Return to site

79. AI-Powered Code Revolution with William Zeng: Unleashing Developers' Potential

· podcast,AI


broken image


Podcast with William Zeng Part 4



Are you a curious fresh graduate eager toembark on a tech journey? Look no further! In this exciting Part 4, William
Zeng shares invaluable advice to spark your tech career. Unleash your coding
potential with Sweep, the AI-powered code assistant that's open-source and
language agnostic. Find out why speed and passion are the secret weapons
startups like Sweep wield against big tech companies. Dive into the world of
tech debt resolution with Sweep and discover how you can get started on this
groundbreaking platform. Whether you're a seasoned developer or a startup
enthusiast, this podcast will revolutionize your coding world.

Join Andrew Liew in this captivating Part 4of the podcast with AI expert William Zeng. They explore valuable advice for
fresh graduates interested in joining the tech industry, emphasizing the power
of curiosity and seeking out new opportunities. Discover the story behind
Sweep, the innovative open-source software revolutionizing coding tasks.
William shares how Sweep's unique approach sets it apart from tech giants and
enables lightning-fast development. As they discuss the advantages of being
programming language agnostic, they highlight how startups can leverage speed
and agility to create a competitive edge. Tune in to learn tips for developers
and how Sweep is transforming the developer's mindset to conquer technical

[00:00:00] Andrew Liew: What, kind of tips you canactually give to the fresh graduates, let's say, who are interested to join
the, tech industry

[00:00:06] William Zeng: now, yeah, for sure. And thengoing back to, I don't think I me, yeah, I mentioned it before and then I,
yeah. I started studying as art architecture student and I changed towards this
path of technology. I think it's really. Important to be curious. I was very
curious when I was starting out.

[00:00:22] William Zeng: I'm still very curious. I thinkthat kind of helped me move in the direction of things that I liked. And then
those things made me also work harder. And it was a nice cycle and also seeking
out new work. That's also part of, I think, being curious. That's my advice.

[00:00:35] Andrew Liew: So being curious use curiosity toyour advantage as you grow and learn something.

[00:00:40] Andrew Liew: Okay, cool. This certainly got methinking about coming back to what you're building. I don't know whether, like,
when you applied to Y Combinator, did they ever ask you that Hey actually,
Microsoft could build this. Why do you want to build this? Did they ever ask

[00:00:54] William Zeng: Yeah, for sure. And I think ouranswer is that We're building this in a really unique [00:01:00]way, like the first part is we're doing this completely open source.

[00:01:02] William Zeng: So letting anyone see and ourcode base and use, share our learnings with the world. And that's the first
interesting part. So not only Microsoft can build this any company can build
this now because they could just copy our code base. But I think our real
advantages are, us spending time to really understand the problem and also
building really fast, like With both of it with our with sweep using it in our
own code base and also just us Naturally being really interested in the

[00:01:30] William Zeng: We think that with our kind ofdevelopment velocity and also focusing on the user we can you're one of the
first to, we're definitely one of the first to market and then just stay around
and build the best product that we can. That's our thoughts on that.

[00:01:44] Andrew Liew: Yeah I, believe that I'm alsoaligned with you on the view that in the modern days, it's really hard to
build, I call it a competitive advantage or competitive mode, especially in
large language model.

[00:01:54] Andrew Liew: A similar context. The, probablythe most formidable is the passion and [00:02:00]the speed of a, team to iterate it fast. Because the more you gain understand of
this complexity of building a, highly intelligent I call it a. a code assistant
or a bug assistant the more you really understand what you don't know and the
more you can actually work on it, whereas if this was a building in a Microsoft
or Google you you probably never work in a, maybe the largest company Roblox, a
company where.

[00:02:26] Andrew Liew: There's 20, 000 headcounts, 30,000 headcounts. And like just to release a feature or just a feature is which
consists of probably like 20 lines of code probably takes three months just to,
okay, we're going to build this. Everybody agrees, reach consensus. That kind
of like 20 lines of code probably to you will be you and Kevin probably two
hours, or even just a half an hour discussion, two hours to get it done, right?

[00:02:48] Andrew Liew: So the amount of time. To, fromthe idea to the write to release, it's so fast you guys can do way much more of
course, now, with computing power [00:03:00]getting cheaper computing speed of the power getting faster now it's just, the
only bottleneck is human beings,

[00:03:06] William Zeng: right?

[00:03:06] William Zeng: Exactly. Yeah I think we justreleased the the thing I mentioned the GitHub actions. We just released the
yesterday. It took us about one day to get all that done between us in terms of
decision making and deciding we're going to do it. We one day we wrote, we
decided we're going to do it. We wrote off code and then we shipped it.

[00:03:23] Andrew Liew: Yeah. It's that kind of speedthat a big company is very hard to match. For me, I work in startups, I work in
big companies. The only good thing about big companies is I get to drink coffee
in the morning and get to listen to all the interesting gossips. Other than
that, of course, there are always nice people to work with.

[00:03:40] Andrew Liew: What else? Okay, for... What areyour tips for developers interested to use such a tools and how should they get
started? What is their mindset

[00:03:49] William Zeng: should be? Yeah. So I have areally a lot of actually some recent learnings from this. I noticed that I
think the best way to approach working with a programming [00:04:00] agent like sweep is to be very specific.

[00:04:02] William Zeng: For example, we provide arecipe. So the three steps of our recipe are. I'll read them out loud. I think
we have the first one is where to look because the AI agent knows, needs to
know where to work in. So for example, I'll say in the UI file and I say what
to do these are two critical pieces what to do with a decent amount of kind of

[00:04:24] William Zeng: Like I'll say. In our UI file,we had to use a OS agnostic temp directory, for example, that was one that we
did I think about five days ago, and it was great because we were, I noticed,
and there's a reason like why, because some of our users are using Microsoft
and they have problems with our app.

[00:04:39] William Zeng: So in the, in our kind of UIfile, we use a temporary directory needs to be OS agnostic. Ah,

[00:04:45] Andrew Liew: yes. Talking about that right nowwhat is your niche in terms of operating systems like Linux, Microsoft? I don't
know, like

[00:04:51] William Zeng: any other stuff. Oh, yeah. We,oh okay. And then for general tips on using Sweep, it's completely in the
browser right now.

[00:04:58] William Zeng: Okay. So we have two things. [00:05:00] The first one is a command line browserapp, which lets you basically talk like how you would to chat GPT, except it
has the ability to understand your code base. And it has the ability to

[00:05:11] William Zeng: To make PRs and kind of workwith you and expose the client. And I can show you that after this, but the
other one is completely based on the browser and GitHub apps.

[00:05:20] William Zeng: So the only thing you need torun this as a GitHub

[00:05:23] Andrew Liew: account. Okay. And it doesn'tmatter whether is it any, so it's also programming language agnostic, right?
Python, JavaScript, C or is it a very programming language specific? Right now

[00:05:35] William Zeng: we're very programming languageagnostic, but we're pretty focused on the use case of building fixing tech debt

[00:05:41] Andrew Liew: so yeah, the tech debt of, sothere's, it's really regardless of what kind, whatever programming language.

[00:05:46] William Zeng: Yes we're, programming languageagnostic. Okay,

[00:05:48] Andrew Liew: which is a very beautiful thing,because every company that I went to, they all use different scripts, right?
Some use Python, some use Java, C and even Frameworks is different, yeah.

[00:05:58] Andrew Liew: Having that agnostic [00:06:00] programming actually enable a wider usecase to try and give feedback and build on it.

[00:06:05] William Zeng: Definitely. Yeah. So we have,that's one thing I really like about starting with dbd4 versus like training
our own model. It definitely would slow us down a lot and how fast we can
iterate on the product and how the product feels.

[00:06:16] William Zeng: Because right now we get to workwith people that, for example, if people that are running Python, they have
PHP, we have JavaScript and TypeScript and different even different testing
frameworks, right? Like you have like in JavaScript, you know They have like
our users have like cypress tests and also like just has just many different
types of frameworks so we're running into all of these like Interesting issues
and learning a lot from them, but that's one really big advantage.

[00:06:38] Andrew Liew: Okay, cool. So thank you so much,William. I think we are coming near to the end of the podcast. Is there any ask
that you want those who are listening to this podcast to do or to act on? Yeah.

[00:06:49] William Zeng: Yeah. So I think the best thingwould be to go to our, GitHub repository, which I think Andrew, you can share
with the audience.

[00:06:58] William Zeng: Oh, yes. And. [00:07:00] Also, we have our sleep pro, which we'rejust rolling out. So giving everyone five GPT for tickets to start out with for
free. And then for the rest of that, it would use GPT 3. 5, which is much
faster, actually. And then we also have our pro tier, which has 60 GPT for
tickets a month.

[00:07:20] William Zeng: And the previous one is also permonth.

[00:07:22] Andrew Liew: Yeah, this is how you, like yourbusiness model all your value proposition is by tickets, right? And my answer
is like, solving tickets by tickets and that rather than so you're going for a
utility pricing kind of model. Is that how it works? Or rather than with some
software service or some application as a service, they go by like unlimited
for X months or something like that.

[00:07:46] Andrew Liew: So it's you guys that buytickets. Is that how it works?

[00:07:49] William Zeng: Yep, that's right. And with yourfirst five, you should be very strict and very detailed, and that will help you
get the best value out of the free tickets. As you go into kind [00:08:00] of the pro subscription and you find valuefrom sleep.

[00:08:03] William Zeng: We can also, there's, we havemore guides as well on how to use it and just be really efficient with it.

[00:08:07] Andrew Liew: Okay, cool. What else I wanted toask Oh, like for those who started using it and we need to seek help, like some,
I don't call it technical support or prompt support. You guys provide that
like, you guys have some kind of intercom or email.

[00:08:21] Andrew Liew: Inbox that they can drop likehey, I started using this sweet thing, but I get stuck I don't know how to use
this prompt properly. Do you guys provide that kind of support or do you guys
have a master class to? Provide that kind of support i'm

[00:08:34] William Zeng: asking definitely I thinkSupport is the most important thing to me.

[00:08:37] William Zeng: I spent almost I think yesterdayI spent half my day providing support and fixing issues and working with
people. That's something that I don't plan to outsource in the next, future. I
think, and of our team of three, I think all of us will be spending a lot of
our time on very hands on support.

[00:08:54] William Zeng: And for kind of our sweet protier, we're provide kind of priority support. So definitely focus on [00:09:00] our paying users first. And make sure thatthey're usually solved. But then when there are no issues, we also help anyone
that helps in. And there's three of us. Very likely that your issue gets fixed.

[00:09:09] William Zeng: So now,

[00:09:10] Andrew Liew: the audience out there, listen,there's this great, wonderful product to try. There's William, Kevin, and who's
that guy? His name is Lucas. Lucas, okay, cool. Yeah, let me switch off the
recording. I think we're pretty much done already, yeah. Yeah we really
appreciate this conversation, and I'm just curious what is the next six months
of Milestones, or what are you, guys aim to, do or to look at,

[00:09:34] William Zeng: I think the biggest things we'refocused on is making sure our... Users are happy. And then we just two ways,
like you just can learn from us. And we're also learning from them. Like I'm
seeing like what patterns work, like how do prompt what's the best way to
prompt sweep, right? That's the best way to use it.

[00:09:49] William Zeng: What repositories work well,what can have actions work well and a whole bunch of customizability. So I
think it's all, it's going to be really product focused and and yeah, it's
mostly just product focus for the [00:10:00]next six months.

[00:10:00] Andrew Liew: So you guys, the first six monthsis like, the typical lingo in the value is, finding product market fit, right?
So finding that that tribe of like maybe a hundred fans that uses your products
and give you great feedback so that they will eventually share it to the next
hundred and so on and so

[00:10:18] William Zeng: forth.

[00:10:18] William Zeng: That's the word. Yeah.

[00:10:19] Andrew Liew: Yeah keep keep me in touch and Iwill also share apart from sharing this podcast if I ever know any friends who wanted
to Solve a problem and want to reduce technical debt. I definitely get them to
try stream what else?

[00:10:33] Andrew Liew: So you guys are actually cateredto startups at the moment, right? So but any users for the now but preferably
startups because big companies will probably want to call you guys for cyber
security legal agreement Donkey stuff,

[00:10:46] William Zeng: right? Yep. Yep.

[00:10:47] Andrew Liew: Okay, cool. Let's keep in touchand i'll update you guys

[00:10:50] William Zeng: Hey, thank you so much Have agreat great, morning.