Free code camp came up,
but also Resilient Coders came up. And Resilient Coders is
like this nonprofit coding boot camp that's based in Boston.
At the time, it was only based in Boston. Now
it's based in Philadelphia as well, and basically it teaches
black and brown folks to code, and you actually get
(03:25):
paid to learn. It's not like a huge amount of money,
but they're trying to basically offset the issue of like, Okay, one,
sometimes people have to pay a ton of money to
go to a coding boot camp, or even if it's
like something that's more free, you have to give up
your job to be able to do it or other
like times and hours. So like this was really good
(03:46):
for me. I did about like a fourteen weeks and
then jump straight into software engineers.
Speaker 3 (03:52):
Nice. Nice, that's awesome. Yeah, I know.
Speaker 2 (03:54):
I had a friend who did a coding boot camp
and he didn't have to pay up front, but afterwards
he had to pay it all back like really fast,
and he like he was literally like living off the
free snacks that he was getting at his first job.
Speaker 3 (04:09):
I'm like, Jane, dude, that sucks.
Speaker 2 (04:11):
But yeah, so there's sometimes there's strings attached. So it's
amazing that there's a program out there attached, no strings attached,
and in fact they hand you some strings and they're like,
go ahead and pay some rent.
Speaker 3 (04:24):
That's amazing. That's amazing.
Speaker 2 (04:26):
If any of our listeners are interested, we'll drop a
link to Resilient Coders in in the chat. But what
we have you on here today to talk about are
AI agents, So we have obviously I think you'd have
to be living under a rock in tech to not
have somebody talking about AI in your life. So so okay, so,
(04:53):
how how hard is your job going on AI?
Speaker 7 (04:59):
It's all I do now.
Speaker 6 (05:01):
So my company actually builds an AI agent, so and
it's open source, so my company is going all in
and my jobs really focused on on championing, championing that
for developers and other people just helping to make their
lives easier. And I'm excited to talk to you about
(05:22):
it as well. For about AI agents in general.
Speaker 2 (05:25):
Yeah, okay, so for our listener who listeners who may
not know because there's there's a lot to AI. It's
not simply like our managers think, which is just simply.
Speaker 3 (05:35):
Do the AI. What is an AI agent?
Speaker 6 (05:39):
Yeah, So an AI agent, I would describe it as
a software that does the task for you. So what
we've been familiar with with these past few years with
lms and stuff like that, it's like we can talk
to or interact with an LM like GPT four ZHO
(06:01):
so on at three point five. These are sometimes they
have like an interface on top of them. Maybe that's
like chat GPT or anthropic or even get hipcopilot. So
you talk to it or you should write a message
and it does a whole bunch of math in the
background to figure out what's the best words to respond
to you with to help you answer your problem. But
(06:22):
with AI agents, it's that plus an additional layer where
it can go out heead and execute the task for you.
So there's something called an agentic loop, and there's different kinds,
but oftentimes it may look like, Okay, your agent it
like I think of it, like the LM as the
brains and the agent as the one that acts on things.
Speaker 7 (06:45):
Or even.
Speaker 6 (06:47):
This might be a weird analogy, but like the movie
ratituly or a little rat inside that guy's head or
outside his hat, and he like the rat's the one
making the moves or the rat's the one like making decisions,
and the chef is the one just being the puppet
kind of, so the agent is the chef knows nothing,
really doesn't know how to cook. It's like rat, tell
(07:09):
me how to cook, and the rat is the m hands, yeah,
tell me what to do, and the LM it has
it's the brain. It's the mastermind behind all of it.
And it's like, okay, I see that you have access
to these different capabilities and the user wants to Let's say,
(07:31):
the user wants to like build a web app or
book a flight or something like that. So the LM
looks and it's like, oh, okay, you're able to go
to Google flights and get this, or you're able to
run a shell command and build a web app or
create a file or something like that, so it'll tell
it here's the plan of what we're going to do,
(07:51):
here's the files that we're going to touch, or the
commands we're going to do, and then the agent simply
just acts it out for you, and then the user
just gets to sit back and relax and give more
feedback on like is this what I wanted to happen?
Or do I want more follow up? I hope that
made sense.
Speaker 3 (08:11):
It does. And I love the ratatuy reference because.
Speaker 2 (08:17):
I feel like it's just kind of it is hard
to get a mental model of some of this stuff,
so hearing it spoken about different ways it's helpful to
kind of help people get their heads around what is
going on there.
Speaker 5 (08:30):
Exactly that your first time using that ratitude reference or
is that like in your back pocket?
Speaker 6 (08:37):
Is in my head and I've been trying to put
it together, So my first time saying it out loud,
I gotta work well.
Speaker 2 (08:44):
And at landed with me because I have kids that
are exactly the age like they were the right age
when ratatuy came out, So like we've.
Speaker 3 (08:52):
Watched it a lot, so.
Speaker 7 (08:56):
Perfect.
Speaker 3 (08:58):
I love it.
Speaker 2 (09:00):
So Okay, so you're working on an AI agent. So
you were talking about like it kind of sets restrictions
for what it can do.
Speaker 3 (09:08):
Why do we need those restrictions?
Speaker 6 (09:11):
Okay, so AI agents, Well, when I say restrictions or
what capabilities it can do, I would to say it's
necessarily setting restrictions. It's just seeing what it has the
ability to do, if that makes any sense, like whoever
developed it? But I guess we would need those restrictions
because an AI agent, you don't want it to go
(09:34):
crazy and go off and do the wrong things. So
a lot of times when people build AI agents, they
may build like restrictions to make sure that like, oh,
maybe it's not necessarily completely autonomous, maybe asks you.
Speaker 7 (09:46):
For approval more.
Speaker 6 (09:48):
But what I described as the capabilities I wouldn't call
those restrictions. In the AI agent world, people call this
like tool calling or function calling, and these are just
like the apps or data that it has access to,
not necessarily restrictions, but more I will say, more freedom
if that makes sense.
Speaker 2 (10:07):
Yeah, yeah, okay, So it's more like the agent is
like you're just kind of fine tuning.
Speaker 3 (10:13):
I mean, like MS can be.
Speaker 2 (10:15):
Huge, right, Like they can be like literally the whole Internet, right,
So you know, if you go out to do a
very specific task, then of course you're going to want
something to be like, Okay, maybe you don't need to
read all of.
Speaker 3 (10:31):
This as well. Ocus here.
Speaker 6 (10:33):
Yeah, yeah exactly, I see what you're saying. Yeah, instead
of that's a good point, instead of using up the
LM's time and energy and even what's called the context
window looking at every single thing on the internet, it
can go ahead and find that specific tool. It's like, oh,
like Laura wants to book a flight, I'm not going
(10:54):
to look at Google Drive or whatever other stuff has
I'm going to look at Expedia dot com or Google
Flights or something like that to help her find the
flight within her her limitations or within her criteria.
Speaker 7 (11:09):
Yeah, I'll wait for you.
Speaker 6 (11:10):
I have more thoughts, but I'll wait for you to
ask questions before I just like word dump everything.
Speaker 3 (11:17):
I'm like, just start word dumping, like I'm into that.
Speaker 6 (11:19):
Okay, I can't. This is where this is where something
called model context protocol comes in. So I don't know
if you've seen people like talking about MCP's model context protocol.
So AI agents like they've been trying to be a
thing for a little bit.
Speaker 7 (11:35):
But because.
Speaker 6 (11:38):
AI agents ecosystem at first was a little bit like
closed off. It's like all right, it's cool, but it
can't autumn Like I still feel like I have to
copy and paste like I did with chat GPT because
it doesn't have access to I don't know, like whatever
platform I'm using, maybe it's work Day or Google Drive
or Snowflake or whatever. I don't know what tools people
(11:58):
are using in their day to day. So the AI
agent wasn't that helpful. But now with Model Context Protocol,
which is an open standard that allows you to have
like or allows your AI agent to get context from
different data platforms, you can literally use those as like
bringing in those tool capabilities to your AI agent. So
(12:21):
literally people around the community, every developer is just like
coming out with their own MCP server, whether it's like oh,
here's the MPCP server for Notion and then you plug
that into your AI agent and now you can tell
your AI agent like build me this Notion page or
look through all of my Notion files or something like that.
Or companies are as well, or like hey, like Slack
(12:43):
is like I'm building we have our own MCP server,
and you can be like AI agent connect to my
my Slack MCP server and go schedule this message for
this time or make a summary of all this, this
entire thread that I missed during this meeting. So that's
that's like my brain dump right now.
Speaker 3 (13:00):
I love it.
Speaker 6 (13:00):
I love it.
Speaker 5 (13:01):
Yeah, this is that's awesome because we just so we
had pieces for developers here and they talked about m
cps as well, Like I that kind of just like, ah,
I remember that because you can you can tie in
context based on what I d E. You're using.
Speaker 3 (13:16):
So that's cool.
Speaker 6 (13:18):
I'm glad you brought them up because they built an
m CP server and Goose is an MCP client, so
you can we were like working with them to like,
oh MCP server to Goose, which is the AI agent
my company is working on.
Speaker 4 (13:32):
Yeah, our company also has an uh an MCP.
Speaker 8 (13:36):
Oh really yeah it is.
Speaker 4 (13:38):
So I worked for all an X, which is like
a like a death tool thing like a like so
we build like an n X MCP so that ll
M or whatever like AI tool do you connect to
it can't understand more about your workspace, project instructures and
all of that stuff.
Speaker 6 (13:56):
I love that everyone's going all in on MCP servers.
I feel like this is kind of cool because, like
I think when AI first came out, it was like, Okay,
this is cool. But then I don't know if y'all
felt this way, but like the excitement kind of died
down a little bit because I'm like, all right, everybody
can write to chat, GPT or whatever it is and
get back text.
Speaker 7 (14:15):
That's cool.
Speaker 6 (14:16):
But now it's like, oh, we can build something, and
just seeing what creations people come up with and how
we can all mix it together, it's kind of exciting.
Speaker 9 (14:25):
Yeah.
Speaker 4 (14:26):
I do feel like it's starting to come together.
Speaker 2 (14:29):
Yeah, yeah, I agree, Yeah, I I you know, even
I'm like, so I'm gonna I can't lie.
Speaker 3 (14:34):
I've been like very I am.
Speaker 2 (14:35):
I feel like a grandma when it comes to AI
stuff because I'm like, oh, there is just like so
much to learn, and then like you'll be at work
and they're like, oh go do go tell us, go
write tickets.
Speaker 3 (14:46):
For this stuff, and you're like, but I need to
learn all this other stuff.
Speaker 2 (14:49):
So I even I am feeling like, Okay, it's starting
to make sense how these tools can be useful, yeah.
Speaker 3 (14:56):
Because it really did.
Speaker 2 (14:57):
At first it was like okay, yeah, I can use
this to generate some data for me and like proof
read my pape.
Speaker 3 (15:03):
Did I make any grammatical errors? Like you know, I'm like.
Speaker 2 (15:07):
Those tools are important too, like because I like, yay,
I don't have to think as hard about my grammar.
But yeah, just finding these ways to make these really
just sort of highly tuned tools to help your workflow.
Speaker 6 (15:21):
It's like we're at another level because, like you just said,
like when before it maybe felt, at least for me
maybe a little bit like we're saying this is saving
my time, but it's not, because now I have to
copy and paste and then I have to edit. But
I'm like, now I'm like, okay, you have context.
Speaker 7 (15:39):
Now I don't have to like copy and paste.
Speaker 6 (15:41):
Okay, here's what I'm doing. This is what happened first.
And I'm like, it's so much worse, right.
Speaker 2 (15:46):
I know, Like I was, I remember trying to like
just give the thing enough context to even come close
to getting back and answer that you wanted. And it's like, dude,
I have to know so much about this, Like by
the time I am able to give this thing enough
context to be able to give me anything relevant back,
I might as well have just done it myself.
Speaker 4 (16:09):
I have I have a question just out of curiosity.
So when you say I build AI agent, what does
this like? What does this mean?
Speaker 8 (16:18):
Really?
Speaker 4 (16:18):
What goes into building an AI agent?
Speaker 6 (16:22):
Okay, I didn't build the AI agent, Just to clear
I'm a developer advocate. My team built the AI agent.
But from what I Okay, first off, y'all can look
at the code. If you you might, you might be
way smarter than me. It's an open source project called Goose,
so givehip dot com flash block slash Goose. But what
(16:43):
they've done is they've created two interfaces. We have like
uh desktop UI interface it's built with Electron and React
and typescript and all that, and then we have a
RUST interface. And from the way I understand it, it's
used agentic loop where they have like functions that are
(17:03):
waiting listening for what the user types in or request,
and then the agent goes and acts upon that request,
reaches out to the MCP server that's plugged into it,
and then it comes back with whatever.
Speaker 7 (17:17):
Tools it has.
Speaker 6 (17:18):
It sends each time the entire context of the conversation,
so the AI agent continues to have that context, and
then it comes back with something for you based on
the plan, and it keeps iterating on that, so like
if something broke or like the command that ran didn't work,
the AI agent will go back and be like that
didn't work to the LM, and we'll go back in
(17:39):
that loop. But I don't know specifically because I didn't
sit there and build it. I've contributed a few things
like oh, let's make this dark mold look a little
bit better.
Speaker 9 (17:51):
Broke, But yeah, yeah, I mean, like I mean, I'm
more like, let's say the company or like the group says, oh,
we want to be like an AI agent?
Speaker 4 (18:05):
What uh what what needs to go?
Speaker 3 (18:08):
Yeah?
Speaker 4 (18:09):
What what the process?
Speaker 8 (18:09):
What do we need?
Speaker 9 (18:11):
Uh?
Speaker 4 (18:11):
Like do we need to have like an LM available
so that we can talk to it. What kind of
like protocols that we have we need to to to
to start investigating or researching to be able to be
like an AI agent.
Speaker 6 (18:24):
Yeah, I mean I can say what I've seen in
the middle or in the early beginnings. I kind of
one off on maternity leave in between it. But from
what I seen, we have started off with like a
more rudimentary version of an AI agent that was just
built in Python, and it was a command line version thing,
and we were just using it internally to help automate
(18:44):
our migrations, and then we were like, why don't we ope.
Speaker 7 (18:47):
A source this. This is really helpful.
Speaker 6 (18:49):
And we can give it more capabilities. Then we also
implemented like why not instead of like being stuck into
one lll M people like the for an lllms, why
don't we allow people to bring their own lms. So
the way it works when you use it is like
you can choose I want to use GPT four point
five or whatever it is, bring your API key into that,
(19:13):
or maybe you want to use OLAM or something. And
then the other thing we thought about, because block is
really about like protecting your data and your privacy, we're like,
let's make it local to your machine. Like a lot
of AI agents right now, they're they're in browser.
Speaker 7 (19:32):
Your data is.
Speaker 6 (19:33):
Going out to somewhere you're not able to use it
at your company.
Speaker 7 (19:36):
So we're like, let's do that as well.
Speaker 6 (19:39):
And there was another thing I was going to say,
let's make it local to your machine, and the thought
went away from me.
Speaker 2 (19:50):
That's all right, Like the local thing is a big
deal though, like because I think that's a huge hurdle
for companies that you know they want to use AI
for things. They want their AI tools to be useful,
but you just can't risk data like you just can't.
Speaker 7 (20:08):
Yeah, that makes it more usable for people. Actually, sorry
to interrupt you. I just remembered what I was gonna
say now.
Speaker 6 (20:14):
And then the version two is when we started working
with Athropics, So we actually helped to build a spec
of MCP with and Thropic, so that that was like
the round two of like, Okay, now we're gonna like
sorry that I interrupted you, but no.
Speaker 3 (20:29):
No, I actually was going so that you might remember
because I you.
Speaker 6 (20:35):
Now, we're gonna build this in Rust would rust back
in and we're gonna make this be an MCP client and.
Speaker 7 (20:41):
Bring an MCP servers there.
Speaker 6 (20:43):
So we were actually from my perspective, I think we
were one of the first MCP clients. And then now
MCP is starting to pick up pace and other companies
are like, oh, we're MCP client too, or we're adopting
this as well. So to me that was like kind
of the process of like figuring out, like you said,
like what are people what's the experience of AI agents?
That's not been great. It's not been like the data
(21:05):
is being shared places it's not easy to connect your
AI agent to other tools, and it's sometimes you want
a specific LM because different lms have different like strengths
and specialties.
Speaker 3 (21:18):
Yeah, yeah, and you can see that and some you know,
like some of the platforms or you can con where you're.
Speaker 2 (21:24):
Like, oh, I'm using Cloud for this and I'm using
chat GPT for this, And I mean there's a huge
difference even with just like copilot and.
Speaker 3 (21:31):
Agentic mode, Like.
Speaker 2 (21:34):
I get way better answers from Cloud than I do
from chat GPT. So yeah, it can definitely make a
big difference to fine tune it for what your industry
is doing.
Speaker 6 (21:45):
So exactly exactly or any task like like you said
with claud, I find it better for like summarizing things
for me maybe g chat GPT, I might use it
like one mini and stufficked that more for like reasoning tasks.
Speaker 3 (22:03):
But yeah, nice, nice.
Speaker 2 (22:07):
So one thing that actually one of the reasons I
reached out to you was because you had posted you
had posted something on LinkedIn about using AI as a
democratizing force, and so one of the things I think
(22:29):
we worry about, you know, with AI, it has a
tent like since a lot of these models are trained
on the Internet, and the Internet is an inherently, frankly
a terrible place sometimes depending on what corners you end
up in, what are.
Speaker 3 (22:44):
How can we use AI for good? I guess is
is a question that I was wondering about.
Speaker 6 (22:53):
First off, we have to educate people on like how
to use it well, because I do think we sometimes
get too overly excited and then meet people in the
wrong direction, Like the vibe coding thing. It's cool, but
then when you're a less experience it's cool and it
allows inexperienced engineers to build. But when you don't know
(23:15):
what to look out for, you might be building something
that can easily break, or you just might be opening
yourself to security risk. So that first thing is just
like educating people. But I think AI is a democratizing
force because it lowers the barrier to coding in general.
I don't I don't think that coding needs to be
(23:37):
this like high up thing that like, oh, only like
special people can do it. Like people have ideas and
they want to bring it to life. And I was
talking to Rachel Neighbors the other day and they had
a really good perspective on this. They were saying like
in the first and first in like our first world,
we're thinking of like, oh my gosh, but what problems
(23:59):
will AI bring? But in other countries they're like this
reduces like the gap between like girls and boys in school,
or just in different countries of people getting to learn
to code, they may not have that access to education.
So we need to start thinking of it that way.
And in the first world, just work on fixing those gaps,
(24:23):
those security gaps or whatever other issues may happen. But
it's not like an inherently bad thing.
Speaker 7 (24:28):
I feel like old, but I hope that made.
Speaker 3 (24:30):
Now, that makes it.
Speaker 2 (24:32):
I love I love that that sentiment because I think
it's easy to always you know, I live.
Speaker 3 (24:38):
I live in the United States. I was born and
raised here.
Speaker 2 (24:40):
Like I I my family has always been sort of
just middle class, Like there's a lot of stuff that
it's easy to forget some of the barriers that people
face in the world. And I do think that that
AI does offer the offers. Early career devs are people
(25:01):
that are learning another tool to use. And one thing
that I've been doing at work is trying to figure
out ways what is the process now, Like, Okay, we
just hired a junior dev, We're like, go use agentic
Mode and co pilot and use it to write your code. Yeah, okay,
So we've encouraged them to do that, What is the
(25:23):
next step, right, because obviously all of that code it
might be fine, it might not be fine.
Speaker 3 (25:28):
So as a senior engineer on the team, what do
I need to do to then take it to the
next step?
Speaker 5 (25:37):
You know?
Speaker 2 (25:37):
And so one of the things that we've I suggested
that we're we have to like actually be raising poll requests.
But you know, it's like, okay, encourage people to explain
what their code is doing. Like, okay, agentic Mode has
written it, now explain what it's doing. Yeah, you know, because.
Speaker 3 (25:56):
I you know, I grew up and I grew up.
Speaker 2 (25:58):
I learned how to code and world of stack over
but overflow where I would just go through and uh
copy paste, right, I didn't always understand the code.
Speaker 3 (26:08):
I didn't have anyone to explain it to. No one
could ound like that's crazy, don't do that.
Speaker 2 (26:14):
But you know, so I think that's a it's an
important responsibility of senior members of the team, and it's
actually I think gives then early career depths an opportunity
to level up faster because you know they're already getting
a sort of decent example, and then how can you
then drive that home that like, Okay, that's a good example,
(26:34):
here's why, or okay, that pattern will cause us problems
this way.
Speaker 7 (26:41):
Yeah, I really I really like that.
Speaker 6 (26:42):
Especially that makes I mean, that makes me also think
it can help a junior engineer to also exercise like
pushing back a little bit if they push back on
an AI agent, like being like why did you make
that decision? Or secure or something like that, like just
asking the AI agent not just having it produce things,
but trying to learn from its decisions. And then I
(27:04):
don't I don't know, I've never seen this happen, but
maybe it'll make them more confident when they have to
interact with their their more senior team members and be.
Speaker 7 (27:12):
Like, oh, why do you make this decision?
Speaker 6 (27:13):
Because I remember when I was more junior, if a
senior team member made a decision in their code, I
will be like, looks good to me, because I'm like,
of course they know more, they're smarter.
Speaker 3 (27:23):
I don't know, right, I know, I don't want to
argue with them. I know, yeah, I had I had
the I had it was like.
Speaker 2 (27:29):
Weird but kind of like one of the best junior
dev experiences on our team. Uh this last year I
had three devs that would be like, why are you
doing it that way?
Speaker 3 (27:39):
Like because but it was great because they were.
Speaker 2 (27:42):
Totally willing to ask questions and push back and you know,
and then it also helped me because I'm like, why
am I doing it that way? But I think in general,
any time you're using anything that's AI generated, the first
thing you should think is is that true?
Speaker 6 (27:57):
Yes? Yeah, because it could be hallucinating exactly exactly.
Speaker 2 (28:02):
I know I use AI a lot, you know, like
it's hard to not use it now when you're googling something, right,
I google something and AI is like, oh, this is
how it is, and it's so confident, and you're like.
Speaker 3 (28:11):
But is that is that? Is that true? Is it relevant?
Speaker 7 (28:14):
Still?
Speaker 3 (28:14):
Is it current?
Speaker 6 (28:15):
Like because it's just summarizing the whole bunch of information
that it might have found. It made me lazier, though,
and I need it.
Speaker 7 (28:22):
I need to keep pushing.
Speaker 2 (28:25):
It's real hard because like I'm just like, yep, looks
good to me. Yeah, nice, nice, Okay, So let's talk
a little bit more about Goose.
Speaker 3 (28:37):
So if somebody what how are people using Goose? Like
what is it? How does it get used?
Speaker 2 (28:43):
Yeah?
Speaker 6 (28:43):
I mean right now mostly for like it's it's heavily
used by engineers, but it doesn't have to be. It
can do non technical workflows as well.
Speaker 7 (28:52):
I can't talk.
Speaker 6 (28:53):
There's people that are doing such awesome things, like.
Speaker 7 (28:56):
Some people at my job have I don't know how
like evil people do things, but I was like listening.
Speaker 6 (29:03):
To them and they're like, apparently like one law legal
document got updated and they want to like compare what
the differences were without like reading through the whole thing,
so they would like ask Goose to like come up
with or like just generate like the top things that
got changed within that legal document. And I was like, oh,
that's interesting. I know different ways that I've used it.
(29:24):
Is like since I'm a developer advocate, I'm coding but
also doing other things. So like let's say I have
a live stream that's coming up and I want to
interview the guests and like make the questions good. I've
done like okay, I've like searched for some YouTube videos
that the the guest has been in before, maybe like
live streams or conference talks, and I've asked Goose to
(29:46):
like come up with like some questions that the person
will be passionate about based on this conference that they
were in or whatever, and they'll like generate a transcript,
it'll generate the agenda, and then I'll also be like, hey,
like here's some idea as I have for.
Speaker 2 (30:03):
Like a.
Speaker 6 (30:05):
Title, but I want something more punchy, so it'll help
me with that. Or it'll give me like ideas for
a thumbnail. So I know those sound like chatchypte like,
but without me. It's without me actually having to like
directly give it the videos, I'll be like, look up
like Rachel Neighbors or this person from open Ai or whatever,
and I'll like generate some questions based on that.
Speaker 7 (30:27):
Let me see what are other things I do.
Speaker 6 (30:29):
A lot there A lot of people are using it
for like Google Drive, Like they'll like give it access
to some of their documents and they're like, hey, like
summarize this, or create some meeting notes on this, or
create an Excel chart for me or Google Sheets chart
for me based on the data that I've found. So
(30:49):
it's it can go from like very technical to like
other like daily jobs, or there's even like silly stuff
that people have been doing just for fun, Like they've
used like an MCP server that makes phone calls for
you and they're like, hey, goo's call and say like
a funny joke or something. So those are just for
like fun demos.
Speaker 3 (31:08):
Yeah, I'll harass my senators for me. Yeah, oh my god,
that's nice. I feel like I'm asking all the questions
chow Q, what do you got?
Speaker 5 (31:24):
No, I mean I'm learning because this is all yeah yelling.
Speaker 4 (31:32):
I'm just learning as well, because I mean I'm probably
using AI agent without knowing it.
Speaker 3 (31:38):
Yeah, so we are you?
Speaker 5 (31:41):
Are you using chapt four point zero?
Speaker 4 (31:44):
No, I'm using uh windsurf.
Speaker 5 (31:47):
That's probably that's probably one.
Speaker 4 (31:50):
Uh I'm using I'm using like cloud cloud call model.
Speaker 5 (31:55):
Yeah.
Speaker 3 (31:57):
Yeah.
Speaker 7 (31:58):
That was my favorite because, in.
Speaker 4 (32:00):
My opinion, I think Kloleu is probably one of the
best at implementing code at the moment, in my opinion.
In my opinion so far, I think I've heard.
Speaker 3 (32:12):
Yan say that too before.
Speaker 2 (32:13):
So now Nicholas is one of our other co hosts,
I'm almost positive he's also said.
Speaker 5 (32:18):
That, so I think he has before too.
Speaker 3 (32:22):
Yeah.
Speaker 2 (32:22):
Yeah, yeah, I think it's it's we are just in
this really interesting spot I think with AI.
Speaker 3 (32:27):
Where I mean, so.
Speaker 2 (32:30):
At Cisco they had all these trainings set up, like okay,
let's go, everyone should level up on AI, right, they
put together this training course, and these courses are less
than a year old, and they already feel outdated.
Speaker 6 (32:42):
Oh wow, it's moving fast.
Speaker 2 (32:44):
Yeah, you know, I mean like some things are there,
you know, like, Okay, this is what prompt engineering is.
This is how you can, like people can jail break
a model, and this is some of the ways that
they can do that. And like knowing that stuff is
good to know. But then like it just feels like
so it So we're all Angular developers, obviously, it's the
Angular plus show. Angular releases a new version every six months,
(33:07):
and it already feels hard to keep up with Angular,
but now we're also keeping up with AI, which honestly
just feels like it's going faster. And probably that's because
all of our companies are throwing just billions of dollars
at it.
Speaker 4 (33:22):
So I told, yeah, sorry, go ahead. I totally angler team.
I totally Angler team to just hey, you're Angler to
cel I just drop the micration story, build an MCP
sver build like an angler and an angler MCP server,
so that when you have like a new version, we
(33:44):
can ask g A I to migrate.
Speaker 7 (33:47):
Right, that's genius exactly.
Speaker 4 (33:52):
Here's here's another idea that I kind of like run
into because because of my workflow is that I'm using
two models like consistently, one with a DIP think capability
and one for like implementing code specifically.
Speaker 5 (34:09):
So my workflow is that.
Speaker 4 (34:11):
I switched between the two and then I have to
reprovide context for the two. It'd be awesome if these
two agents I don't know agents or models can.
Speaker 8 (34:21):
Talk to h other.
Speaker 4 (34:22):
So like hey, uh, let's say let's say Goose. Let's
I'm gonna pick like AA agent Goose, Hey Gooz, I asked, uh,
DIP think or DIP sick this question it comes up
with like an implement implementation plan? Can you do it
with using call?
Speaker 2 (34:38):
I was thinking about like because at work we're building,
like we have to really make sure that the data
that an agent can access is like contained, you know,
like we need to make sure that like whatever it
responds with, the person has permission to see.
Speaker 3 (34:54):
It's not like bleeding in the other parts.
Speaker 2 (34:56):
Of the data, but like like a Kubernetes for AI,
like some sort of centralized hub that just knows where
to go to ask these questions. So yeah, somebody can
just go ahead. I think there's something kind of like that,
but I don't know what it is.
Speaker 7 (35:15):
That would be cool.
Speaker 6 (35:17):
Yeah, I think for for child, I feel.
Speaker 7 (35:20):
Like you could do that with Goose. If you had
a session and you had.
Speaker 6 (35:25):
Access to both l ms, you can just switch it
and be like, okay, I'm going to use this LEM
for this, but you don't have to get out of
the session.
Speaker 7 (35:35):
Yeah, you can just switch.
Speaker 6 (35:36):
And then it'll start responding with that.
Speaker 5 (35:39):
But if there was a Kubernetes, you're saying, the context
would still be there for both.
Speaker 6 (35:44):
At that point the context would it will still be
in the sane ecosystem.
Speaker 4 (35:49):
Yeah, okay, cool that that'd be great Because I used
like h l ms with dip think capability to plan
something like reason I say that this sound that does
this reasoning sound enough to start implementing it, and it's
been pretty good and better at the outer other ms.
Speaker 3 (36:12):
Yeah, yeah, nice.
Speaker 2 (36:14):
I need AI where I can feed it a pigma
drawing and then fit another figma drawing and be like different.
Speaker 8 (36:22):
I think you could do that.
Speaker 6 (36:23):
I'm like, I'm like, I'm not trying to Like, I
think you could like use the Figma MCP server and
then just be like hey, here's me.
Speaker 3 (36:31):
I just we're gonna bring you on.
Speaker 2 (36:32):
Hold on, I'm gonna go and get a list of
questions you're coming on every week?
Speaker 5 (36:36):
Yeah, hang on, You're you're saying that you can do
that with what the what's the Pigma thing.
Speaker 7 (36:43):
Server?
Speaker 6 (36:43):
Yeah, we have a I mean it's not doing Laura's
exact case, but we have a video of like using
an m c P server with Goose. I'm gonna just
drop it.
Speaker 7 (36:56):
In the chat.
Speaker 4 (36:57):
Do it?
Speaker 3 (36:57):
Do it?
Speaker 2 (36:58):
Yes, because I'm taking this to work because what always
happens is that they'll be like, go see what you
need to build and then like a couple of days later,
they changed the drawings, but they don't tell me how,
and you know, like the history is not like I
don't know.
Speaker 3 (37:11):
It's hard.
Speaker 5 (37:12):
Yeah.
Speaker 2 (37:13):
One of the most frustrating parts of my job is
playing where's waldough of like where are the changes?
Speaker 3 (37:18):
Like good luck? Was that there before? I don't know.
Speaker 5 (37:24):
Yeah, we get hit with that often, like hey, that's
that should be a link when the link when when
I designed it?
Speaker 2 (37:31):
Shots now and put it on our tickets, so we
can be like, when I wrote this ticket, this is
what it did.
Speaker 3 (37:36):
Okay, I watch this.
Speaker 6 (37:37):
It's not it's not your exact use case. In the video,
my coworker just uses the pigma file to.
Speaker 7 (37:43):
Create a web AAP.
Speaker 6 (37:44):
But I'm literally like, I'm like, I'm sure you could
do that. You can just be like, here's the two
here's the Pigma file. How did it change without like
having to? I think you just give it your Figma
API tokyo.
Speaker 3 (37:55):
I think I'm going to play around with an Oliver
report back.
Speaker 2 (38:00):
But yeah, it's I think that, Like, I think I
can be excited about AI the more it starts solving
these terrible problems I have at work, right.
Speaker 7 (38:13):
When when the more its useful, it gets.
Speaker 2 (38:15):
Right right I have, I'm not really excited for it
to write all of my code.
Speaker 3 (38:20):
For me, reviewing code is not my favorite job. Like
what wanted to write all the code?
Speaker 2 (38:26):
And I just have to review code for the rest
of my life. That sounds a little bit terrible, But
to lower the bar on you know, some of the
stuff that's just takes time that's tedious, Like that's fantastic.
Speaker 4 (38:41):
Yeah, you know, you know what like what guests meto.
What pisses me off is that people use AI uh.
Speaker 2 (38:47):
And then.
Speaker 4 (38:49):
The the result they don't have a human touch to it.
So what what the AI produces? They can they produce
that entirely, so from the without touching it, like without
putting there like.
Speaker 6 (39:04):
Flair to it.
Speaker 5 (39:05):
Yeah, and everyone has their flare. You know you can
tell like all that Laura wrote that code, like I
could tell.
Speaker 2 (39:12):
My unit and see what kind of example, like what
the mock data looks like?
Speaker 5 (39:17):
So all everyone has their tails. But yeah, and then
and then you get an AI AI has its tail too.
You're like someone someone colly pasted that from from chat, DBT.
Speaker 6 (39:27):
Or even when it's not code, they just wrote something.
They're like in today's long, everlasting world.
Speaker 4 (39:34):
Yes, exactly what a my co workers? What did my
co workers start like generating their pr description entirely with
a I.
Speaker 5 (39:46):
Our manager told us to do that too, they because
get lap has its tone. You can just summarize my
MR for me with based on the ticket and the code.
There's no way I can't be caught.
Speaker 3 (39:57):
I don't.
Speaker 5 (39:57):
I don't know heop like it you used. I don't know.
It's something if you feel dirty about it, I don't know.
Speaker 3 (40:02):
I don't know.
Speaker 5 (40:03):
At some point I'll get over it, but.
Speaker 2 (40:04):
Like maybe the at least initially it to be like
I did read through this, I did.
Speaker 7 (40:12):
You guys are good engineers.
Speaker 5 (40:15):
Now I won't want I do use I do use
the AI like I've been using developers for our pieces
for developers a lot, because I don't write LUA normally,
but my kids like to play roadblocks and all the
code is written in LUA, and so I'll just say, hey,
I don't want to do it this way an angular,
I do it this way? Can I do it this
way in Lula? And it's like, well, you can do it.
(40:36):
People don't generally do it this way, but I'll just
it works for me, and it's and it's been it's
been nice to kind of learn, yeah, arcade code.
Speaker 6 (40:45):
That's what it's good for for me too, is like
discovering maybe technologies I'm not used to working with.
Speaker 7 (40:51):
I just want to like do an experiment real quick.
Speaker 6 (40:54):
Nothing for production, yeah, yeah, I do.
Speaker 2 (40:59):
I have learned to like it. Okay, for writing unit tests.
I love writing tests, but I don't really love writing
them that much.
Speaker 3 (41:07):
So it is nice to be like, what looking at
this file, what coverage am I missing?
Speaker 2 (41:13):
Yeah?
Speaker 3 (41:13):
That's solid. Yeah, I'm like, okay, I can get behind that.
Speaker 2 (41:17):
And then we we put a thing in ours where
we've instructed it to read our read mews first. So
I spent all this time documenting how we intend to
write the code in our application, but then we go
tell our junior devs to use Copilot. It's like, but
we need Copilot to know that, Like, that's not how
(41:39):
we do it. This is how we do it, and
this is where you can get that context from.
Speaker 6 (41:43):
So right, Cursor and Goose are probably good at that
because you can give like Cursor rules or Goose hints,
and you can be like, this is our stylistic way
of doing things that Cisco on this particular I don't know.
Speaker 2 (42:01):
Yeah, yeah, no, I mean that's true because it is like,
and it's so much easier to review code if it's
all following kind of similar patterns, and you know, you
don't get some like off the cuff, like completely different
thing that you have to be like, okay, how does
this work?
Speaker 4 (42:18):
Yeah, I'm a little bit different, I mean and next
is a small team, so we don't have a lot
of standards. I guess the standard keeps changing as we
as we learned the project, as we learn the technology
that we use. So every other day I would fight
with the head of.
Speaker 2 (42:36):
Lam Yeah stop stop, Well we are, we're getting close
to the end here.
Speaker 3 (42:49):
I think we've all kind of got heart outs today.
Speaker 2 (42:52):
Was there anything that we didn't talk about that you
were hoping to talk about today?
Speaker 6 (42:57):
I don't think so we covered a lot they should definitely.
Speaker 2 (43:02):
Uh, do you have any upcoming talks or appearances that
you're excited about.
Speaker 6 (43:08):
I'm going to speak at talking about pieces they're doing,
like the AI Productivity Summit. It's online. I'll be speaking
at that on May first, and I think that's the
main thing that's on top of my mouth right now.
Speaker 3 (43:23):
Yeah, I'm going to put that in the show notes
as well. We had we had.
Speaker 2 (43:31):
Jump from Pieces on earlier and it was so great
to have him on to have him kind of.
Speaker 3 (43:36):
Talk about it.
Speaker 2 (43:36):
So so yeah, that's awesome.
Speaker 3 (43:41):
Well, it's been really great to have you on.
Speaker 2 (43:43):
If people in the community would like to reach out
to you to learn more about resilient coders, to learn
more about goose to learn more about you?
Speaker 3 (43:56):
What's the best way to reach you?
Speaker 6 (43:58):
Uh, my hand on all social media platforms is black Girl,
But like I'm on everything blue Sky threads every day.
Speaker 3 (44:07):
Nice love it all right? Well, thank you for joining
us today. We always appreciate your time.
Speaker 2 (44:17):
Obviously our podcast wouldn't It would be something without our guests.
Speaker 6 (44:21):
But it would still be amazing.
Speaker 7 (44:24):
I'll ask good questions.
Speaker 2 (44:25):
It'd still be amazing. It would just get like really
deep in the weeds, I think. So thank you for
keeping us on topic. To our listener, if.
Speaker 3 (44:35):
If you like what here, be sure to subscribe to
the podcast.
Speaker 2 (44:39):
And if you haven't yet, the talk proposals are still
open at energie Comp, so go tog comp dot org.
Speaker 3 (44:49):
Get your proposals in for that. Get your tickets.
Speaker 2 (44:51):
That will be October seventeenth through the nineteenth. I believe
I may have thrown out the wrong dates because I
didn't look them up ahead of time, but I know
it's on October seventeenth because that's my kid's birthday. I
was like, oh good, I'll be somewhere near Baltimore at
that time. So yeah to the listener, thank you for
(45:11):
joining us today. Razelle Thank you so much for joining us,
and I look forward to seeing more from you in
the future.
Speaker 7 (45:18):
Thanks for having me absolutely thank you.
Speaker 8 (45:22):
Hey, this is Prestol. I'm one of the NGI Champions writers.
In our daily battle to crush out code, we run
into problems and sometimes those problems aren't easily solved. NGCOMF
broadcasts articles and tutorials from angie champions like myself that
help make other developers' lives just a little bit easier.
To access these articles, visit medium dot com, forward Slash ngcomm.
Speaker 1 (45:44):
Thank you for listening to the Angular Plus show in
Chiecoff podcast, we'd like to thank our sponsors, the NGCOMF
organizers Joe Eames and Aaron Frost, our producer Gene Bourne,
and our podcast editor and engineer Patrick Ky's. You can
find him at spoonful of Media dot com.
Welcome to the Angular plus Show. We're app developers of
all kinds share their insights and experiences. Let's get started.
Speaker 2 (00:21):
Hello and welcome to another episode of the Angular Plus
Show podcast. My name is Laura Newsome.
Speaker 3 (00:27):
I'll be one of your hosts today with me. Today
I have my co host Chao Chow. How's it going?
Speaker 4 (00:34):
Hey, he's doing pretty well. Nice spring weather.
Speaker 2 (00:39):
Yes, I've got my door open, so like, if you
hear birds, I doubt if you will, but if you do,
it's me because the birds are out and it's great.
Speaker 3 (00:48):
My other co host today is que Q. How's it going.
Speaker 5 (00:51):
It is going excellent. Thanks for asking, Laura. How are
you today?
Speaker 3 (00:54):
I am excellent as well. I'm doing well.
Speaker 2 (00:57):
And we have a guest today. Oh no, Risel, Oh no,
I did that right, Riselle Scarlet? Yeah right, okay, so
full the sculpture all afternoon. I've been saying Riseel, Riseell,
and so like, I'm so afraid.
Speaker 3 (01:12):
So it's Riselle.
Speaker 6 (01:14):
Don't know if you call me rise, I'm not gonna
I'm not gonna be like, oh my gosh.
Speaker 3 (01:19):
Yeah, that's how I feel about.
Speaker 2 (01:20):
Like. My name is l A r A and so
it's Lara but people sometimes call me Lara or Laura
and I'm like, I don't care, You're like things I'm
worry about. Soll welcome to the podcast. Uh do you
want to introduce yourself to the listeners?
Speaker 6 (01:38):
Yeah? So, like you said, my name is Roselle Scarlett.
I'm a developer advocate at Block. Specifically, I just became
a tech lead for the open source several team, so
Block is going all in non open source, which is
really exciting. Other things about me, I'm a new mom
and that's it I could think of off the top
(01:58):
of my head.
Speaker 3 (01:59):
Nice. Well, congratulations on the promotion. That's really exciting.
Speaker 2 (02:04):
Do you want to so you have the way that
you got into tech or you went through the Resilient
Counters program?
Speaker 7 (02:12):
Yeah?
Speaker 3 (02:13):
Do you want to tell us a little bit about that?
Speaker 2 (02:16):
Yeah?
Speaker 7 (02:16):
Definitely, so I did.
Speaker 6 (02:19):
So. Okay, when I wanted to get into tech, I
had initially been studying like psychology in college and I realized, Okay,
this is not going to work for me financially, because
I started asking around and people like, yeah, you need
like a master's to like start getting a real like
job in that, And I'm like, oh no, I don't
know if I have master's money right now, so let
(02:40):
me like step back and figure out. And then I
also ended up having to drop out of college. So
I like googled how to make a lot of money
and I love it. Like text things stuff started coming up,
and I was like, Okay, maybe I can like study
computer science. But I still couldn't go to college at
the time, so I I started looking up like free
(03:01):
ways to learn to code. Free code camp came up,
but also Resilient Coders came up. And Resilient Coders is
like this nonprofit coding boot camp that's based in Boston.
At the time, it was only based in Boston. Now
it's based in Philadelphia as well, and basically it teaches
black and brown folks to code, and you actually get
(03:25):
paid to learn. It's not like a huge amount of money,
but they're trying to basically offset the issue of like, Okay, one,
sometimes people have to pay a ton of money to
go to a coding boot camp, or even if it's
like something that's more free, you have to give up
your job to be able to do it or other
like times and hours. So like this was really good
(03:46):
for me. I did about like a fourteen weeks and
then jump straight into software engineers.
Speaker 3 (03:52):
Nice. Nice, that's awesome. Yeah, I know.
Speaker 2 (03:54):
I had a friend who did a coding boot camp
and he didn't have to pay up front, but afterwards
he had to pay it all back like really fast,
and he like he was literally like living off the
free snacks that he was getting at his first job.
Speaker 3 (04:09):
I'm like, Jane, dude, that sucks.
Speaker 2 (04:11):
But yeah, so there's sometimes there's strings attached. So it's
amazing that there's a program out there attached, no strings attached,
and in fact they hand you some strings and they're like,
go ahead and pay some rent.
Speaker 3 (04:24):
That's amazing. That's amazing.
Speaker 2 (04:26):
If any of our listeners are interested, we'll drop a
link to Resilient Coders in in the chat. But what
we have you on here today to talk about are
AI agents, So we have obviously I think you'd have
to be living under a rock in tech to not
have somebody talking about AI in your life. So so okay, so,
(04:53):
how how hard is your job going on AI?
Speaker 7 (04:59):
It's all I do now.
Speaker 6 (05:01):
So my company actually builds an AI agent, so and
it's open source, so my company is going all in
and my jobs really focused on on championing, championing that
for developers and other people just helping to make their
lives easier. And I'm excited to talk to you about
(05:22):
it as well. For about AI agents in general.
Speaker 2 (05:25):
Yeah, okay, so for our listener who listeners who may
not know because there's there's a lot to AI. It's
not simply like our managers think, which is just simply.
Speaker 3 (05:35):
Do the AI. What is an AI agent?
Speaker 6 (05:39):
Yeah, So an AI agent, I would describe it as
a software that does the task for you. So what
we've been familiar with with these past few years with
lms and stuff like that, it's like we can talk
to or interact with an LM like GPT four ZHO
(06:01):
so on at three point five. These are sometimes they
have like an interface on top of them. Maybe that's
like chat GPT or anthropic or even get hipcopilot. So
you talk to it or you should write a message
and it does a whole bunch of math in the
background to figure out what's the best words to respond
to you with to help you answer your problem. But
(06:22):
with AI agents, it's that plus an additional layer where
it can go out heead and execute the task for you.
So there's something called an agentic loop, and there's different kinds,
but oftentimes it may look like, Okay, your agent it
like I think of it, like the LM as the
brains and the agent as the one that acts on things.
Speaker 7 (06:45):
Or even.
Speaker 6 (06:47):
This might be a weird analogy, but like the movie
ratituly or a little rat inside that guy's head or
outside his hat, and he like the rat's the one
making the moves or the rat's the one like making decisions,
and the chef is the one just being the puppet
kind of, so the agent is the chef knows nothing,
really doesn't know how to cook. It's like rat, tell
(07:09):
me how to cook, and the rat is the m hands, yeah,
tell me what to do, and the LM it has
it's the brain. It's the mastermind behind all of it.
And it's like, okay, I see that you have access
to these different capabilities and the user wants to Let's say,
(07:31):
the user wants to like build a web app or
book a flight or something like that. So the LM
looks and it's like, oh, okay, you're able to go
to Google flights and get this, or you're able to
run a shell command and build a web app or
create a file or something like that, so it'll tell
it here's the plan of what we're going to do,
(07:51):
here's the files that we're going to touch, or the
commands we're going to do, and then the agent simply
just acts it out for you, and then the user
just gets to sit back and relax and give more
feedback on like is this what I wanted to happen?
Or do I want more follow up? I hope that
made sense.
Speaker 3 (08:11):
It does. And I love the ratatuy reference because.
Speaker 2 (08:17):
I feel like it's just kind of it is hard
to get a mental model of some of this stuff,
so hearing it spoken about different ways it's helpful to
kind of help people get their heads around what is
going on there.
Speaker 5 (08:30):
Exactly that your first time using that ratitude reference or
is that like in your back pocket?
Speaker 6 (08:37):
Is in my head and I've been trying to put
it together, So my first time saying it out loud,
I gotta work well.
Speaker 2 (08:44):
And at landed with me because I have kids that
are exactly the age like they were the right age
when ratatuy came out, So like we've.
Speaker 3 (08:52):
Watched it a lot, so.
Speaker 7 (08:56):
Perfect.
Speaker 3 (08:58):
I love it.
Speaker 2 (09:00):
So Okay, so you're working on an AI agent. So
you were talking about like it kind of sets restrictions
for what it can do.
Speaker 3 (09:08):
Why do we need those restrictions?
Speaker 6 (09:11):
Okay, so AI agents, Well, when I say restrictions or
what capabilities it can do, I would to say it's
necessarily setting restrictions. It's just seeing what it has the
ability to do, if that makes any sense, like whoever
developed it? But I guess we would need those restrictions
because an AI agent, you don't want it to go
(09:34):
crazy and go off and do the wrong things. So
a lot of times when people build AI agents, they
may build like restrictions to make sure that like, oh,
maybe it's not necessarily completely autonomous, maybe asks you.
Speaker 7 (09:46):
For approval more.
Speaker 6 (09:48):
But what I described as the capabilities I wouldn't call
those restrictions. In the AI agent world, people call this
like tool calling or function calling, and these are just
like the apps or data that it has access to,
not necessarily restrictions, but more I will say, more freedom
if that makes sense.
Speaker 2 (10:07):
Yeah, yeah, okay, So it's more like the agent is
like you're just kind of fine tuning.
Speaker 3 (10:13):
I mean, like MS can be.
Speaker 2 (10:15):
Huge, right, Like they can be like literally the whole Internet, right,
So you know, if you go out to do a
very specific task, then of course you're going to want
something to be like, Okay, maybe you don't need to
read all of.
Speaker 3 (10:31):
This as well. Ocus here.
Speaker 6 (10:33):
Yeah, yeah exactly, I see what you're saying. Yeah, instead
of that's a good point, instead of using up the
LM's time and energy and even what's called the context
window looking at every single thing on the internet, it
can go ahead and find that specific tool. It's like, oh,
like Laura wants to book a flight, I'm not going
(10:54):
to look at Google Drive or whatever other stuff has
I'm going to look at Expedia dot com or Google
Flights or something like that to help her find the
flight within her her limitations or within her criteria.
Speaker 7 (11:09):
Yeah, I'll wait for you.
Speaker 6 (11:10):
I have more thoughts, but I'll wait for you to
ask questions before I just like word dump everything.
Speaker 3 (11:17):
I'm like, just start word dumping, like I'm into that.
Speaker 6 (11:19):
Okay, I can't. This is where this is where something
called model context protocol comes in. So I don't know
if you've seen people like talking about MCP's model context protocol.
So AI agents like they've been trying to be a
thing for a little bit.
Speaker 7 (11:35):
But because.
Speaker 6 (11:38):
AI agents ecosystem at first was a little bit like
closed off. It's like all right, it's cool, but it
can't autumn Like I still feel like I have to
copy and paste like I did with chat GPT because
it doesn't have access to I don't know, like whatever
platform I'm using, maybe it's work Day or Google Drive
or Snowflake or whatever. I don't know what tools people
(11:58):
are using in their day to day. So the AI
agent wasn't that helpful. But now with Model Context Protocol,
which is an open standard that allows you to have
like or allows your AI agent to get context from
different data platforms, you can literally use those as like
bringing in those tool capabilities to your AI agent. So
(12:21):
literally people around the community, every developer is just like
coming out with their own MCP server, whether it's like oh,
here's the MPCP server for Notion and then you plug
that into your AI agent and now you can tell
your AI agent like build me this Notion page or
look through all of my Notion files or something like that.
Or companies are as well, or like hey, like Slack
(12:43):
is like I'm building we have our own MCP server,
and you can be like AI agent connect to my
my Slack MCP server and go schedule this message for
this time or make a summary of all this, this
entire thread that I missed during this meeting. So that's
that's like my brain dump right now.
Speaker 3 (13:00):
I love it.
Speaker 6 (13:00):
I love it.
Speaker 5 (13:01):
Yeah, this is that's awesome because we just so we
had pieces for developers here and they talked about m
cps as well, Like I that kind of just like, ah,
I remember that because you can you can tie in
context based on what I d E. You're using.
Speaker 3 (13:16):
So that's cool.
Speaker 6 (13:18):
I'm glad you brought them up because they built an
m CP server and Goose is an MCP client, so
you can we were like working with them to like,
oh MCP server to Goose, which is the AI agent
my company is working on.
Speaker 4 (13:32):
Yeah, our company also has an uh an MCP.
Speaker 8 (13:36):
Oh really yeah it is.
Speaker 4 (13:38):
So I worked for all an X, which is like
a like a death tool thing like a like so
we build like an n X MCP so that ll
M or whatever like AI tool do you connect to
it can't understand more about your workspace, project instructures and
all of that stuff.
Speaker 6 (13:56):
I love that everyone's going all in on MCP servers.
I feel like this is kind of cool because, like
I think when AI first came out, it was like, Okay,
this is cool. But then I don't know if y'all
felt this way, but like the excitement kind of died
down a little bit because I'm like, all right, everybody
can write to chat, GPT or whatever it is and
get back text.
Speaker 7 (14:15):
That's cool.
Speaker 6 (14:16):
But now it's like, oh, we can build something, and
just seeing what creations people come up with and how
we can all mix it together, it's kind of exciting.
Speaker 9 (14:25):
Yeah.
Speaker 4 (14:26):
I do feel like it's starting to come together.
Speaker 2 (14:29):
Yeah, yeah, I agree, Yeah, I I you know, even
I'm like, so I'm gonna I can't lie.
Speaker 3 (14:34):
I've been like very I am.
Speaker 2 (14:35):
I feel like a grandma when it comes to AI
stuff because I'm like, oh, there is just like so
much to learn, and then like you'll be at work
and they're like, oh go do go tell us, go
write tickets.
Speaker 3 (14:46):
For this stuff, and you're like, but I need to
learn all this other stuff.
Speaker 2 (14:49):
So I even I am feeling like, Okay, it's starting
to make sense how these tools can be useful, yeah.
Speaker 3 (14:56):
Because it really did.
Speaker 2 (14:57):
At first it was like okay, yeah, I can use
this to generate some data for me and like proof
read my pape.
Speaker 3 (15:03):
Did I make any grammatical errors? Like you know, I'm like.
Speaker 2 (15:07):
Those tools are important too, like because I like, yay,
I don't have to think as hard about my grammar.
But yeah, just finding these ways to make these really
just sort of highly tuned tools to help your workflow.
Speaker 6 (15:21):
It's like we're at another level because, like you just said,
like when before it maybe felt, at least for me
maybe a little bit like we're saying this is saving
my time, but it's not, because now I have to
copy and paste and then I have to edit. But
I'm like, now I'm like, okay, you have context.
Speaker 7 (15:39):
Now I don't have to like copy and paste.
Speaker 6 (15:41):
Okay, here's what I'm doing. This is what happened first.
And I'm like, it's so much worse, right.
Speaker 2 (15:46):
I know, Like I was, I remember trying to like
just give the thing enough context to even come close
to getting back and answer that you wanted. And it's like, dude,
I have to know so much about this, Like by
the time I am able to give this thing enough
context to be able to give me anything relevant back,
I might as well have just done it myself.
Speaker 4 (16:09):
I have I have a question just out of curiosity.
So when you say I build AI agent, what does
this like? What does this mean?
Speaker 8 (16:18):
Really?
Speaker 4 (16:18):
What goes into building an AI agent?
Speaker 6 (16:22):
Okay, I didn't build the AI agent, Just to clear
I'm a developer advocate. My team built the AI agent.
But from what I Okay, first off, y'all can look
at the code. If you you might, you might be
way smarter than me. It's an open source project called Goose,
so givehip dot com flash block slash Goose. But what
(16:43):
they've done is they've created two interfaces. We have like
uh desktop UI interface it's built with Electron and React
and typescript and all that, and then we have a
RUST interface. And from the way I understand it, it's
used agentic loop where they have like functions that are
(17:03):
waiting listening for what the user types in or request,
and then the agent goes and acts upon that request,
reaches out to the MCP server that's plugged into it,
and then it comes back with whatever.
Speaker 7 (17:17):
Tools it has.
Speaker 6 (17:18):
It sends each time the entire context of the conversation,
so the AI agent continues to have that context, and
then it comes back with something for you based on
the plan, and it keeps iterating on that, so like
if something broke or like the command that ran didn't work,
the AI agent will go back and be like that
didn't work to the LM, and we'll go back in
(17:39):
that loop. But I don't know specifically because I didn't
sit there and build it. I've contributed a few things
like oh, let's make this dark mold look a little
bit better.
Speaker 9 (17:51):
Broke, But yeah, yeah, I mean, like I mean, I'm
more like, let's say the company or like the group says, oh,
we want to be like an AI agent?
Speaker 4 (18:05):
What uh what what needs to go?
Speaker 3 (18:08):
Yeah?
Speaker 4 (18:09):
What what the process?
Speaker 8 (18:09):
What do we need?
Speaker 9 (18:11):
Uh?
Speaker 4 (18:11):
Like do we need to have like an LM available
so that we can talk to it. What kind of
like protocols that we have we need to to to
to start investigating or researching to be able to be
like an AI agent.
Speaker 6 (18:24):
Yeah, I mean I can say what I've seen in
the middle or in the early beginnings. I kind of
one off on maternity leave in between it. But from
what I seen, we have started off with like a
more rudimentary version of an AI agent that was just
built in Python, and it was a command line version thing,
and we were just using it internally to help automate
(18:44):
our migrations, and then we were like, why don't we ope.
Speaker 7 (18:47):
A source this. This is really helpful.
Speaker 6 (18:49):
And we can give it more capabilities. Then we also
implemented like why not instead of like being stuck into
one lll M people like the for an lllms, why
don't we allow people to bring their own lms. So
the way it works when you use it is like
you can choose I want to use GPT four point
five or whatever it is, bring your API key into that,
(19:13):
or maybe you want to use OLAM or something. And
then the other thing we thought about, because block is
really about like protecting your data and your privacy, we're like,
let's make it local to your machine. Like a lot
of AI agents right now, they're they're in browser.
Speaker 7 (19:32):
Your data is.
Speaker 6 (19:33):
Going out to somewhere you're not able to use it
at your company.
Speaker 7 (19:36):
So we're like, let's do that as well.
Speaker 6 (19:39):
And there was another thing I was going to say,
let's make it local to your machine, and the thought
went away from me.
Speaker 2 (19:50):
That's all right, Like the local thing is a big
deal though, like because I think that's a huge hurdle
for companies that you know they want to use AI
for things. They want their AI tools to be useful,
but you just can't risk data like you just can't.
Speaker 7 (20:08):
Yeah, that makes it more usable for people. Actually, sorry
to interrupt you. I just remembered what I was gonna
say now.
Speaker 6 (20:14):
And then the version two is when we started working
with Athropics, So we actually helped to build a spec
of MCP with and Thropic, so that that was like
the round two of like, Okay, now we're gonna like
sorry that I interrupted you, but no.
Speaker 3 (20:29):
No, I actually was going so that you might remember
because I you.
Speaker 6 (20:35):
Now, we're gonna build this in Rust would rust back
in and we're gonna make this be an MCP client and.
Speaker 7 (20:41):
Bring an MCP servers there.
Speaker 6 (20:43):
So we were actually from my perspective, I think we
were one of the first MCP clients. And then now
MCP is starting to pick up pace and other companies
are like, oh, we're MCP client too, or we're adopting
this as well. So to me that was like kind
of the process of like figuring out, like you said,
like what are people what's the experience of AI agents?
That's not been great. It's not been like the data
(21:05):
is being shared places it's not easy to connect your
AI agent to other tools, and it's sometimes you want
a specific LM because different lms have different like strengths
and specialties.
Speaker 3 (21:18):
Yeah, yeah, and you can see that and some you know,
like some of the platforms or you can con where you're.
Speaker 2 (21:24):
Like, oh, I'm using Cloud for this and I'm using
chat GPT for this, And I mean there's a huge
difference even with just like copilot and.
Speaker 3 (21:31):
Agentic mode, Like.
Speaker 2 (21:34):
I get way better answers from Cloud than I do
from chat GPT. So yeah, it can definitely make a
big difference to fine tune it for what your industry
is doing.
Speaker 6 (21:45):
So exactly exactly or any task like like you said
with claud, I find it better for like summarizing things
for me maybe g chat GPT, I might use it
like one mini and stufficked that more for like reasoning tasks.
Speaker 3 (22:03):
But yeah, nice, nice.
Speaker 2 (22:07):
So one thing that actually one of the reasons I
reached out to you was because you had posted you
had posted something on LinkedIn about using AI as a
democratizing force, and so one of the things I think
(22:29):
we worry about, you know, with AI, it has a
tent like since a lot of these models are trained
on the Internet, and the Internet is an inherently, frankly
a terrible place sometimes depending on what corners you end
up in, what are.
Speaker 3 (22:44):
How can we use AI for good? I guess is
is a question that I was wondering about.
Speaker 6 (22:53):
First off, we have to educate people on like how
to use it well, because I do think we sometimes
get too overly excited and then meet people in the
wrong direction, Like the vibe coding thing. It's cool, but
then when you're a less experience it's cool and it
allows inexperienced engineers to build. But when you don't know
(23:15):
what to look out for, you might be building something
that can easily break, or you just might be opening
yourself to security risk. So that first thing is just
like educating people. But I think AI is a democratizing
force because it lowers the barrier to coding in general.
I don't I don't think that coding needs to be
(23:37):
this like high up thing that like, oh, only like
special people can do it. Like people have ideas and
they want to bring it to life. And I was
talking to Rachel Neighbors the other day and they had
a really good perspective on this. They were saying like
in the first and first in like our first world,
we're thinking of like, oh my gosh, but what problems
(23:59):
will AI bring? But in other countries they're like this
reduces like the gap between like girls and boys in school,
or just in different countries of people getting to learn
to code, they may not have that access to education.
So we need to start thinking of it that way.
And in the first world, just work on fixing those gaps,
(24:23):
those security gaps or whatever other issues may happen. But
it's not like an inherently bad thing.
Speaker 7 (24:28):
I feel like old, but I hope that made.
Speaker 3 (24:30):
Now, that makes it.
Speaker 2 (24:32):
I love I love that that sentiment because I think
it's easy to always you know, I live.
Speaker 3 (24:38):
I live in the United States. I was born and
raised here.
Speaker 2 (24:40):
Like I I my family has always been sort of
just middle class, Like there's a lot of stuff that
it's easy to forget some of the barriers that people
face in the world. And I do think that that
AI does offer the offers. Early career devs are people
(25:01):
that are learning another tool to use. And one thing
that I've been doing at work is trying to figure
out ways what is the process now, Like, Okay, we
just hired a junior dev, We're like, go use agentic
Mode and co pilot and use it to write your code. Yeah, okay,
So we've encouraged them to do that, What is the
(25:23):
next step, right, because obviously all of that code it
might be fine, it might not be fine.
Speaker 3 (25:28):
So as a senior engineer on the team, what do
I need to do to then take it to the
next step?
Speaker 5 (25:37):
You know?
Speaker 2 (25:37):
And so one of the things that we've I suggested
that we're we have to like actually be raising poll requests.
But you know, it's like, okay, encourage people to explain
what their code is doing. Like, okay, agentic Mode has
written it, now explain what it's doing. Yeah, you know, because.
Speaker 3 (25:56):
I you know, I grew up and I grew up.
Speaker 2 (25:58):
I learned how to code and world of stack over
but overflow where I would just go through and uh
copy paste, right, I didn't always understand the code.
Speaker 3 (26:08):
I didn't have anyone to explain it to. No one
could ound like that's crazy, don't do that.
Speaker 2 (26:14):
But you know, so I think that's a it's an
important responsibility of senior members of the team, and it's
actually I think gives then early career depths an opportunity
to level up faster because you know they're already getting
a sort of decent example, and then how can you
then drive that home that like, Okay, that's a good example,
(26:34):
here's why, or okay, that pattern will cause us problems
this way.
Speaker 7 (26:41):
Yeah, I really I really like that.
Speaker 6 (26:42):
Especially that makes I mean, that makes me also think
it can help a junior engineer to also exercise like
pushing back a little bit if they push back on
an AI agent, like being like why did you make
that decision? Or secure or something like that, like just
asking the AI agent not just having it produce things,
but trying to learn from its decisions. And then I
(27:04):
don't I don't know, I've never seen this happen, but
maybe it'll make them more confident when they have to
interact with their their more senior team members and be.
Speaker 7 (27:12):
Like, oh, why do you make this decision?
Speaker 6 (27:13):
Because I remember when I was more junior, if a
senior team member made a decision in their code, I
will be like, looks good to me, because I'm like,
of course they know more, they're smarter.
Speaker 3 (27:23):
I don't know, right, I know, I don't want to
argue with them. I know, yeah, I had I had
the I had it was like.
Speaker 2 (27:29):
Weird but kind of like one of the best junior
dev experiences on our team. Uh this last year I
had three devs that would be like, why are you
doing it that way?
Speaker 3 (27:39):
Like because but it was great because they were.
Speaker 2 (27:42):
Totally willing to ask questions and push back and you know,
and then it also helped me because I'm like, why
am I doing it that way? But I think in general,
any time you're using anything that's AI generated, the first
thing you should think is is that true?
Speaker 6 (27:57):
Yes? Yeah, because it could be hallucinating exactly exactly.
Speaker 2 (28:02):
I know I use AI a lot, you know, like
it's hard to not use it now when you're googling something, right,
I google something and AI is like, oh, this is
how it is, and it's so confident, and you're like.
Speaker 3 (28:11):
But is that is that? Is that true? Is it relevant?
Speaker 7 (28:14):
Still?
Speaker 3 (28:14):
Is it current?
Speaker 6 (28:15):
Like because it's just summarizing the whole bunch of information
that it might have found. It made me lazier, though,
and I need it.
Speaker 7 (28:22):
I need to keep pushing.
Speaker 2 (28:25):
It's real hard because like I'm just like, yep, looks
good to me. Yeah, nice, nice, Okay, So let's talk
a little bit more about Goose.
Speaker 3 (28:37):
So if somebody what how are people using Goose? Like
what is it? How does it get used?
Speaker 2 (28:43):
Yeah?
Speaker 6 (28:43):
I mean right now mostly for like it's it's heavily
used by engineers, but it doesn't have to be. It
can do non technical workflows as well.
Speaker 7 (28:52):
I can't talk.
Speaker 6 (28:53):
There's people that are doing such awesome things, like.
Speaker 7 (28:56):
Some people at my job have I don't know how
like evil people do things, but I was like listening.
Speaker 6 (29:03):
To them and they're like, apparently like one law legal
document got updated and they want to like compare what
the differences were without like reading through the whole thing,
so they would like ask Goose to like come up
with or like just generate like the top things that
got changed within that legal document. And I was like, oh,
that's interesting. I know different ways that I've used it.
(29:24):
Is like since I'm a developer advocate, I'm coding but
also doing other things. So like let's say I have
a live stream that's coming up and I want to
interview the guests and like make the questions good. I've
done like okay, I've like searched for some YouTube videos
that the the guest has been in before, maybe like
live streams or conference talks, and I've asked Goose to
(29:46):
like come up with like some questions that the person
will be passionate about based on this conference that they
were in or whatever, and they'll like generate a transcript,
it'll generate the agenda, and then I'll also be like, hey,
like here's some idea as I have for.
Speaker 2 (30:03):
Like a.
Speaker 6 (30:05):
Title, but I want something more punchy, so it'll help
me with that. Or it'll give me like ideas for
a thumbnail. So I know those sound like chatchypte like,
but without me. It's without me actually having to like
directly give it the videos, I'll be like, look up
like Rachel Neighbors or this person from open Ai or whatever,
and I'll like generate some questions based on that.
Speaker 7 (30:27):
Let me see what are other things I do.
Speaker 6 (30:29):
A lot there A lot of people are using it
for like Google Drive, Like they'll like give it access
to some of their documents and they're like, hey, like
summarize this, or create some meeting notes on this, or
create an Excel chart for me or Google Sheets chart
for me based on the data that I've found. So
(30:49):
it's it can go from like very technical to like
other like daily jobs, or there's even like silly stuff
that people have been doing just for fun, Like they've
used like an MCP server that makes phone calls for
you and they're like, hey, goo's call and say like
a funny joke or something. So those are just for
like fun demos.
Speaker 3 (31:08):
Yeah, I'll harass my senators for me. Yeah, oh my god,
that's nice. I feel like I'm asking all the questions
chow Q, what do you got?
Speaker 5 (31:24):
No, I mean I'm learning because this is all yeah yelling.
Speaker 4 (31:32):
I'm just learning as well, because I mean I'm probably
using AI agent without knowing it.
Speaker 3 (31:38):
Yeah, so we are you?
Speaker 5 (31:41):
Are you using chapt four point zero?
Speaker 4 (31:44):
No, I'm using uh windsurf.
Speaker 5 (31:47):
That's probably that's probably one.
Speaker 4 (31:50):
Uh I'm using I'm using like cloud cloud call model.
Speaker 5 (31:55):
Yeah.
Speaker 3 (31:57):
Yeah.
Speaker 7 (31:58):
That was my favorite because, in.
Speaker 4 (32:00):
My opinion, I think Kloleu is probably one of the
best at implementing code at the moment, in my opinion.
In my opinion so far, I think I've heard.
Speaker 3 (32:12):
Yan say that too before.
Speaker 2 (32:13):
So now Nicholas is one of our other co hosts,
I'm almost positive he's also said.
Speaker 5 (32:18):
That, so I think he has before too.
Speaker 3 (32:22):
Yeah.
Speaker 2 (32:22):
Yeah, yeah, I think it's it's we are just in
this really interesting spot I think with AI.
Speaker 3 (32:27):
Where I mean, so.
Speaker 2 (32:30):
At Cisco they had all these trainings set up, like okay,
let's go, everyone should level up on AI, right, they
put together this training course, and these courses are less
than a year old, and they already feel outdated.
Speaker 6 (32:42):
Oh wow, it's moving fast.
Speaker 2 (32:44):
Yeah, you know, I mean like some things are there,
you know, like, Okay, this is what prompt engineering is.
This is how you can, like people can jail break
a model, and this is some of the ways that
they can do that. And like knowing that stuff is
good to know. But then like it just feels like
so it So we're all Angular developers, obviously, it's the
Angular plus show. Angular releases a new version every six months,
(33:07):
and it already feels hard to keep up with Angular,
but now we're also keeping up with AI, which honestly
just feels like it's going faster. And probably that's because
all of our companies are throwing just billions of dollars
at it.
Speaker 4 (33:22):
So I told, yeah, sorry, go ahead. I totally angler team.
I totally Angler team to just hey, you're Angler to
cel I just drop the micration story, build an MCP
sver build like an angler and an angler MCP server,
so that when you have like a new version, we
(33:44):
can ask g A I to migrate.
Speaker 7 (33:47):
Right, that's genius exactly.
Speaker 4 (33:52):
Here's here's another idea that I kind of like run
into because because of my workflow is that I'm using
two models like consistently, one with a DIP think capability
and one for like implementing code specifically.
Speaker 5 (34:09):
So my workflow is that.
Speaker 4 (34:11):
I switched between the two and then I have to
reprovide context for the two. It'd be awesome if these
two agents I don't know agents or models can.
Speaker 8 (34:21):
Talk to h other.
Speaker 4 (34:22):
So like hey, uh, let's say let's say Goose. Let's
I'm gonna pick like AA agent Goose, Hey Gooz, I asked, uh,
DIP think or DIP sick this question it comes up
with like an implement implementation plan? Can you do it
with using call?
Speaker 2 (34:38):
I was thinking about like because at work we're building,
like we have to really make sure that the data
that an agent can access is like contained, you know,
like we need to make sure that like whatever it
responds with, the person has permission to see.
Speaker 3 (34:54):
It's not like bleeding in the other parts.
Speaker 2 (34:56):
Of the data, but like like a Kubernetes for AI,
like some sort of centralized hub that just knows where
to go to ask these questions. So yeah, somebody can
just go ahead. I think there's something kind of like that,
but I don't know what it is.
Speaker 7 (35:15):
That would be cool.
Speaker 6 (35:17):
Yeah, I think for for child, I feel.
Speaker 7 (35:20):
Like you could do that with Goose. If you had
a session and you had.
Speaker 6 (35:25):
Access to both l ms, you can just switch it
and be like, okay, I'm going to use this LEM
for this, but you don't have to get out of
the session.
Speaker 7 (35:35):
Yeah, you can just switch.
Speaker 6 (35:36):
And then it'll start responding with that.
Speaker 5 (35:39):
But if there was a Kubernetes, you're saying, the context
would still be there for both.
Speaker 6 (35:44):
At that point the context would it will still be
in the sane ecosystem.
Speaker 4 (35:49):
Yeah, okay, cool that that'd be great Because I used
like h l ms with dip think capability to plan
something like reason I say that this sound that does
this reasoning sound enough to start implementing it, and it's
been pretty good and better at the outer other ms.
Speaker 3 (36:12):
Yeah, yeah, nice.
Speaker 2 (36:14):
I need AI where I can feed it a pigma
drawing and then fit another figma drawing and be like different.
Speaker 8 (36:22):
I think you could do that.
Speaker 6 (36:23):
I'm like, I'm like, I'm not trying to Like, I
think you could like use the Figma MCP server and
then just be like hey, here's me.
Speaker 3 (36:31):
I just we're gonna bring you on.
Speaker 2 (36:32):
Hold on, I'm gonna go and get a list of
questions you're coming on every week?
Speaker 5 (36:36):
Yeah, hang on, You're you're saying that you can do
that with what the what's the Pigma thing.
Speaker 7 (36:43):
Server?
Speaker 6 (36:43):
Yeah, we have a I mean it's not doing Laura's
exact case, but we have a video of like using
an m c P server with Goose. I'm gonna just
drop it.
Speaker 7 (36:56):
In the chat.
Speaker 4 (36:57):
Do it?
Speaker 3 (36:57):
Do it?
Speaker 2 (36:58):
Yes, because I'm taking this to work because what always
happens is that they'll be like, go see what you
need to build and then like a couple of days later,
they changed the drawings, but they don't tell me how,
and you know, like the history is not like I
don't know.
Speaker 3 (37:11):
It's hard.
Speaker 5 (37:12):
Yeah.
Speaker 2 (37:13):
One of the most frustrating parts of my job is
playing where's waldough of like where are the changes?
Speaker 3 (37:18):
Like good luck? Was that there before? I don't know.
Speaker 5 (37:24):
Yeah, we get hit with that often, like hey, that's
that should be a link when the link when when
I designed it?
Speaker 2 (37:31):
Shots now and put it on our tickets, so we
can be like, when I wrote this ticket, this is
what it did.
Speaker 3 (37:36):
Okay, I watch this.
Speaker 6 (37:37):
It's not it's not your exact use case. In the video,
my coworker just uses the pigma file to.
Speaker 7 (37:43):
Create a web AAP.
Speaker 6 (37:44):
But I'm literally like, I'm like, I'm sure you could
do that. You can just be like, here's the two
here's the Pigma file. How did it change without like
having to? I think you just give it your Figma
API tokyo.
Speaker 3 (37:55):
I think I'm going to play around with an Oliver
report back.
Speaker 2 (38:00):
But yeah, it's I think that, Like, I think I
can be excited about AI the more it starts solving
these terrible problems I have at work, right.
Speaker 7 (38:13):
When when the more its useful, it gets.
Speaker 2 (38:15):
Right right I have, I'm not really excited for it
to write all of my code.
Speaker 3 (38:20):
For me, reviewing code is not my favorite job. Like
what wanted to write all the code?
Speaker 2 (38:26):
And I just have to review code for the rest
of my life. That sounds a little bit terrible, But
to lower the bar on you know, some of the
stuff that's just takes time that's tedious, Like that's fantastic.
Speaker 4 (38:41):
Yeah, you know, you know what like what guests meto.
What pisses me off is that people use AI uh.
Speaker 2 (38:47):
And then.
Speaker 4 (38:49):
The the result they don't have a human touch to it.
So what what the AI produces? They can they produce
that entirely, so from the without touching it, like without
putting there like.
Speaker 6 (39:04):
Flair to it.
Speaker 5 (39:05):
Yeah, and everyone has their flare. You know you can
tell like all that Laura wrote that code, like I
could tell.
Speaker 2 (39:12):
My unit and see what kind of example, like what
the mock data looks like?
Speaker 5 (39:17):
So all everyone has their tails. But yeah, and then
and then you get an AI AI has its tail too.
You're like someone someone colly pasted that from from chat, DBT.
Speaker 6 (39:27):
Or even when it's not code, they just wrote something.
They're like in today's long, everlasting world.
Speaker 4 (39:34):
Yes, exactly what a my co workers? What did my
co workers start like generating their pr description entirely with
a I.
Speaker 5 (39:46):
Our manager told us to do that too, they because
get lap has its tone. You can just summarize my
MR for me with based on the ticket and the code.
There's no way I can't be caught.
Speaker 3 (39:57):
I don't.
Speaker 5 (39:57):
I don't know heop like it you used. I don't know.
It's something if you feel dirty about it, I don't know.
Speaker 3 (40:02):
I don't know.
Speaker 5 (40:03):
At some point I'll get over it, but.
Speaker 2 (40:04):
Like maybe the at least initially it to be like
I did read through this, I did.
Speaker 7 (40:12):
You guys are good engineers.
Speaker 5 (40:15):
Now I won't want I do use I do use
the AI like I've been using developers for our pieces
for developers a lot, because I don't write LUA normally,
but my kids like to play roadblocks and all the
code is written in LUA, and so I'll just say, hey,
I don't want to do it this way an angular,
I do it this way? Can I do it this
way in Lula? And it's like, well, you can do it.
(40:36):
People don't generally do it this way, but I'll just
it works for me, and it's and it's been it's
been nice to kind of learn, yeah, arcade code.
Speaker 6 (40:45):
That's what it's good for for me too, is like
discovering maybe technologies I'm not used to working with.
Speaker 7 (40:51):
I just want to like do an experiment real quick.
Speaker 6 (40:54):
Nothing for production, yeah, yeah, I do.
Speaker 2 (40:59):
I have learned to like it. Okay, for writing unit tests.
I love writing tests, but I don't really love writing
them that much.
Speaker 3 (41:07):
So it is nice to be like, what looking at
this file, what coverage am I missing?
Speaker 2 (41:13):
Yeah?
Speaker 3 (41:13):
That's solid. Yeah, I'm like, okay, I can get behind that.
Speaker 2 (41:17):
And then we we put a thing in ours where
we've instructed it to read our read mews first. So
I spent all this time documenting how we intend to
write the code in our application, but then we go
tell our junior devs to use Copilot. It's like, but
we need Copilot to know that, Like, that's not how
(41:39):
we do it. This is how we do it, and
this is where you can get that context from.
Speaker 6 (41:43):
So right, Cursor and Goose are probably good at that
because you can give like Cursor rules or Goose hints,
and you can be like, this is our stylistic way
of doing things that Cisco on this particular I don't know.
Speaker 2 (42:01):
Yeah, yeah, no, I mean that's true because it is like,
and it's so much easier to review code if it's
all following kind of similar patterns, and you know, you
don't get some like off the cuff, like completely different
thing that you have to be like, okay, how does
this work?
Speaker 4 (42:18):
Yeah, I'm a little bit different, I mean and next
is a small team, so we don't have a lot
of standards. I guess the standard keeps changing as we
as we learned the project, as we learn the technology
that we use. So every other day I would fight
with the head of.
Speaker 2 (42:36):
Lam Yeah stop stop, Well we are, we're getting close
to the end here.
Speaker 3 (42:49):
I think we've all kind of got heart outs today.
Speaker 2 (42:52):
Was there anything that we didn't talk about that you
were hoping to talk about today?
Speaker 6 (42:57):
I don't think so we covered a lot they should definitely.
Speaker 2 (43:02):
Uh, do you have any upcoming talks or appearances that
you're excited about.
Speaker 6 (43:08):
I'm going to speak at talking about pieces they're doing,
like the AI Productivity Summit. It's online. I'll be speaking
at that on May first, and I think that's the
main thing that's on top of my mouth right now.
Speaker 3 (43:23):
Yeah, I'm going to put that in the show notes
as well. We had we had.
Speaker 2 (43:31):
Jump from Pieces on earlier and it was so great
to have him on to have him kind of.
Speaker 3 (43:36):
Talk about it.
Speaker 2 (43:36):
So so yeah, that's awesome.
Speaker 3 (43:41):
Well, it's been really great to have you on.
Speaker 2 (43:43):
If people in the community would like to reach out
to you to learn more about resilient coders, to learn
more about goose to learn more about you?
Speaker 3 (43:56):
What's the best way to reach you?
Speaker 6 (43:58):
Uh, my hand on all social media platforms is black Girl,
But like I'm on everything blue Sky threads every day.
Speaker 3 (44:07):
Nice love it all right? Well, thank you for joining
us today. We always appreciate your time.
Speaker 2 (44:17):
Obviously our podcast wouldn't It would be something without our guests.
Speaker 6 (44:21):
But it would still be amazing.
Speaker 7 (44:24):
I'll ask good questions.
Speaker 2 (44:25):
It'd still be amazing. It would just get like really
deep in the weeds, I think. So thank you for
keeping us on topic. To our listener, if.
Speaker 3 (44:35):
If you like what here, be sure to subscribe to
the podcast.
Speaker 2 (44:39):
And if you haven't yet, the talk proposals are still
open at energie Comp, so go tog comp dot org.
Speaker 3 (44:49):
Get your proposals in for that. Get your tickets.
Speaker 2 (44:51):
That will be October seventeenth through the nineteenth. I believe
I may have thrown out the wrong dates because I
didn't look them up ahead of time, but I know
it's on October seventeenth because that's my kid's birthday. I
was like, oh good, I'll be somewhere near Baltimore at
that time. So yeah to the listener, thank you for
(45:11):
joining us today. Razelle Thank you so much for joining us,
and I look forward to seeing more from you in
the future.
Speaker 7 (45:18):
Thanks for having me absolutely thank you.
Speaker 8 (45:22):
Hey, this is Prestol. I'm one of the NGI Champions writers.
In our daily battle to crush out code, we run
into problems and sometimes those problems aren't easily solved. NGCOMF
broadcasts articles and tutorials from angie champions like myself that
help make other developers' lives just a little bit easier.
To access these articles, visit medium dot com, forward Slash ngcomm.
Speaker 1 (45:44):
Thank you for listening to the Angular Plus show in
Chiecoff podcast, we'd like to thank our sponsors, the NGCOMF
organizers Joe Eames and Aaron Frost, our producer Gene Bourne,
and our podcast editor and engineer Patrick Ky's. You can
find him at spoonful of Media dot com.
Popular Podcasts
New Heights with Jason & Travis Kelce
Football’s funniest family duo — Jason Kelce of the Philadelphia Eagles and Travis Kelce of the Kansas City Chiefs — team up to provide next-level access to life in the league as it unfolds. The two brothers and Super Bowl champions drop weekly insights about the weekly slate of games and share their INSIDE perspectives on trending NFL news and sports headlines. They also endlessly rag on each other as brothers do, chat the latest in pop culture and welcome some very popular and well-known friends to chat with them. Check out new episodes every Wednesday. Follow New Heights on the Wondery App, YouTube or wherever you get your podcasts. You can listen to new episodes early and ad-free, and get exclusive content on Wondery+. Join Wondery+ in the Wondery App, Apple Podcasts or Spotify. And join our new membership for a unique fan experience by going to the New Heights YouTube channel now!
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy, Jess Hilarious, And Charlamagne Tha God!
Fudd Around And Find Out
UConn basketball star Azzi Fudd brings her championship swag to iHeart Women’s Sports with Fudd Around and Find Out, a weekly podcast that takes fans along for the ride as Azzi spends her final year of college trying to reclaim the National Championship and prepare to be a first round WNBA draft pick. Ever wonder what it’s like to be a world-class athlete in the public spotlight while still managing schoolwork, friendships and family time? It’s time to Fudd Around and Find Out!