Episode 107
LIVE DEMO: BUILDING A GO BACKEND WITH JULES THE AI DEVELOPER
summary In this episode, Dalton Anderson explores the capabilities of Google Jules and Google Flow, comparing them with other AI platforms like OpenAI and Anthropic. He provides a live demo…
summary
In this episode, Dalton Anderson explores the capabilities of Google Jules and Google Flow, comparing them with other AI platforms like OpenAI and Anthropic. He provides a live demo of Jules, discusses its architecture, and shares insights on AI tool integration for content creation and coding automation.
Episode content
Explore every layer of this episode.
Each article, guide, analysis, and field note has its own focused page and stays linked to this source conversation.
Articles & stories
Narrative and editorial pieces that carry the conversation forward.
Research & analysis
Evidence-led work that tests and expands the claims in the conversation.
Parallel Coding Agent Operations Research Note
Parallel coding agents improve throughput only when task definition, execution isolation, review, integration, and release capacity scale with them.
Coding Agent Secret and Environment Boundary Research Note
A coding agent's execution environment is a security boundary. Repository contents, setup scripts, dependency installers, network access, environment variables, logs, art
Asynchronous Coding Agent Control Loop Research Note
An asynchronous coding agent accepts a repository task, prepares an isolated environment, forms a plan, changes code, runs tools, and returns an artifact later. The devel
AI-Generated Pull Request Review Research Note
Review an AI-generated pull request from the requirement and threat model inward. Reading the agent's summary or diff first can anchor the reviewer to the implementation'
Field notes
Focused observations and durable ideas worth carrying into other work.
What Is Google Jules? How the Coding Agent Works
Google Jules is an asynchronous coding agent that plans repository work, runs it in a cloud VM, and returns code as a reviewable branch or pull request.
How to Review an AI-Generated Pull Request
Review AI-generated code from the requirement and threat model inward, then test behavior, inspect dependencies and secrets, verify operations, and preserve rollback.
How to Run Parallel Coding Agents Without Losing Control
Scale coding agents through bounded tasks, isolated execution, minimal permissions, explicit contracts, queued integration, independent review, and clear stop conditions.
Asynchronous Coding Agents vs IDE Assistants
Choose an asynchronous coding agent for bounded repository tasks and an IDE assistant for exploratory work that benefits from rapid steering and local context.
Full episode
Read the complete record.
The show notes, transcript, and source trail remain on this canonical episode page.
TranscriptRead the full conversation.
E107 LIVE DEMO_ BUILDING A GO BACKEND WITH JULES THE AI DEVELOPER
Timestamped transcript
1 00:00:00,750 --> 00:00:06,550 Welcome to Mitch Step Podcasts where we discuss entrepreneurship, industry trends, and the occasional book review.
2 00:00:06,730 --> 00:00:13,350 Today it's just gonna be me, no guests, in the lab, talking tools with you.
3 00:00:13,450 --> 00:00:17,350 We're gonna be talking about Google Flow, Google Jules.
4 00:00:17,370 --> 00:00:24,590 Those are the two tools mainly gonna be covering the product Google Jules versus Google Flow.
5 00:00:24,630 --> 00:00:30,690 Google Flow is a content creation platform where you can generate high fidelity videos.
6 00:00:30,700 --> 00:00:45,261 Jules is an autonomous coding platform where you manage agents and can have agents work in sync and or independently of each other on the same code base and work on various topics,
7 00:00:45,662 --> 00:00:50,446 various topics, various tasks, tasks makes more sense.
8 00:00:51,607 --> 00:00:58,273 That's today's episode, but I do realize that it's been a while since I've done a host intro.
9 00:00:58,273 --> 00:01:00,098 So if you're unfamiliar who I am,
10 00:01:00,098 --> 00:01:01,649 My name is Dalton Anderson.
11 00:01:01,649 --> 00:01:05,222 the host of Interest Step Podcast and I'm really happy to be here.
12 00:01:05,222 --> 00:01:07,383 I've worked on all sorts of crazy stuff.
13 00:01:07,383 --> 00:01:12,226 I've been really fortunate with the guest lineup we had recently and there's more to come on that.
14 00:01:12,226 --> 00:01:16,429 But as far as myself, my background is in insurance and data science.
15 00:01:16,950 --> 00:01:21,293 I like to read, run and build stuff in my free time.
16 00:01:21,293 --> 00:01:23,944 One of these things is what we're going to be talking about today.
17 00:01:23,944 --> 00:01:29,198 We have a live demo today and so I built some code or I didn't build it.
18 00:01:29,198 --> 00:01:33,038 I orchestrated it and Jules agents built it.
19 00:01:33,198 --> 00:01:35,178 So a little bit of clarification there.
20 00:01:35,178 --> 00:01:40,118 I didn't build anything, which is quite nice, but that's going to be today's episode.
21 00:01:40,118 --> 00:01:50,342 So we're just going to be talking about Jules and why I'm pursuing Jules versus other platforms and my thought process behind it.
22 00:01:51,991 --> 00:01:54,113 So we'll get into it.
23 00:01:54,995 --> 00:01:56,637 But the first thing is.
24 00:01:57,619 --> 00:01:58,780 I think.
25 00:01:59,201 --> 00:02:03,346 Is most important is to make sure everybody's got the lay of the land.
26 00:02:05,366 --> 00:02:07,888 with the environment.
27 00:02:07,888 --> 00:02:17,376 So when I talk big companies, we've got OpenAI, not Gemini, Google, and then Anthropic.
28 00:02:17,376 --> 00:02:19,797 They all have similar offerings.
29 00:02:20,378 --> 00:02:22,980 It's really preference on what you want to do.
30 00:02:23,301 --> 00:02:33,989 And there's Grok, which is XAI, which is owned by, I guess, SpaceX now, which in all, or Tesla, I'm not sure who owns them now.
31 00:02:33,989 --> 00:02:35,170 It was a weird...
32 00:02:35,170 --> 00:02:39,652 weird thing going on with all that, but it was quite confusing.
33 00:02:40,773 --> 00:02:43,655 So there is a couple of companies called four.
34 00:02:43,655 --> 00:02:55,881 So, uh, Grok is positioned to be the best real time AI there is because it's hooked directly into X.com.
35 00:02:55,901 --> 00:03:04,766 And so if you're asking about trade news or breaking news, either like crypto stocks, world news, all that stuff,
36 00:03:05,160 --> 00:03:10,014 X AI with like rock is undisputed the best.
37 00:03:10,014 --> 00:03:18,391 If you're talking about coding and like human speech, there is Claude, which is owned by Anthropic.
38 00:03:18,391 --> 00:03:20,182 Their product is Claude.
39 00:03:20,923 --> 00:03:23,425 Anthropic, it's a weird thing.
40 00:03:23,425 --> 00:03:30,531 Like how strong the Claude branding is, like Claude's branding is almost stronger than Anthropic.
41 00:03:30,531 --> 00:03:34,994 Like when people say Claude, they're like, okay Claude, but they're not like
42 00:03:35,160 --> 00:03:37,290 because it's just a weird thing.
43 00:03:37,290 --> 00:03:40,936 I feel like the brand of the model is stronger than the actual brand.
44 00:03:42,883 --> 00:03:43,583 Okay.
45 00:03:43,583 --> 00:03:52,198 And then there's OpenAI, which OpenAI I would say is more like a corporate consultant and it's great at coding.
46 00:03:52,198 --> 00:03:53,205 It's great at everything.
47 00:03:53,205 --> 00:03:54,429 It's just all around great.
48 00:03:54,429 --> 00:04:03,414 And it has the biggest access or the largest repository of third party integrations.
49 00:04:03,414 --> 00:04:05,575 So you can integrate pretty much anything.
50 00:04:05,975 --> 00:04:11,592 Claude is right up there with OpenAI, not to the same extent as OpenAI, but it's there.
51 00:04:11,592 --> 00:04:14,898 And they've got they both got all the important stuff for the most.
52 00:04:16,494 --> 00:04:19,797 Google is like none of those things.
53 00:04:19,858 --> 00:04:28,366 Google is really, I would say the only thing that really stands out on Google's product offering is.
54 00:04:30,018 --> 00:04:33,520 the context window.
55 00:04:33,520 --> 00:04:38,042 So you can send a lot of information through these Google models, like a lot.
56 00:04:38,123 --> 00:04:53,031 And that is reflective in these offerings like Jules and Codex, which is owned by OpenAI, and then Cloud Code, which is owned by Anthropic.
57 00:04:53,052 --> 00:04:57,574 So Anthropic's context window is around like 250,000.
58 00:04:58,574 --> 00:05:06,434 Codex is a million and then jewels is two million.
59 00:05:07,074 --> 00:05:08,974 What is a context window?
60 00:05:09,254 --> 00:05:12,254 So a context window is basically your memory.
61 00:05:12,254 --> 00:05:13,794 Like what can you remember?
62 00:05:14,294 --> 00:05:15,354 And what can you handle?
63 00:05:15,354 --> 00:05:27,374 So that includes your own thoughts, documentation, resources, information, instructions, things that you've done prior and also the things you're doing now.
64 00:05:27,374 --> 00:05:28,522 All of that stuff.
65 00:05:28,522 --> 00:05:35,526 So if you give a task to somebody and they can't remember what they did prior, then they'll get confused sometimes.
66 00:05:35,807 --> 00:05:42,332 If they can't remember the changes that they made and then the changes they're working on, that also causes issues.
67 00:05:42,332 --> 00:05:53,159 So with a big context window, theoretically, you should be able to remember more stuff and make less mistakes and handle more complex topics.
68 00:05:54,200 --> 00:05:57,496 If that's true or not, I think it's still up to debate, but
69 00:05:57,496 --> 00:06:04,639 Another thing that I would say Google and XAI are really good at is cost per compute.
70 00:06:04,859 --> 00:06:08,880 So cost per compute is quite low for both of those models.
71 00:06:08,880 --> 00:06:15,963 Like their flash, the Google flash model, and then XAI is like basic model, they are quite cheap.
72 00:06:15,963 --> 00:06:26,401 And when I say quite cheap, like flash, if I recall correctly, like the brand new flash model is, which is 3.1 is
73 00:06:26,575 --> 00:06:30,155 40 cents per million tokens, which is cheap.
74 00:06:30,155 --> 00:06:36,335 I think like the older models that are production models, like the 3.1 is still in beta.
75 00:06:36,335 --> 00:06:37,795 It's like global beta.
76 00:06:38,335 --> 00:06:43,635 The 2.5, I think it might be in like a 20 cents area.
77 00:06:43,635 --> 00:06:44,535 So it's quite cheap.
78 00:06:44,535 --> 00:06:45,895 Like it's super cheap.
79 00:06:46,495 --> 00:06:48,679 So saying all that stuff.
80 00:06:50,551 --> 00:06:52,932 It's not that one is better than the other.
81 00:06:52,932 --> 00:06:54,733 They're just different approaches.
82 00:06:54,733 --> 00:06:57,155 There's pros and cons to all of them.
83 00:06:57,155 --> 00:07:00,126 For myself, I use Google products.
84 00:07:00,126 --> 00:07:01,254 I've got a Google phone.
85 00:07:01,254 --> 00:07:07,981 I'm a Google fanboy type of thing, but I am quite critical of Google because I think that they can do way better than what they're doing.
86 00:07:07,981 --> 00:07:14,624 They're doing a good job, but like how they handle their open, what is it?
87 00:07:14,624 --> 00:07:15,094 I.O.
88 00:07:15,094 --> 00:07:15,998 Google I.O.
89 00:07:15,998 --> 00:07:18,707 I was like, Whoa, hold on.
90 00:07:18,707 --> 00:07:20,157 What is going on here?
91 00:07:20,611 --> 00:07:24,202 So I am very critical of them, but also enjoy using their products.
92 00:07:24,202 --> 00:07:28,915 And I would like a bigger context window just so I can do more stuff.
93 00:07:30,728 --> 00:07:43,177 And the main thing that was attractive about this Jules platform was the ability to kick off many agents all at once and just let them do their thing.
94 00:07:43,258 --> 00:07:51,404 And so with the Google Ultra subscription, you can do up to 60 simultaneous tasks at once.
95 00:07:51,404 --> 00:07:57,908 I don't even know how that looks or feels like, but I'll try to figure that out.
96 00:07:58,586 --> 00:08:07,983 how I review all the code or I get some agent to review all the code and approve the commits to the repository, I'm not completely certain.
97 00:08:07,983 --> 00:08:18,221 But I did work on it today and I built a very simple train or a very simple claims triage.
98 00:08:20,315 --> 00:08:26,219 app that has a UI or you type in your claim and then you submit it.
99 00:08:26,380 --> 00:08:41,992 And then in the back end, AI examines the claim, it scores it, it looks for injuries, it provides an urgency score, and then it gives you a detection if it's fraud or not.
100 00:08:42,654 --> 00:08:49,949 So we'll be testing that out live later on this episode, but I just wanted to give you the lay of the land.
101 00:08:51,223 --> 00:08:54,204 So I think that we just jump right into it.
102 00:08:54,204 --> 00:08:55,805 Or maybe one more thing.
103 00:08:56,965 --> 00:09:06,950 When you're thinking about these platforms, you're no longer a IC or individual contributor.
104 00:09:06,950 --> 00:09:13,993 What is happening is the agents are becoming individual contributors and you were becoming an orchestrator.
105 00:09:13,993 --> 00:09:16,173 And that's the big thing that people are building.
106 00:09:16,173 --> 00:09:20,615 Like Google just came out with a
107 00:09:20,665 --> 00:09:31,090 forked version of VS code, anti-gravity and anti-gravity, the whole thing about anti-gravity is mission control and you can put missions on these agents and the agents
108 00:09:31,090 --> 00:09:34,841 can go on their mission and come back to mission control with the results.
109 00:09:35,762 --> 00:09:47,957 Similar here, similar to Claude code or Codex is like, you are the leader in this many organization, like you're the leader of the project.
110 00:09:48,111 --> 00:09:56,391 and then you've got all these ICs working, these agents, they're doing their thing and you've got to review their stuff.
111 00:09:56,391 --> 00:10:06,091 And I think it gets to a certain point to where you can't review all their stuff and some agent needs to review their stuff and then maybe the stuff that they're not sure about
112 00:10:06,091 --> 00:10:11,811 kicks out to you, but there's no way I can review like 300 commits a day.
113 00:10:12,231 --> 00:10:13,095 But.
114 00:10:14,691 --> 00:10:18,316 Without further ado, I'll share my screen.
115 00:10:20,623 --> 00:10:21,723 of my project.
116 00:10:21,723 --> 00:10:24,403 So share screen, this tab.
117 00:10:25,243 --> 00:10:26,603 OK, so I've got a session.
118 00:10:26,603 --> 00:10:30,823 So the first thing that happens when you connect and I guess I'm sharing my GitHub.
119 00:10:30,823 --> 00:10:33,363 If you guys want to connect or follow me, you can.
120 00:10:33,363 --> 00:10:35,303 But a lot of my stuff is going to be private.
121 00:10:37,815 --> 00:10:49,478 So you connect and the first thing it'll ask you to do is it will ask you to connect your GitHub and you can connect other like Git repositories.
122 00:10:50,258 --> 00:10:54,039 But it's best to use GitHub.
123 00:10:54,039 --> 00:11:04,221 And so if you don't like using GitHub, then this might be, I don't know, like inconvenient for you versus Codex is easier to connect to two different vendors that.
124 00:11:04,922 --> 00:11:06,302 No organized get.
125 00:11:08,547 --> 00:11:13,039 Okay, so after you do that, another thing you'll have to do is you'll have to set up your render account, which was super easy.
126 00:11:13,039 --> 00:11:24,053 They make it like a three click process so you can get set up probably in under five minutes if you don't have any issues with authenticating in GitHub.
127 00:11:24,114 --> 00:11:26,995 If you can do all that just fine, you're all good.
128 00:11:26,995 --> 00:11:36,857 Another thing, I don't have currently Ultra on here and so I can only look at Flash, but I think they also put in the
129 00:11:36,857 --> 00:11:42,029 Pro model as well, so you could use that, but currently it can only use Flash, which I'm okay with.
130 00:11:42,029 --> 00:11:43,390 Flash is pretty good at coding.
131 00:11:43,390 --> 00:11:50,722 If I'm coding, I typically use Flash because it gets it done like 95 % of the time and it's way cheaper.
132 00:11:50,722 --> 00:11:53,103 So that's what I rock with a lot of times.
133 00:11:53,103 --> 00:12:00,856 And if I have something complicated, then I would call on anthropic models, the golden goose.
134 00:12:00,856 --> 00:12:03,657 So I didn't mean to do that.
135 00:12:04,488 --> 00:12:05,314 yeah, so.
136 00:12:05,314 --> 00:12:10,538 I think it takes 24 hours for you to handshake with your workspace or with your upgrade.
137 00:12:10,538 --> 00:12:15,191 So keep that in mind when you do use this app and you're like, I'm so excited to get started.
138 00:12:15,191 --> 00:12:18,884 Like myself, it takes 24 hours for the handshake to happen.
139 00:12:19,184 --> 00:12:23,167 But the sessions are quite generous.
140 00:12:23,528 --> 00:12:27,311 I, it says I only have one session, but I did quite a bit.
141 00:12:27,311 --> 00:12:32,394 So I'm not necessarily sure, completely sure how that all works to be honest with you.
142 00:12:34,690 --> 00:12:36,831 But let's get into the session.
143 00:12:36,831 --> 00:12:48,954 I have a session and another thing that bothers me a lot is, and I haven't figured out how to remove it, is I have this old thing that I did in college, a time series regression,
144 00:12:49,655 --> 00:12:54,516 and it doesn't let me remove it from the code base, which is really annoying.
145 00:12:54,736 --> 00:13:00,928 So I don't know if the only way to remove it is to create a new session.
146 00:13:01,326 --> 00:13:16,322 but I think there should be an easier way to remove stuff from your code base because I like was, I said like, I was scrolling over and I accidentally clicked this roof, I guess.
147 00:13:18,380 --> 00:13:24,931 recommended repository and I clicked on it on accident and it added it to my code base.
148 00:13:24,931 --> 00:13:26,924 I was like, wait, no, I don't want it.
149 00:13:26,924 --> 00:13:30,449 I don't want that on my code base and I can't remove it.
150 00:13:32,366 --> 00:13:38,440 I'm pretty sure that the way that you would remove it is you just create a new session and restart.
151 00:13:39,121 --> 00:13:44,624 But come on, like there's there should be like a way to just exit this like all right, like I didn't mean to add it like this.
152 00:13:44,624 --> 00:13:45,555 Just drop it off.
153 00:13:45,555 --> 00:13:46,865 But regardless.
154 00:13:47,526 --> 00:13:51,368 I thought it was really cool, so if we get in here and we go to our session.
155 00:13:52,076 --> 00:13:53,140 It starts off.
156 00:13:53,140 --> 00:13:55,931 At the top here.
157 00:13:56,352 --> 00:14:00,655 So I gave it an actual plan and I told the structure how I wanted to do it.
158 00:14:00,655 --> 00:14:01,327 This.
159 00:14:01,327 --> 00:14:15,763 Project is based on Go and I'm taking a hexagonal architectural approach, which long story short, it makes it so the system doesn't necessarily recognize inputs and outputs.
160 00:14:15,763 --> 00:14:23,676 And so as technology changes, like it is changing really fast today, you can just switch out the different inputs and different outputs.
161 00:14:23,676 --> 00:14:27,758 The system doesn't necessarily care or know what that input is.
162 00:14:27,758 --> 00:14:31,079 it's nothing is, nothing is
163 00:14:31,405 --> 00:14:32,876 like hard coded in.
164 00:14:32,876 --> 00:14:34,517 And so the system runs.
165 00:14:34,517 --> 00:14:36,569 You can just switch out your tires really.
166 00:14:36,569 --> 00:14:48,497 So like think about like a hexagonal architecture is like a car and you it's snowing out and you put on your all weather and it's summer.
167 00:14:48,497 --> 00:14:49,151 All right.
168 00:14:49,151 --> 00:14:49,828 Not all weather.
169 00:14:49,828 --> 00:14:52,570 think like I'm not I've never driven in the winter time.
170 00:14:52,570 --> 00:14:56,302 I know you have winter tires, whatever it is, your winter tires.
171 00:14:56,302 --> 00:15:00,815 And then you switch them to your your your all weather tires, I guess.
172 00:15:01,039 --> 00:15:03,559 Maybe I switched them up, but you get what I mean.
173 00:15:03,559 --> 00:15:04,879 You can just interchange it.
174 00:15:04,879 --> 00:15:05,679 It's not a big deal.
175 00:15:05,679 --> 00:15:10,719 It's not like, all right, now, now we've got to change out the interior and the engine.
176 00:15:10,719 --> 00:15:11,719 Like that's a lot of work.
177 00:15:11,719 --> 00:15:12,919 You don't want to do that.
178 00:15:12,919 --> 00:15:17,079 So what this does is you can switch swap, swap, swap.
179 00:15:17,179 --> 00:15:19,159 So that that's kind of the deal.
180 00:15:19,299 --> 00:15:21,139 So it kicks off.
181 00:15:21,139 --> 00:15:23,519 It tells me what it's up to.
182 00:15:23,519 --> 00:15:26,119 It just starts its project.
183 00:15:26,119 --> 00:15:27,519 It's a little task.
184 00:15:27,799 --> 00:15:30,055 And it goes through, it sets everything up.
185 00:15:30,055 --> 00:15:34,816 And when it sets things up, you can see what it's up to on the right side.
186 00:15:34,816 --> 00:15:37,157 If it was the recent task.
187 00:15:37,737 --> 00:15:45,599 And you just go through here, it tells it what it's doing and then the plan was approved and it goes through and it does all the updates that it talked about.
188 00:15:45,599 --> 00:15:49,320 And then what happens is it makes it commit.
189 00:15:49,360 --> 00:15:55,082 And so it goes under the alias of Arc Skeleton Triage API.
190 00:15:55,082 --> 00:15:58,883 And then this is some kind of hash.
191 00:15:59,222 --> 00:16:08,734 ID agent hash ID so so the agents don't commit over each other or work on the same code or it doesn't mess up the repository.
192 00:16:08,734 --> 00:16:11,195 So it's not necessarily an API key.
193 00:16:11,195 --> 00:16:14,476 It's really an agent ID is how I would say it.
194 00:16:14,476 --> 00:16:17,916 It's an agent ID that identifies this sessions agent.
195 00:16:17,916 --> 00:16:22,898 So if you had like 60 agents working on your code base, they'd all have different IDs.
196 00:16:22,898 --> 00:16:28,591 And so that what that what that prevents is things getting next mixed up when
197 00:16:28,591 --> 00:16:32,331 they might have the same, like I guess sub alias.
198 00:16:32,331 --> 00:16:38,511 And then if they don't have this, if they have different IDs, then it wouldn't overwrite.
199 00:16:38,511 --> 00:16:41,211 But if it was the same exact thing, then it would have issues.
200 00:16:41,211 --> 00:16:44,791 So it just prevents, just prevents issues.
201 00:16:44,891 --> 00:16:48,131 It's a bit odd, but it is what it is.
202 00:16:48,131 --> 00:16:49,451 mean, simple enough.
203 00:16:49,451 --> 00:16:50,671 I don't really care.
204 00:16:50,671 --> 00:16:53,851 And if you want, can change, you can change the name.
205 00:16:54,291 --> 00:16:54,631 Okay.
206 00:16:54,631 --> 00:16:58,391 So that was one session, one task.
207 00:16:58,585 --> 00:16:59,595 Completed.
208 00:17:00,156 --> 00:17:03,068 And then it's important to think about these as tasks.
209 00:17:03,068 --> 00:17:06,481 It's not a token thing or usage limit.
210 00:17:06,481 --> 00:17:08,122 It's task based.
211 00:17:08,122 --> 00:17:09,123 So.
212 00:17:09,864 --> 00:17:13,786 I give it a task and I have a certain amount of tasks I can use per day.
213 00:17:14,087 --> 00:17:18,650 And without Ultra, I think I'm I think I'm out at like six tasks or something.
214 00:17:18,650 --> 00:17:20,331 Six or so.
215 00:17:20,331 --> 00:17:22,693 The daily session limit says 15.
216 00:17:22,693 --> 00:17:25,495 From what I read, it was six tasks.
217 00:17:25,923 --> 00:17:29,284 So I was trying to limit my tasks and get a lot of stuff done.
218 00:17:29,304 --> 00:17:30,914 But these, aren't big texts.
219 00:17:30,914 --> 00:17:36,026 It's not, it's not too much going on, but the next one I kind of go through and does its thing.
220 00:17:36,246 --> 00:17:42,558 And one thing I do like is once it starts building it out, it does, it does verify the build.
221 00:17:42,558 --> 00:17:46,269 So it builds it and then verifies that it's good to go.
222 00:17:46,269 --> 00:17:55,344 And what's happening in the backend is it creates a virtual machine and makes a copy of the repository, makes a virtual machine.
223 00:17:55,344 --> 00:18:07,164 runs the code, does the updates, tests it, verifies the build is good, packs up the virtual machine, brings it back into here and then commits it.
224 00:18:07,564 --> 00:18:13,564 So there's a lot going on in the background that you don't see, but it is very cool that it does all those things.
225 00:18:13,564 --> 00:18:18,064 It is quite slow though, because sometimes it takes a while to run.
226 00:18:18,064 --> 00:18:24,085 And these time minutes, like the time says 43 minutes, it didn't actually take 43 minutes, I think.
227 00:18:24,085 --> 00:18:26,817 One time I skipped over a commit.
228 00:18:26,817 --> 00:18:33,883 I think my total session was 43 minutes, but I skipped over commit and it caused it to increase the time.
229 00:18:34,404 --> 00:18:35,214 I'm not so sure.
230 00:18:35,214 --> 00:18:37,265 Like sometimes this information's wrong.
231 00:18:37,906 --> 00:18:39,187 Let's see if I could get an example.
232 00:18:39,187 --> 00:18:40,148 Yeah, right here.
233 00:18:40,148 --> 00:18:41,609 So this freaked me out.
234 00:18:41,609 --> 00:18:45,572 So I'm working on a simple app, very simple app.
235 00:18:45,613 --> 00:18:54,049 It's got a couple lines of go in there, a package, and it's got some HTML and some other stuff.
236 00:18:54,285 --> 00:18:55,305 but it's very simple app.
237 00:18:55,305 --> 00:19:03,438 And then I saw a commit come in for review and like the PR, the push request and the pull request.
238 00:19:03,438 --> 00:19:05,936 And then it had 88,000.
239 00:19:05,936 --> 00:19:12,380 was like, Hey, yo, what, what, where did you get 88,000 from?
240 00:19:12,380 --> 00:19:14,488 Like nothing in there could be 88,000.
241 00:19:14,488 --> 00:19:16,301 That's a lot of lines of code.
242 00:19:16,701 --> 00:19:21,783 And it was just a, a UI mistake.
243 00:19:21,783 --> 00:19:23,727 I'm not sure how it happens, but
244 00:19:23,727 --> 00:19:28,607 Sometimes it happened to me twice in this session and it happened earlier.
245 00:19:28,607 --> 00:19:29,487 I reviewed the code.
246 00:19:29,487 --> 00:19:31,267 I was like, there's no way what could have done.
247 00:19:31,267 --> 00:19:36,547 So I downloaded the zip, looked at it, ran it locally and I was like, this doesn't have 88,000 lines of code.
248 00:19:36,667 --> 00:19:37,907 Why does it say that?
249 00:19:37,907 --> 00:19:39,427 So that's wrong.
250 00:19:39,427 --> 00:19:42,447 I don't know why it's doing that, but it is doing that.
251 00:19:43,187 --> 00:19:48,427 So the first two tasks I had to just build, build out the app.
252 00:19:48,427 --> 00:19:51,527 And then the third task I had it put, put together.
253 00:19:51,544 --> 00:20:05,631 a very lightweight build of the UI and then it took a screenshot of the UI or maybe it used some kind of app like Figma and put this together.
254 00:20:05,992 --> 00:20:14,126 But then it put it in there, which is quite nice where it says I've inspected the front end changes visually and then it sent a screenshot in there uh and it's building it.
255 00:20:14,126 --> 00:20:20,783 I couldn't even see the commit yet and then it's like, okay, I've colluded my pre-commit stages, I've tested it.
256 00:20:20,783 --> 00:20:24,784 I've tested the build, it looks good, and let's move forward.
257 00:20:25,424 --> 00:20:28,655 And then it does the commit.
258 00:20:28,655 --> 00:20:39,748 And I think this process would be similar if you were a developer, like you would work on a feature, you would make your additions to the code base, then you would build the app,
259 00:20:39,748 --> 00:20:49,921 make sure there's no errors, or you don't get an error when you try to execute, and then from there you would bring it back in, and then you'd try to push it into the code base.
260 00:20:50,029 --> 00:20:54,244 And so that's basically what it's doing is mimicking what you would do on a day to day, which is kind of slow.
261 00:20:54,244 --> 00:21:01,833 But if you could do that at scale with 10 agents or so at a time, then you could get a lot done.
262 00:21:02,595 --> 00:21:06,118 Cause I was able to get this done in like 40 minutes and.
263 00:21:07,929 --> 00:21:09,530 Definitely wouldn't have taken me 40 minutes.
264 00:21:09,530 --> 00:21:15,062 I don't think it's a multi-day project, but there's some stuff going on for sure.
265 00:21:15,702 --> 00:21:17,763 And we'll look into that in a sec.
266 00:21:18,283 --> 00:21:21,884 And after that, it kind of just goes through its thing.
267 00:21:22,225 --> 00:21:30,328 It does the UI, and then the only thing I said was I can't see the commit because it committed it to like an old commit and the commit was closed.
268 00:21:30,328 --> 00:21:31,769 So I couldn't see it.
269 00:21:31,829 --> 00:21:34,650 There was like a sync issue or something like that.
270 00:21:35,030 --> 00:21:37,392 So I had to force it to commit again.
271 00:21:37,392 --> 00:21:39,152 And so could see it again.
272 00:21:39,792 --> 00:21:41,332 A lot going on there.
273 00:21:41,332 --> 00:21:42,812 The UI is nice.
274 00:21:42,812 --> 00:21:44,772 I like how you can see.
275 00:21:45,052 --> 00:21:49,532 The additions and what it's doing and then.
276 00:21:50,452 --> 00:21:54,752 I liked that it puts it into.
277 00:21:55,492 --> 00:21:57,952 GitHub right right away and I think there is it.
278 00:21:57,952 --> 00:22:02,452 There is a setting to where you can have it auto commit to to the to.
279 00:22:02,692 --> 00:22:06,375 The main, but I have that turned off because I'm I'm checking the code.
280 00:22:07,481 --> 00:22:08,551 This is it.
281 00:22:08,611 --> 00:22:11,554 I mean, it was two hours ago, but it didn't take me two hours.
282 00:22:11,854 --> 00:22:22,352 The only thing that I did want to add and only mistake it did have was I had, when I was testing it locally, I had a timeout error on my stuff.
283 00:22:22,352 --> 00:22:23,923 was like, what do mean?
284 00:22:23,923 --> 00:22:25,264 This is simple enough.
285 00:22:25,264 --> 00:22:35,711 And it had a processing time for the AI to think and receive the query, send that to Google.
286 00:22:35,727 --> 00:22:42,411 Google to send back the results within 10 seconds, which wasn't enough time.
287 00:22:42,872 --> 00:22:44,432 So I had to increase it.
288 00:22:44,633 --> 00:22:49,736 So if I stop sharing again, let's stop sharing here.
289 00:22:49,977 --> 00:22:53,799 And then if I come back and I will share.
290 00:22:55,567 --> 00:22:57,494 I'll share my...
291 00:23:00,015 --> 00:23:04,197 and then anti-gravity.
292 00:23:05,839 --> 00:23:06,669 Is this anti-gravity?
293 00:23:06,669 --> 00:23:07,219 Cool.
294 00:23:07,219 --> 00:23:08,800 Yeah, that's the right one.
295 00:23:09,680 --> 00:23:13,181 Okay, so this is it.
296 00:23:13,181 --> 00:23:16,212 This is the project.
297 00:23:16,212 --> 00:23:21,982 I did not do anything besides increase the timeout, so we changed it to 60 from 10.
298 00:23:21,982 --> 00:23:22,861 That was it.
299 00:23:22,861 --> 00:23:23,804 You know, I didn't even do that.
300 00:23:23,804 --> 00:23:25,544 That was an AI agent.
301 00:23:25,865 --> 00:23:29,722 So this is anti-gravity.
302 00:23:29,722 --> 00:23:31,359 I haven't necessarily used it.
303 00:23:31,359 --> 00:23:35,055 I personally, like said, quite a bit.
304 00:23:35,055 --> 00:23:39,708 I feel more productive in Zed versus VS code, but anti-gravity is not that bad.
305 00:23:39,949 --> 00:23:48,616 I'll get used to the mission control thing because I think I get a lot of credits per month to use it, but also I could just use the API within Zed.
306 00:23:48,616 --> 00:23:50,397 So it's pretty much the same thing.
307 00:23:50,678 --> 00:23:54,600 So I don't necessarily know what I'll do, but I'll test it out some more.
308 00:23:55,201 --> 00:23:56,302 Regardless.
309 00:23:56,302 --> 00:23:57,363 So this is the app.
310 00:23:57,363 --> 00:24:03,888 One thing that I do like about what this did is see this main.go.
311 00:24:04,898 --> 00:24:07,810 Main.go is more or less the instructions.
312 00:24:07,810 --> 00:24:12,912 There's not supposed to be a lot of logic built into here.
313 00:24:13,648 --> 00:24:21,517 like it's subjective, but if your main.go is hundreds of lines, then your app is crap.
314 00:24:21,517 --> 00:24:26,580 So you want your main.go to be like 50 to 75 lines, like max.
315 00:24:27,180 --> 00:24:33,954 And if you have a lot of extra stuff in there, you should probably have that built into the package or
316 00:24:33,954 --> 00:24:36,586 somewhere else, not within the main.go.
317 00:24:36,586 --> 00:24:46,793 Like I would think about it is if you were looking at a cooking recipe and you were looking at like just the ingredients, not how to cook it, not anything like that, just
318 00:24:46,793 --> 00:24:49,294 like what ingredients you need.
319 00:24:49,535 --> 00:25:03,074 That's that's basically main.go main.go is the ingredients and then how to cook it, what to cook it, what it is, what is a fish, what is what is the ocean, what is salt, all that
320 00:25:03,074 --> 00:25:03,588 stuff.
321 00:25:03,588 --> 00:25:05,489 That's done in these other files.
322 00:25:06,090 --> 00:25:10,993 So if you've a whole bunch of nonsense in your main.go, that means your app is jacked up.
323 00:25:10,993 --> 00:25:17,618 And so if there's not anything in here, and in this instance there isn't, then it's a good app.
324 00:25:18,359 --> 00:25:21,481 Subjective, but a very good sniff test.
325 00:25:22,722 --> 00:25:24,583 Okay, so let's run this.
326 00:25:24,583 --> 00:25:25,984 So I'm gonna run it.
327 00:25:27,898 --> 00:25:30,209 So it's gonna start up the local host.
328 00:25:32,354 --> 00:25:36,674 And then I will stop sharing this and I'll flip back over.
329 00:25:36,756 --> 00:25:37,568 Stop sharing.
330 00:25:37,568 --> 00:25:40,364 I'll flip back over and I'll share my local.
331 00:25:42,873 --> 00:25:44,331 Share window.
332 00:25:45,924 --> 00:25:47,212 back to.
333 00:25:48,954 --> 00:25:50,782 So many windows open, that's a problem.
334 00:25:58,193 --> 00:25:59,353 No, all right, there it is.
335 00:25:59,353 --> 00:26:01,373 I was like, have an hour time.
336 00:26:01,513 --> 00:26:02,993 Okay, so you should be able to see this.
337 00:26:02,993 --> 00:26:08,433 Sometimes the font is kind of small on the big screen if you're watching it at your house or your TV.
338 00:26:08,433 --> 00:26:12,013 So sometimes I remember, a lot of times I don't, but today I remembered.
339 00:26:12,393 --> 00:26:16,253 So this is the claims triage AI.
340 00:26:16,253 --> 00:26:19,833 Paste the details of the incident below to triage the claim.
341 00:26:19,833 --> 00:26:26,233 The first thing was I was talking to my mom on the phone and she was mean, dot, dot, dot.
342 00:26:26,233 --> 00:26:27,853 So I'm gonna submit this claim.
343 00:26:36,816 --> 00:26:38,645 Okay, it's processing.
344 00:26:38,645 --> 00:26:42,199 Sometimes it takes longer than what I would like.
345 00:26:46,128 --> 00:26:47,429 It's still going.
346 00:26:47,650 --> 00:26:50,052 Maybe I'll get refresh, hold on.
347 00:26:50,813 --> 00:26:55,328 Refresh, because I had the local host running before and I closed down the server.
348 00:26:55,328 --> 00:26:57,590 So let's see again.
349 00:27:04,492 --> 00:27:07,132 Okay, this might be a live demo issue.
350 00:27:07,132 --> 00:27:08,157 I tested it earlier.
351 00:27:08,157 --> 00:27:09,498 It's crazy.
352 00:27:10,240 --> 00:27:11,031 That's crazy.
353 00:27:11,031 --> 00:27:15,866 I'm gonna go back to my other screen to see.
354 00:27:16,949 --> 00:27:18,550 So it's running.
355 00:27:18,831 --> 00:27:19,882 It's running.
356 00:27:25,327 --> 00:27:26,469 Okay, there we go.
357 00:27:26,469 --> 00:27:29,890 I was like, there's no way I've tested this so many times.
358 00:27:29,890 --> 00:27:30,590 What?
359 00:27:30,590 --> 00:27:31,510 It's going to break.
360 00:27:31,510 --> 00:27:35,141 I I tested all sorts of prompts to make sure the AI was working correctly.
361 00:27:35,141 --> 00:27:40,132 And then I put together like three prompts for us to discuss in the episode.
362 00:27:40,393 --> 00:27:49,555 So in this instance, I think it had a hard time because this isn't a real claim and I didn't build any logic or guidance on how to answer this.
363 00:27:49,555 --> 00:27:52,316 So the AI has
364 00:27:52,611 --> 00:27:55,143 some parameters on how it could answer.
365 00:27:55,143 --> 00:28:03,228 So for category, it can do auto, home, slash like property and medical.
366 00:28:03,228 --> 00:28:04,229 That's it.
367 00:28:04,229 --> 00:28:08,972 So it can't say, it can't say this is not a claim.
368 00:28:08,972 --> 00:28:10,333 It's not allowed.
369 00:28:10,413 --> 00:28:17,397 So I specified what it can answer and it's got to figure out what is the most likely answer of the answers available.
370 00:28:17,398 --> 00:28:20,974 And so if I ask it something where it doesn't have an answer available,
371 00:28:20,974 --> 00:28:27,796 I didn't build anything and when I say I, didn't ask AI to build anything that handles that.
372 00:28:29,540 --> 00:28:31,851 But that's kind of an edge case because it's not really a claim.
373 00:28:31,851 --> 00:28:35,932 People should not submit things that aren't claims, but you never know how people are going to use your apps.
374 00:28:35,932 --> 00:28:39,643 So I thought it would be interesting to see what AI thought.
375 00:28:39,883 --> 00:28:44,124 So it says it's a medical claim and it's from a mental health.
376 00:28:44,124 --> 00:28:45,745 My mom is being mean to me.
377 00:28:45,745 --> 00:28:46,952 This is not true, by the way.
378 00:28:46,952 --> 00:28:47,615 I love my mom.
379 00:28:47,615 --> 00:28:48,365 She's super nice.
380 00:28:48,365 --> 00:28:51,106 She can be hard sometimes, but hard in a good way.
381 00:28:51,606 --> 00:28:55,767 Making sure I'm on track and it's a medical claim.
382 00:28:55,767 --> 00:28:57,368 The urgency is low.
383 00:28:57,368 --> 00:28:58,804 It's flagged.
384 00:28:58,804 --> 00:29:10,087 So flagged is if it has an injury so flag is not flagged is like it's suspicious or anything like that flagged is is if it has an injury related claim and then the
385 00:29:10,087 --> 00:29:24,171 probability of fraud on this is high so I'm gonna put in one where It seems like a small claim, so I'll read it out, too Maybe I could figure out a different voice
386 00:29:26,666 --> 00:29:32,038 I was backing out of my driveway this morning and I lightly scraped my car.
387 00:29:33,199 --> 00:29:35,160 I can't even read this.
388 00:29:35,160 --> 00:29:37,740 I lightly scraped my car bumper.
389 00:29:38,121 --> 00:29:42,202 I lightly scraped my rear bumper on the brick mailbox.
390 00:29:42,423 --> 00:29:45,244 There is a small scratch on the paint.
391 00:29:45,244 --> 00:29:47,325 The car drives perfectly fine.
392 00:29:47,325 --> 00:29:49,386 There's no one has been hurt.
393 00:29:49,386 --> 00:29:53,847 I just need to get an estimate for the cosmetic damage.
394 00:29:54,087 --> 00:29:55,998 OK, that was my accent.
395 00:29:56,849 --> 00:30:02,789 I don't know what the next accent I'll do.
396 00:30:02,789 --> 00:30:04,609 So this one is triage result.
397 00:30:04,609 --> 00:30:06,389 It did it pretty quickly.
398 00:30:06,449 --> 00:30:12,849 Triage result auto urgency is three and it's flagged as no, no one got hurt.
399 00:30:12,849 --> 00:30:17,309 Probability of fraud is low.
400 00:30:17,829 --> 00:30:19,649 All right, let's see.
401 00:30:19,669 --> 00:30:21,256 So I'm gonna do this one.
402 00:30:24,016 --> 00:30:26,216 I'm going to be very suspicious on this one.
403 00:30:26,216 --> 00:30:30,736 I'll say I was walking Pat.
404 00:30:30,736 --> 00:30:32,316 That's the same voice.
405 00:30:32,316 --> 00:30:35,196 I don't know if this is art.
406 00:30:35,256 --> 00:30:37,256 How is suspicious?
407 00:30:37,256 --> 00:30:38,936 How does suspicious.
408 00:30:41,614 --> 00:30:47,948 I was walking past a parked car and the side mirrors spontaneous exploded.
409 00:30:47,948 --> 00:30:50,730 A piece of glass allegedly hit my leg.
410 00:30:50,730 --> 00:30:55,584 There is no witnesses, no police reports and the glass melted.
411 00:30:55,584 --> 00:31:04,879 I have severe backlash and I demand the 50,000 wire transfer for emotional distress before I call my lawyer.
412 00:31:06,501 --> 00:31:08,983 That was my best, that was my best go.
413 00:31:08,983 --> 00:31:09,883 I don't know.
414 00:31:09,883 --> 00:31:10,333 I don't know.
415 00:31:10,333 --> 00:31:11,204 That was pretty bad.
416 00:31:11,204 --> 00:31:14,245 but we're staying in character, we're having fun.
417 00:31:14,665 --> 00:31:17,146 No one's getting hurt.
418 00:31:17,146 --> 00:31:30,311 Okay, so this one is triage result, categories auto, urgencies 10, flagged as yes, because there was bodily harm here, and high probability of fraud, which makes sense.
419 00:31:30,311 --> 00:31:32,031 And so the next one is...
420 00:31:36,144 --> 00:31:37,642 I guess I can try to like.
421 00:31:40,069 --> 00:31:41,269 This is so weird.
422 00:31:41,269 --> 00:31:43,570 I don't know why I committed to doing this bit.
423 00:31:44,251 --> 00:31:45,371 All right.
424 00:31:45,371 --> 00:31:46,792 I could do like meows.
425 00:31:46,792 --> 00:31:47,442 Meow meow.
426 00:31:47,442 --> 00:31:47,872 All right.
427 00:31:47,872 --> 00:31:48,771 Nevermind.
428 00:31:48,853 --> 00:31:50,033 That was, that was a joke.
429 00:31:50,033 --> 00:31:54,075 If you, if you didn't catch on, but I guess I'll just read this in my normal voice.
430 00:31:54,075 --> 00:31:55,055 I'm getting shy now.
431 00:31:55,055 --> 00:31:58,556 So my house is currently engulfed in flames.
432 00:31:58,556 --> 00:32:02,558 The fire department is here, but the roof just collapsed.
433 00:32:02,758 --> 00:32:08,140 We all got out, but my grandfather inhaled a lot of smoke and his
434 00:32:08,228 --> 00:32:10,698 being loaded into the ambulance right now.
435 00:32:10,983 --> 00:32:13,342 We lost everything.
436 00:32:18,842 --> 00:32:19,822 So then it's gonna run.
437 00:32:19,822 --> 00:32:30,301 Okay, so it says category's home, vertices 10, flagged no, probability low, fraud probability.
438 00:32:30,362 --> 00:32:38,668 So I thought that was pretty cool that it built this app, everything, the API connections, the whole nine yards.
439 00:32:39,069 --> 00:32:42,311 I honestly don't know how it got my API connection.
440 00:32:42,872 --> 00:32:44,914 I don't know how it got.
441 00:32:44,914 --> 00:32:47,656 I'm running this locally, but they ran it.
442 00:32:48,072 --> 00:32:54,834 on their own and the code shows that it used it used a Google AI.
443 00:32:55,134 --> 00:32:59,875 P like the Google API, uh AI API.
444 00:33:00,215 --> 00:33:08,748 And I don't know how it got the secret keys or or or what, but apparently it was running it and it's already in the code base.
445 00:33:08,748 --> 00:33:14,839 So when I was running it locally, all I had to do is assign the key, the secret key, and it just worked.
446 00:33:15,040 --> 00:33:16,910 So I don't know what.
447 00:33:16,910 --> 00:33:24,683 magic it was doing on its own, but it was pretty cool that it was able to build this app with limited instruction.
448 00:33:24,683 --> 00:33:25,854 The code was crisp.
449 00:33:25,854 --> 00:33:27,824 It never had any issues.
450 00:33:28,065 --> 00:33:30,326 There wasn't any spaghetti code involved.
451 00:33:30,326 --> 00:33:32,286 Everything was fine.
452 00:33:32,286 --> 00:33:41,240 The only thing that I had issues with was I accidentally added that item to the code base, that different, a different repository that isn't related to my project.
453 00:33:41,240 --> 00:33:44,451 And I don't know how to remove it, which is annoying.
454 00:33:44,551 --> 00:33:46,432 And if there is a way to remove it,
455 00:33:47,118 --> 00:33:51,361 That's fine, but I don't necessarily care that there's a way to remove it because it's.
456 00:33:51,942 --> 00:33:56,526 So hard to remove it as is like if you click on it, you go, you go through the little tabs.
457 00:33:56,526 --> 00:34:01,169 I don't see anywhere where it's very clear to the user how to remove it from the code base.
458 00:34:01,230 --> 00:34:06,354 So in that instance, like whatever answer it is, is invalid.
459 00:34:06,615 --> 00:34:12,779 And then the weird stuff about the commits where the commits is coming in at 88000 sometimes.
460 00:34:12,940 --> 00:34:14,781 And it's done that twice.
461 00:34:16,175 --> 00:34:17,426 Within 40 minutes.
462 00:34:17,426 --> 00:34:27,374 So I'm not sure if that's a bug or what that is, but for the price that I'm paying and I know that's deeply subsidized by Google.
463 00:34:27,374 --> 00:34:37,141 So I can't be like, Oh yeah, I'm paying so much because like Google, it's costing Google a lot of money to, offer this product for what they're allowing customers to pay for it.
464 00:34:38,737 --> 00:34:43,640 But that's a problem in my opinion.
465 00:34:44,241 --> 00:34:45,362 So that was Jules.
466 00:34:45,362 --> 00:34:46,963 Overall excited.
467 00:34:46,963 --> 00:34:56,309 I want to get very advanced with Jules and I want to run it nonstop all day and have things scheduled and running throughout the middle of the day, throughout the night.
468 00:34:56,489 --> 00:35:02,777 Just have things scheduled to kick off and check it in the morning and check it in the evening and then have all these things kicked off.
469 00:35:02,777 --> 00:35:08,387 like in the evening, I review all the code that happened throughout the day and then
470 00:35:08,625 --> 00:35:12,025 I approve everything or have some kind of approval process.
471 00:35:12,025 --> 00:35:17,525 And then after that, we walk through, man, it's getting late.
472 00:35:17,525 --> 00:35:30,865 We walk through the next steps for the next day or I guess throughout the night, run that and then hopefully in the morning I review it and I'm turning out like three coding
473 00:35:30,865 --> 00:35:31,505 sessions.
474 00:35:31,505 --> 00:35:38,129 So I do one in the morning or I approve everything from night, do it for the
475 00:35:38,129 --> 00:35:48,709 work day, after work, in the nighttime, look at what was done, approve it, schedule for nighttime, look at it in the morning, and so on.
476 00:35:48,709 --> 00:35:51,329 So it's just a constant cycle of workers.
477 00:35:51,969 --> 00:35:55,289 And if I can get everybody running all the time, that would be cool.
478 00:35:55,289 --> 00:35:56,309 So I'm gonna stop sharing.
479 00:35:56,309 --> 00:35:58,669 Ooh, I almost stopped the recording.
480 00:36:01,721 --> 00:36:07,469 Okay, so real quickly, I'm just going to cover the flow aspect of
481 00:36:09,605 --> 00:36:12,198 what was built, so I'm gonna share my screen again.
482 00:36:12,198 --> 00:36:14,330 Really cool stuff today.
483 00:36:14,671 --> 00:36:19,887 Very light, nothing serious, just keeping it breezy.
484 00:36:28,369 --> 00:36:30,049 So tabs, there we go.
485 00:36:30,109 --> 00:36:33,069 So I made three videos.
486 00:36:33,069 --> 00:36:35,409 It's supposed to have like a similar character.
487 00:36:35,409 --> 00:36:42,589 I didn't put like a crazy amount of effort into it, but I'm sure that if you wanted to have one character, you could.
488 00:36:43,229 --> 00:36:44,929 All right, so this is the first one.
489 00:36:44,929 --> 00:36:46,089 Let me click into it.
490 00:36:47,664 --> 00:36:48,447 Wait.
491 00:36:51,119 --> 00:36:52,819 And this was made in...
492 00:36:54,961 --> 00:36:56,781 I don't know, a minute.
493 00:36:57,521 --> 00:37:02,381 I it looks pretty cool for how quickly it was made and how much effort I put into it.
494 00:37:02,381 --> 00:37:03,921 This one's pretty cool, too.
495 00:37:03,921 --> 00:37:05,341 I like the vibes.
496 00:37:14,936 --> 00:37:16,376 This one's interesting.
497 00:37:18,033 --> 00:37:21,721 It's like a robotic alien from the alien movies, Alien Predator.
498 00:37:29,777 --> 00:37:31,077 A lot going on.
499 00:37:31,678 --> 00:37:36,341 Well, this is supposed to allow you to do content creation and make videos.
500 00:37:36,342 --> 00:37:44,888 Apparently I get 25,000 credits a month, which I have no idea how I'm supposed to use all that.
501 00:37:44,888 --> 00:37:52,253 It's supposed to be like an hour and a half worth of videos and you can only generate shorter videos, I think.
502 00:37:52,253 --> 00:37:56,696 And each of these videos cost me a hundred credits.
503 00:37:56,696 --> 00:37:58,687 So I have...
504 00:38:01,377 --> 00:38:05,437 24,000 and I guess 700 more.
505 00:38:06,077 --> 00:38:09,157 24,700 more credits to spend.
506 00:38:09,157 --> 00:38:10,917 I hope I can get that done.
507 00:38:10,917 --> 00:38:12,897 But I thought this was really cool.
508 00:38:12,897 --> 00:38:15,277 I would like to dive into this a bit more.
509 00:38:15,456 --> 00:38:20,717 You could probably make some pretty cool marketing stuff and I'm going to dive into the tool.
510 00:38:20,837 --> 00:38:24,197 But I'm really mostly interested in Jules.
511 00:38:24,197 --> 00:38:28,785 But I thought that since I'm doing a podcast episode about Jules.
512 00:38:28,785 --> 00:38:33,005 Might as well flip some stuff into flow, dual purpose.
513 00:38:33,785 --> 00:38:35,805 It's economical for everybody.
514 00:38:35,805 --> 00:38:41,265 While I was waiting for the AI workers, I was playing around with flow.
515 00:38:41,265 --> 00:38:47,165 And then when, when flow was done, I was working with the workers, the AI workers and jewels, and then I would flip back and forth.
516 00:38:47,165 --> 00:38:48,885 So it made sense.
517 00:38:48,885 --> 00:38:50,065 It kept me entertained.
518 00:38:51,185 --> 00:38:57,145 But of course, wherever you are in this world, good evening, good afternoon.
519 00:38:59,745 --> 00:39:01,209 and good morning.
520 00:39:02,016 --> 00:39:05,548 Thank you for listening in and I'll talk to you next week.
521 00:39:07,090 --> 00:39:07,951 Cheers.
522 00:39:07,951 --> 00:39:10,415 And next time you see me, I should be on a new camera.
523 00:39:10,415 --> 00:39:17,164 I've been trying to upgrade my camera situation for some time, but I'll talk to you soon.
524 00:39:17,164 --> 00:39:17,707 See ya.
525 00:39:17,707 --> 00:39:18,564 Bye.
SourcesFollow the source trail.
E107 Sources
[[E107 - Transcript - dalton-take-15 (SRT only 1)]] is the only preserved transcript source. It was imported from subtitles, so names, commands, product terms, punctuation, and technical details may contain speech-recognition errors. The recording, repository state, pull requests, and test outputs should be consulted before quoting or reproducing the live demonstration.
The transcript supports Dalton's first-person experience connecting a repository, planning work, running a Jules task, reviewing a proposed change, testing a small Go claims-triage demonstration, and generating short videos in Flow. It does not establish code quality, security, model performance, production readiness, or comparative superiority.
Current Jules sources
The current Jules getting-started guide explains that Jules connects to GitHub, clones a selected repository into a fresh virtual machine, installs dependencies, generates a plan, modifies files after approval, and uses repository instructions such as AGENTS.md.
The current Jules product page describes repository and branch selection, cloud VM execution, test-suite use, pull-request workflows, CLI access, and API workflows. Product model, limits, plans, integrations, permissions, and availability change quickly and must be checked at drafting time.
The Jules FAQ controls current statements about task execution and security. The API quickstart documents programmatic session creation, plan approval, activity inspection, and messaging. A public guide should link current documentation rather than reproduce commands that may age.
Current Flow source
The Google Flow help center describes Flow as an AI filmmaking tool using Veo, Imagen, and Gemini. The transcript contains a brief experiment with three generated clips and credit use. It is enough for the Episode Story, not for a durable product review or tutorial.
Evidence boundaries
The private repository, application name, source code, claims-triage logic, prompts, API configuration, account details, identifiers, branch names, screenshots, and secrets remain private unless separately approved. The transcript includes uncertainty about how an API key became available. No public page may reproduce or speculate about a secret-handling path without inspecting the actual environment and redacting sensitive information.
A successful build and a clean-looking diff do not establish correctness. The demonstration used a small application and an unspecified test surface. Insurance classification output is a software demonstration, not claims, fraud, coverage, underwriting, or legal guidance.
The episode's price, credit, model, interface, repository-support, concurrency, and scheduling comments are dated observations. Current Google documentation controls present behavior. The transcript's ambition to run agents continuously is not a recommended operating pattern.
Current product and workflow research
The current environment guide, plan-review guide, task and repository guide, code-review guide, usage page, and changelog control the present Jules workflow.
The environment-variable announcement, MCP announcement, and CI Fixer announcement control those dated capabilities.
The current Jules API overview, authentication guide, and sessions reference control programmatic session, plan approval, API key, state, message, activity, and automatic pull-request claims.
GitHub's Copilot IDE guide provides the representative interactive assistant workflow. The comparison is about control loops, not a model or vendor ranking.
Review, security, and integration sources
NIST's Secure Software Development Framework controls the risk-based secure-development framing. GitHub's status checks, protected branches, pull-request management, and dependency review control repository governance claims.
Verification record
The five public drafts, five Page Plans, one new product profile, one updated company profile, and four research notes were completed on July 27, 2026.
No private repository, code, recording, pull request, branch, test log, credential, Jules account, Google account, GitHub application, API, VM, MCP server, Flow project, or insurance system was accessed or changed.
The raw transcript SHA-256 hash is recorded in [[E107 Draft Validation]] and remained unchanged.