Meeting Title: AI Consulting and Project Updates Catch-up Date: 2025-12-12 Meeting participants: Neil Cohen, Uttam Kumaran
WEBVTT
1 00:01:07.110 ⇒ 00:01:07.930 Neil Cohen: Hey.
2 00:01:09.460 ⇒ 00:01:10.469 Neil Cohen: Can you hear me?
3 00:01:10.780 ⇒ 00:01:12.170 Uttam Kumaran: Hey, can you hear me?
4 00:01:12.170 ⇒ 00:01:13.690 Neil Cohen: Yeah, yeah, how’s it going?
5 00:01:13.690 ⇒ 00:01:14.889 Uttam Kumaran: Good, how are you?
6 00:01:14.890 ⇒ 00:01:15.510 Neil Cohen: Good.
7 00:01:16.300 ⇒ 00:01:17.290 Uttam Kumaran: How’s everything?
8 00:01:17.660 ⇒ 00:01:19.489 Neil Cohen: Look at my earphones.
9 00:01:19.490 ⇒ 00:01:20.679 Uttam Kumaran: Can you hear me?
10 00:01:20.680 ⇒ 00:01:25.240 Neil Cohen: Yeah, yeah, I gotta change the, audio too loud.
11 00:01:29.030 ⇒ 00:01:30.300 Neil Cohen: Yeah, can you hear me now?
12 00:01:30.440 ⇒ 00:01:30.940 Uttam Kumaran: Yes.
13 00:01:30.940 ⇒ 00:01:35.900 Neil Cohen: Yeah, okay, no, that’s up. It went through my external for some reason.
14 00:01:36.200 ⇒ 00:01:36.990 Uttam Kumaran: Bye.
15 00:01:37.420 ⇒ 00:01:38.550 Neil Cohen: Good, how you doing?
16 00:01:38.550 ⇒ 00:01:40.249 Uttam Kumaran: Good, how’s the family?
17 00:01:40.250 ⇒ 00:01:41.860 Neil Cohen: Pretty good, pretty good.
18 00:01:44.000 ⇒ 00:01:45.719 Neil Cohen: You froze for a bit.
19 00:01:45.920 ⇒ 00:01:46.790 Uttam Kumaran: Can you hear me?
20 00:01:46.790 ⇒ 00:01:49.120 Neil Cohen: Yeah, yeah, okay, now it’s better.
21 00:01:49.300 ⇒ 00:01:52.719 Uttam Kumaran: I forgot, I didn’t know if you had one or two sons, I forgot.
22 00:01:52.720 ⇒ 00:01:55.510 Neil Cohen: I have two. One’s in college, Virginia.
23 00:01:55.510 ⇒ 00:02:00.289 Uttam Kumaran: Yeah, I was gonna ask if he ended up going to college, and like, where’d he… what’s he studying?
24 00:02:00.290 ⇒ 00:02:03.219 Neil Cohen: Yeah, he’s still in computer science at Virginia Tech.
25 00:02:03.510 ⇒ 00:02:04.760 Uttam Kumaran: Oh, hell yeah!
26 00:02:04.760 ⇒ 00:02:11.679 Neil Cohen: Yeah, he did. It was a struggle to find a school, so it’s so hard these days to get into colleges, so…
27 00:02:12.470 ⇒ 00:02:14.520 Uttam Kumaran: He’s not too far from you, then.
28 00:02:14.520 ⇒ 00:02:17.439 Neil Cohen: Yeah, yeah, it’s just, like, 4 hours, so that’s, that’s good.
29 00:02:17.600 ⇒ 00:02:22.860 Uttam Kumaran: Cool. Is your dog sleeping, or… Yes, he is sleeping.
30 00:02:22.860 ⇒ 00:02:24.629 Neil Cohen: It’s okay, okay, it’s like…
31 00:02:24.630 ⇒ 00:02:27.570 Uttam Kumaran: Oh, no, yeah, no, he’s alive.
32 00:02:27.570 ⇒ 00:02:34.539 Neil Cohen: Me yelling at the monitor is, like, white noise to him now. Oh, okay.
33 00:02:34.540 ⇒ 00:02:37.359 Uttam Kumaran: That’s all I do, so…
34 00:02:37.360 ⇒ 00:02:38.710 Neil Cohen: Where are you based out of? Texas?
35 00:02:38.710 ⇒ 00:02:39.820 Uttam Kumaran: I’m in Austin, yeah.
36 00:02:39.820 ⇒ 00:02:41.240 Neil Cohen: Yeah, okay. How’s that?
37 00:02:41.600 ⇒ 00:02:52.250 Uttam Kumaran: It’s great! Yeah, I moved here, like, 3 years ago, and it’s awesome here. It’s so nice. It’s like, I’m kind of in the suburbs,
38 00:02:52.750 ⇒ 00:02:54.430 Uttam Kumaran: And yeah, I just, like…
39 00:02:55.110 ⇒ 00:03:01.989 Uttam Kumaran: started this company, like, you know, maybe, like, two and a half years ago, and just been doing a lot of data and AI work for clients, so I’ve just been working out of here.
40 00:03:02.140 ⇒ 00:03:04.339 Uttam Kumaran: I have, like, a little office room in my house.
41 00:03:04.740 ⇒ 00:03:09.819 Uttam Kumaran: Just been, like, sitting here and, like, pushing things out, but it’s been great.
42 00:03:10.240 ⇒ 00:03:14.159 Neil Cohen: But when you left Flowcode, you went to a different company, I remember.
43 00:03:14.160 ⇒ 00:03:21.960 Uttam Kumaran: Oh, prequel. Yeah, yeah. So I helped them build the first product. But then, like, I don’t know, I felt like it was just, like, another startup where…
44 00:03:22.070 ⇒ 00:03:37.420 Uttam Kumaran: they’re just trying to get more and more funding, they didn’t really care about the product, like, working super well, and they didn’t invest a lot in, like, getting the right resources, and they just didn’t have a plan, and I was like.
45 00:03:37.790 ⇒ 00:03:41.330 Uttam Kumaran: I did that for… Almost 2 years.
46 00:03:41.470 ⇒ 00:03:47.099 Uttam Kumaran: Like, we, like, basically lead product built the whole, you know, built the initial product team, and then…
47 00:03:47.210 ⇒ 00:03:50.270 Uttam Kumaran: Built the first version of the product, and then just left.
48 00:03:50.580 ⇒ 00:04:06.649 Uttam Kumaran: And I was like, whatever. And then I didn’t want to go back into startups, at least, like, someone else’s startup. And then I also didn’t want to go to, like, a big corporation, and I… I had some familiar… familiarity with, like, consulting and contracting, and I was like.
49 00:04:06.900 ⇒ 00:04:19.210 Uttam Kumaran: Okay, well, it’s, like, not like the stuff we’re doing is not necessary, and it would be fun to do it, you know, for… for different clients, but it’s really, like, the politics and, like, stuff like that that’s really, like.
50 00:04:19.390 ⇒ 00:04:24.159 Uttam Kumaran: crappy, and so I was like, okay, maybe we just start a little bit of an agency doing data work.
51 00:04:24.360 ⇒ 00:04:35.449 Uttam Kumaran: So we just… for most of the company, we’ve just been standing up, like, data infrastructure, so warehouses, ETL, you know, dbt and, like, data models, and then…
52 00:04:35.590 ⇒ 00:04:52.349 Uttam Kumaran: BI and reporting. So, kind of, like, full-stack reporting. And then I was using AI tons to build a company, like, everything, because I just didn’t have any money to… to, like… we didn’t raise any money for the business. And so, after doing that.
53 00:04:52.410 ⇒ 00:05:00.520 Uttam Kumaran: you know, this is, like, right after GPT 3.5 came out. We just got really good at, like, running a business with AI, and then
54 00:05:00.600 ⇒ 00:05:03.320 Uttam Kumaran: Of course, like, a lot of people are interested in that now.
55 00:05:03.320 ⇒ 00:05:03.710 Neil Cohen: Yeah.
56 00:05:03.710 ⇒ 00:05:12.729 Uttam Kumaran: built a lot of workflows, we built, like, an internal platform, everybody uses Cursor, and it just helped us, like, be faster and better for, like.
57 00:05:12.810 ⇒ 00:05:24.439 Uttam Kumaran: clients, and then… so now we actually are building, you know, workflows, agentic workflows, doing, like, sort of AI training for clients as well, you know, which has been really, really cool.
58 00:05:24.730 ⇒ 00:05:26.469 Neil Cohen: Do you use cursor, that’s what you’re… you’re.
59 00:05:26.470 ⇒ 00:05:27.510 Uttam Kumaran: Yeah, we use Purser a ton.
60 00:05:27.510 ⇒ 00:05:35.089 Neil Cohen: I didn’t like her. I use Claude code. I mean, most of my code now is all AI-generated. Oh, really? Claude just does everything for me.
61 00:05:35.380 ⇒ 00:05:38.110 Uttam Kumaran: Yeah, I feel like, I don’t know, I still like,
62 00:05:38.550 ⇒ 00:05:42.980 Uttam Kumaran: I… for a lot of data work, it’s tough, because you’re writing, like, SQL, like, a little piecemeal.
63 00:05:42.980 ⇒ 00:05:43.530 Neil Cohen: Huh.
64 00:05:43.530 ⇒ 00:05:53.970 Uttam Kumaran: But I use it a lot for writing, too, for, like, writing project plans, documents, but yeah, some people… I feel like we’re gonna end up using codecs or cursor, like.
65 00:05:54.400 ⇒ 00:06:03.979 Uttam Kumaran: probably in the next few months here to start. As we start to… tickets get more well-defined, we’ll just ship it to Codex, probably, to take the first pass.
66 00:06:03.980 ⇒ 00:06:05.170 Neil Cohen: Heck of a year, yeah.
67 00:06:05.170 ⇒ 00:06:05.860 Uttam Kumaran: Yeah.
68 00:06:05.860 ⇒ 00:06:10.620 Neil Cohen: Yeah, I mean, I use Cloud Code for everything. Code, the documents, or you can do architecture diagrams, everything, just.
69 00:06:10.620 ⇒ 00:06:11.189 Uttam Kumaran: Yeah, yeah.
70 00:06:11.190 ⇒ 00:06:14.650 Neil Cohen: It’ll do, like, Visio file, or not Visio, a Dryo file.
71 00:06:14.650 ⇒ 00:06:16.169 Uttam Kumaran: The mermaid, yeah, the dry out.
72 00:06:16.170 ⇒ 00:06:17.400 Neil Cohen: Yeah, yeah, pretty much.
73 00:06:17.400 ⇒ 00:06:20.510 Uttam Kumaran: No, it’s out of this world. It’s a completely different world.
74 00:06:20.750 ⇒ 00:06:24.999 Neil Cohen: It is, yeah, I mean… I mean, I still… I still review a lot of the code, or most of the code, but…
75 00:06:25.000 ⇒ 00:06:25.530 Uttam Kumaran: Yeah.
76 00:06:25.530 ⇒ 00:06:29.399 Neil Cohen: Because it’s still… sometimes it goes off the… off the rails and does crazy shit.
77 00:06:29.400 ⇒ 00:06:33.490 Uttam Kumaran: What did you tell your son, like, now that AI’s here about, like, computer science?
78 00:06:33.590 ⇒ 00:06:36.040 Neil Cohen: Yeah, so you better take AI class.
79 00:06:37.920 ⇒ 00:06:48.429 Neil Cohen: And he uses it at school, too, because he was asking me questions. I was like, I’m just gonna ask Cloud Code, you should just use it yourself, even if they… because he doesn’t want to, because they tell him, oh, you should try to learn it on your own, which he does, but when he gets stuck, it’s like…
80 00:06:48.430 ⇒ 00:07:05.799 Neil Cohen: well, then you should use the AI, because he did… I was showing him one thing, because it was… he was doing some mobile thing, and he couldn’t figure it out, and he was trying to use Claude code, and he still couldn’t figure it out. So what do you got to do is you gotta tell Claude to debug it for you, tell Claude to add print lines, and then let it see all the output, and then, you know, of course it figures it out.
81 00:07:05.800 ⇒ 00:07:08.389 Uttam Kumaran: Yeah, you have to give it logs, and it’ll figure everything out.
82 00:07:08.390 ⇒ 00:07:08.780 Neil Cohen: everything.
83 00:07:08.780 ⇒ 00:07:12.189 Uttam Kumaran: But, like, if you give it no context, or, like, you have no testing environment.
84 00:07:12.190 ⇒ 00:07:12.650 Neil Cohen: Ray.
85 00:07:12.650 ⇒ 00:07:14.320 Uttam Kumaran: A way for it to run things.
86 00:07:14.320 ⇒ 00:07:14.830 Neil Cohen: It’ll just come.
87 00:07:14.830 ⇒ 00:07:32.509 Uttam Kumaran: I use it for even all Git-related actions, too. I’m like, do this, commit this, like, then push this, and… I think the bigger problem is it’s less about me, it’s more about, like, how do I get our team, like, if you’re just an average engineer, and you come into my company, like, how do we teach this, like, new way of, like, working?
88 00:07:32.510 ⇒ 00:07:46.320 Neil Cohen: Yeah, it’s like, it’s prompt-first development. I mean, my team, because, well, the front-end guy’s fine, he uses it, but the back-end guys, they’re like senior guys, and a lot of… that’s the thing, there’s some senior guys who… the senior guys who want to get shit done, they adapt AI.
89 00:07:46.320 ⇒ 00:07:46.730 Uttam Kumaran: Yeah.
90 00:07:46.750 ⇒ 00:07:58.599 Neil Cohen: the ones who are, like, into, oh, I love the code, and refactoring, and twiddling the bytes, they’re, like, they’re like skeptic, but, you know, I get them to use it, they point out, look, it did this wrong. I say, well, that’s fine, you tell it what to do, and then it put it in the CloudMD file, and then it’ll do right.
91 00:07:58.600 ⇒ 00:08:11.720 Uttam Kumaran: Yeah, but it’s like, what, if you… you saw, like, any one of us makes, like, doesn’t make mistakes, it’s like… but you know what you could do, like, on average, your thing could be, like, 50 to 100% faster. Exactly. Like, to the mistake, or to the whatever.
92 00:08:12.270 ⇒ 00:08:21.789 Neil Cohen: I mean, I’m, like, 5 times faster. I’ve measured it with all the stuff I’m doing. It’s around 5X, and that doesn’t count, like, I do some side projects as well, and that’s even more.
93 00:08:21.790 ⇒ 00:08:22.670 Uttam Kumaran: Yeah.
94 00:08:22.670 ⇒ 00:08:37.000 Neil Cohen: Because the stuff I’m doing… the stuff I’m doing at work is all Rust code, so I gotta be careful with the cloud stuff up in there. It’s a lot of infrastructure. But I do some, like, side projects that are, like, TypeScript, and TRPC and UI, and that stuff is just blazingly fast within.
95 00:08:37.000 ⇒ 00:08:37.780 Uttam Kumaran: Yeah.
96 00:08:37.789 ⇒ 00:08:40.179 Neil Cohen: It’s just because I’ve seen so much stuff that’s exactly like it, it’s like…
97 00:08:40.179 ⇒ 00:08:54.079 Uttam Kumaran: Well, now we can do browser, like, basically, it’ll build up the browser, see it, use the image models to then bring it back down. Like, that’s what cursor does. So it’ll actually render the browser C, and then can just, like, iterate from there.
98 00:08:54.080 ⇒ 00:08:55.990 Neil Cohen: I can have Claude use Puppeteer so it can launch.
99 00:08:55.990 ⇒ 00:08:56.360 Uttam Kumaran: Yeah.
100 00:08:56.360 ⇒ 00:09:01.480 Neil Cohen: puppeteer, and then use that same thing, MPC or MTP to the,
101 00:09:01.480 ⇒ 00:09:01.860 Uttam Kumaran: Yeah.
102 00:09:01.860 ⇒ 00:09:02.950 Neil Cohen: And figure everything out.
103 00:09:03.190 ⇒ 00:09:08.359 Uttam Kumaran: Yeah. What’s, like, what, what is the job? Like, what are you guys… what are you guys building?
104 00:09:08.360 ⇒ 00:09:12.160 Neil Cohen: It’s just… it’s not going anywhere. I mean, I’m not the founder, so I don’t really care.
105 00:09:12.160 ⇒ 00:09:12.570 Uttam Kumaran: Okay.
106 00:09:12.570 ⇒ 00:09:30.079 Neil Cohen: There’s funding till, like, I think till summer. I think we had, because, initially we had, like, 3, 4 years of runway, so now it’s coming to an end. We’ll see where it goes. But it’s essentially… I mean, it’s an interesting project. It’s essentially a platform to do all your, essentially all your stuff in confidential computing.
107 00:09:30.090 ⇒ 00:09:41.920 Neil Cohen: So, you build stuff in confidential computing, and then no one has access to the data or the database, because it’s all protected by hardware cryptography. So, it’s good for, like, government-type stuff. You know, typical startups aren’t going to need that level of security.
108 00:09:41.920 ⇒ 00:09:42.490 Uttam Kumaran: Yeah.
109 00:09:42.490 ⇒ 00:09:46.020 Neil Cohen: But, like, government, so we’ve been doing stuff with NATO and DHS.
110 00:09:46.020 ⇒ 00:09:46.849 Uttam Kumaran: Oh, dope.
111 00:09:46.850 ⇒ 00:09:50.730 Neil Cohen: But it’s just so slow getting those onboarded, so it’s, you know.
112 00:09:50.730 ⇒ 00:09:51.120 Uttam Kumaran: Yeah.
113 00:09:51.120 ⇒ 00:09:54.630 Neil Cohen: if the money’s gonna run out, or if that’s gonna go anywhere. We had something with,
114 00:09:54.790 ⇒ 00:10:03.329 Neil Cohen: Department of Homeland Security, but then the new administration took over, and that department got gutted by Elon, or whoever, Christine Elm, or someone.
115 00:10:03.480 ⇒ 00:10:17.319 Uttam Kumaran: Yeah, I feel like we’re not… it’s… it’s good in a way, because we just now are… we’re growing the number of clients, but a lot of the work is the same thing. So, our innovation is just doing it faster, and we’re… I’m starting to build, like, kind of, like.
116 00:10:17.550 ⇒ 00:10:29.540 Uttam Kumaran: an operating system for the company that is, like, all AI-driven. Like, from stand-ups, to sprint planning, to grooming to writing, like, contracts, SOWs, everything is AI-assisted in the company.
117 00:10:29.660 ⇒ 00:10:36.299 Uttam Kumaran: Like, even our marketing… like, and also everything in our company runs on one-week sprints. Even marketing and sales.
118 00:10:36.820 ⇒ 00:10:41.300 Uttam Kumaran: And so it’s sort of been fun to, like, engineer the company in a way where it’s, like.
119 00:10:41.760 ⇒ 00:10:45.789 Uttam Kumaran: Like, the average task in a company is now so much faster.
120 00:10:45.920 ⇒ 00:10:50.880 Uttam Kumaran: and everybody’s using AI, and so I feel like that’s been really, really helpful.
121 00:10:50.980 ⇒ 00:10:55.670 Uttam Kumaran: To, like, just try to build, like, a faster company that, like, does things way faster, you know?
122 00:10:55.890 ⇒ 00:10:58.779 Neil Cohen: Have you looked at that BMAD method? Have done anything with that?
123 00:10:59.030 ⇒ 00:11:00.180 Uttam Kumaran: Wait.
124 00:11:00.180 ⇒ 00:11:05.070 Neil Cohen: It’s the thing where they have all these different agents set up for different things, like Mark… Oh, yeah.
125 00:11:05.070 ⇒ 00:11:07.560 Uttam Kumaran: I think a lot of that is, like, kind of like…
126 00:11:08.290 ⇒ 00:11:10.170 Uttam Kumaran: It’s kind of like a dream, I think.
127 00:11:10.170 ⇒ 00:11:23.399 Neil Cohen: It is, yeah, I mean, because I do the side thing I’m doing is with another guy I’m friends with, and we’re just doing this SaaS. Essentially, there’s these founders who… they’re construction guys, and they want to build a site for construction workers and so forth, or contractors.
128 00:11:23.400 ⇒ 00:11:23.780 Uttam Kumaran: Yeah.
129 00:11:23.780 ⇒ 00:11:33.420 Neil Cohen: So it’s just, you know, now you can build the whole… you don’t need a big team, you can build everything with, you know, one or two people, and that’s what we’re doing. But we did look at that, but it just generated so many artifacts, all these different spots.
130 00:11:33.420 ⇒ 00:11:34.100 Uttam Kumaran: No, I mean…
131 00:11:34.100 ⇒ 00:11:34.830 Neil Cohen: So much so, like…
132 00:11:34.830 ⇒ 00:11:43.549 Uttam Kumaran: Like, yeah, I mean, I also think that, it’s like you’re trying to one-shot, like, too big of a problem. Like, instead, we just dissect…
133 00:11:43.680 ⇒ 00:12:00.640 Uttam Kumaran: like, we just dissect a workflow, and my job is, like, I don’t need it to be 100% faster. I’m just happy if it’s 20-40% better, because the impact on our margin would be so much better if that’s it. So, I don’t… I tell my team to really be cautious of anything that’s, like.
134 00:12:01.050 ⇒ 00:12:03.469 Uttam Kumaran: oh, you could just do one thing, and then you get the outlet.
135 00:12:03.470 ⇒ 00:12:04.120 Neil Cohen: You get everything.
136 00:12:04.120 ⇒ 00:12:11.719 Uttam Kumaran: I like that. And so I really try to root people in, like, look at the workflow, identify where the human needs to be in the loop.
137 00:12:11.900 ⇒ 00:12:22.580 Uttam Kumaran: Focus on the context, focus on nailing the prompt, and focus on, like, the integration or, like, the next step. But don’t just be like, do this and go, like, you’re gonna get crappy results.
138 00:12:23.100 ⇒ 00:12:28.220 Uttam Kumaran: We’re also using it for design, to do, like, lo-fi.
139 00:12:28.240 ⇒ 00:12:35.830 Uttam Kumaran: mid-Fi, like, prototypes, and then taking that, putting it into Codex to, like, do the first version. Of course, for code reviews.
140 00:12:35.830 ⇒ 00:12:49.989 Uttam Kumaran: But we’re gonna… we’ll start using it for, like, helping for recruiting, and like… but again, it sort of starts as, like, a prompt, then moving into a workflow, then maybe having an agent that, like, does a couple things that does tool calls and stuff like that, but…
141 00:12:50.090 ⇒ 00:12:53.239 Uttam Kumaran: you can’t, like, jump to that, and a lot of this software, I think they just, like.
142 00:12:53.340 ⇒ 00:12:55.980 Uttam Kumaran: they jumped to it, but it’s all demos. It’s like, none of it really.
143 00:12:55.980 ⇒ 00:13:04.330 Neil Cohen: Yeah, yeah, you get so much stuff. That was the problem when we tried to get so many artifacts you have to review. It’s like, here’s the plan for this, here’s the… and it’s just too… too much to consider.
144 00:13:04.330 ⇒ 00:13:10.369 Uttam Kumaran: Yeah, it’s just, like, also, Codex will be in a different… Codex Claw Code, it gets better every 6 months, so, like…
145 00:13:10.370 ⇒ 00:13:10.990 Neil Cohen: Yeah.
146 00:13:11.120 ⇒ 00:13:16.790 Uttam Kumaran: what, like, I’ll just, like, keep doing the same thing. It’s not like it’s getting any slower, you know?
147 00:13:16.790 ⇒ 00:13:20.550 Neil Cohen: I mean, that’s a big problem with cloud code now, it’s just managing the context window.
148 00:13:20.550 ⇒ 00:13:20.940 Uttam Kumaran: Yeah.
149 00:13:20.940 ⇒ 00:13:25.490 Neil Cohen: Go deal with that, and, you know, go back in time to when it was working, and clear it.
150 00:13:25.490 ⇒ 00:13:37.010 Uttam Kumaran: But that’s why you just have to… but you have to do what we always know, which is, like, write a great PRD, break up the plan, then have it dissect the plan. It’s like… it’s actually not much different to when we just did it as humans.
151 00:13:37.150 ⇒ 00:13:39.809 Uttam Kumaran: I just think we assumed that, like.
152 00:13:39.990 ⇒ 00:13:51.749 Uttam Kumaran: you could get away with not doing that. And, like, there will be… there will be agents that will build out all those pieces, like Claude, like, Cursor has plan mode, where you give it something, it’ll literally plan it out first.
153 00:13:51.750 ⇒ 00:13:56.480 Neil Cohen: Yeah, that’s what Cloud does. You do it in plan mode, then you iterate and you go until you get the plan you want.
154 00:13:56.480 ⇒ 00:14:15.949 Uttam Kumaran: Yeah, so usually I use, like, Opus to plan, and then I’ll use, like, Codex to do a lot of the writing, and then… and then that’s better, but again, still, like, okay, I tell my team, like, focus your time more on writing the best PRD, as specific as possible, before giving it to it, because, like, otherwise it’s gonna be slop, like, the whole way out, you know?
155 00:14:18.260 ⇒ 00:14:26.430 Uttam Kumaran: And then we’re trying to figure out its effectiveness for, like, data analysis, which is, like, that’s what Patrick’s company, you know, is doing right now.
156 00:14:26.430 ⇒ 00:14:30.390 Neil Cohen: I haven’t talked to him. I talked to him after he left Flowcode, or was told to leave Flowcode, or whatever.
157 00:14:30.390 ⇒ 00:14:49.389 Uttam Kumaran: Yeah, they’re doing, like, AI… it’s like an… it’s like an AI data analyst for large e-commerce, so they’re doing a lot of Shopify data, but, like, that’s a lot of what we’re trying to do also, is like, how do we speed up the analytics part? Like, actually getting insights from data.
158 00:14:49.600 ⇒ 00:14:58.670 Uttam Kumaran: but it’s tailored to a company, you know? Not just, like, how many orders did I get, but, like, doing really sophisticated, like, cohort forecasting, like, retention analysis.
159 00:14:59.070 ⇒ 00:15:00.339 Uttam Kumaran: Things like that.
160 00:15:02.090 ⇒ 00:15:08.539 Uttam Kumaran: Because, like, it’s… the analysts who could do that are very, very expensive, and it takes us really time-consuming, and…
161 00:15:08.900 ⇒ 00:15:22.590 Uttam Kumaran: A lot of that is, like, the AI is very, very familiar, but you just have to have a great input data set, you know, and so we’re starting to work on that as well. That’s what a lot of our clients are asking us for, is they’re like, how do I… I want to start enabling chat with data use cases, you know?
162 00:15:23.150 ⇒ 00:15:23.860 Neil Cohen: Yeah.
163 00:15:23.860 ⇒ 00:15:24.450 Uttam Kumaran: Yeah.
164 00:15:27.100 ⇒ 00:15:30.920 Uttam Kumaran: What’s, what’s an interesting side project that you’re working on?
165 00:15:32.120 ⇒ 00:15:37.530 Neil Cohen: So, I mean, so, I mean, the one I’m doing, it’s a paid project, is essentially a friend of mine, he’s,
166 00:15:37.970 ⇒ 00:15:49.999 Neil Cohen: building this site for these, these two guys. They’re, like, I think they’re actually in Texas, but they own a bunch of contracting companies, so they want to build something similar to Jobber, but with more AI and different…
167 00:15:50.000 ⇒ 00:16:08.400 Neil Cohen: more workflow-type stuff to it, to do that, so that small contracting companies can use it, and then larger, it’s also targeting the larger companies that, I guess, they buy these small contracting companies, and then they run them as businesses, reporting to the top company, but we’re… I mean, we’re doing it all with cloud code, it’s all.
168 00:16:08.820 ⇒ 00:16:13.930 Neil Cohen: vibe, vibe coded, and I think we built the initial thing in, like, 4 weeks, so…
169 00:16:14.190 ⇒ 00:16:15.160 Uttam Kumaran: Wow.
170 00:16:15.760 ⇒ 00:16:21.590 Uttam Kumaran: Yeah, there’s a bunch of… I have a bunch of… but see, all my ideas I end up just doing for the company, like…
171 00:16:21.830 ⇒ 00:16:27.460 Neil Cohen: I’m like, oh, I think we should speed up the way we do stand-ups, and like, I’m like, okay, we should just build something.
172 00:16:27.460 ⇒ 00:16:30.349 Uttam Kumaran: A little platform where people can start to build agents.
173 00:16:30.350 ⇒ 00:16:35.330 Neil Cohen: But, like, I think we’re gonna stop paying for some software, too, because I could just build it our own.
174 00:16:35.330 ⇒ 00:16:38.340 Uttam Kumaran: That’s exactly what we needed to do, you know?
175 00:16:38.340 ⇒ 00:16:41.429 Neil Cohen: Yeah, you don’t need to buy all these fast projects, like…
176 00:16:41.430 ⇒ 00:16:42.230 Uttam Kumaran: Yeah.
177 00:16:44.810 ⇒ 00:16:54.410 Neil Cohen: Yeah, I mean, like, everything, even setting up on AWS, the cloud will do it all for you, you don’t have to… Yeah, yeah, yeah. You still have to… you still have to go over, you still have to review everything and stuff like that, you don’t want to…
178 00:16:54.410 ⇒ 00:16:56.190 Uttam Kumaran: And you have to… possible. Like, if.
179 00:16:56.190 ⇒ 00:16:56.920 Neil Cohen: You don’t go with…
180 00:16:56.920 ⇒ 00:17:00.680 Uttam Kumaran: possible, how are you gonna ask it to do it? So that’s the thing, is like.
181 00:17:01.010 ⇒ 00:17:10.279 Uttam Kumaran: it’s still helpful that I knew how to do a lot of these things, and then I know how to… now I can just have it do it faster, but if you never knew what, like… I guess…
182 00:17:10.770 ⇒ 00:17:14.590 Uttam Kumaran: Eventually, AI will read through all the docs and, like, find the best stack.
183 00:17:14.599 ⇒ 00:17:26.879 Neil Cohen: I don’t know, maybe, that’s the thing. I mean, that would be the problem for juniors right now, because they won’t know, because, I mean, you get the clog, comes up with a plan, they tell, no, that’s… we don’t want to do that, we want to cache this, or we want to… we don’t want, like, a linear lookup, and of course, it says, you’re absolutely right, and then it changes.
184 00:17:26.880 ⇒ 00:17:27.960 Uttam Kumaran: I know, I know.
185 00:17:27.960 ⇒ 00:17:38.399 Neil Cohen: But, if you didn’t know that, if you just let… I think that’s where I see a lot of, like, the people complaining on the different sites, that they got slopped, is because they didn’t know any better, they just let… let the tools do whatever.
186 00:17:38.400 ⇒ 00:17:51.300 Uttam Kumaran: like, you need to still do, like, a TDD process, right? There’s, like, no avoiding… I just think there’s no avoiding design docs, because that is actually where the human creativity needs to be, to think about, like.
187 00:17:51.330 ⇒ 00:18:00.800 Uttam Kumaran: the actual product development process, think about the design and the architecture. Executing it is, like, out, you know? I feel like that’s really, really out, but…
188 00:18:00.960 ⇒ 00:18:08.109 Uttam Kumaran: Understanding the trade-offs, or understanding the constraints that your company or your customers are dealing… how is the AI gonna know that, you know?
189 00:18:08.380 ⇒ 00:18:16.640 Neil Cohen: Yeah, I don’t know. I mean, you hear some people talk about how they just keep feeding it from different AIs to different agents and so forth, like from Grok to… to Anthropist.
190 00:18:16.640 ⇒ 00:18:25.399 Uttam Kumaran: Yeah, but these are all… but I also just think there’s, like, a level of optimization where people are avoiding doing the hard thing, which is, I think, sitting down and being like, what does I actually…
191 00:18:25.400 ⇒ 00:18:26.940 Neil Cohen: Well, yeah, yeah, yeah.
192 00:18:26.940 ⇒ 00:18:30.029 Uttam Kumaran: You know, but that’s where, like, I tell our people, like.
193 00:18:30.340 ⇒ 00:18:35.469 Uttam Kumaran: just think about, like, what the proof of concept, the MVP, the V1 is, do some features.
194 00:18:35.900 ⇒ 00:18:39.879 Uttam Kumaran: Work with design to do some sketching, and then it’ll come out, like, way better.
195 00:18:43.650 ⇒ 00:18:59.290 Uttam Kumaran: Yeah, but that’s a lot of, like, that’s a lot of what we’re doing. We’re starting to do more, like, AI application development for clients, too, now. But we’re using a tool called, like, LangFuse for prompt management and, like, observability that’s really, really good. I don’t know if you’ve heard of that before.
196 00:18:59.290 ⇒ 00:19:01.020 Neil Cohen: No, I haven’t used that.
197 00:19:01.350 ⇒ 00:19:09.760 Uttam Kumaran: And then we’re using, then we’re using, like, Mastra, which is, like, a JavaScript framework for…
198 00:19:09.990 ⇒ 00:19:12.749 Uttam Kumaran: you know, building agents, it’s sort of like the…
199 00:19:13.570 ⇒ 00:19:15.730 Uttam Kumaran: You know, really well favored right now.
200 00:19:22.400 ⇒ 00:19:25.859 Uttam Kumaran: Because previously, everything was, like, Python to build, like, build agents.
201 00:19:25.860 ⇒ 00:19:26.770 Neil Cohen: Yeah, yeah.
202 00:19:31.740 ⇒ 00:19:32.410 Uttam Kumaran: Yeah.
203 00:19:33.450 ⇒ 00:19:35.020 Uttam Kumaran: That’s been most of it.
204 00:19:35.790 ⇒ 00:19:40.630 Uttam Kumaran: But it’s been fun, yeah, so we’re just slowly growing. We have, like, 16 people now.
205 00:19:40.630 ⇒ 00:19:41.319 Neil Cohen: Oh, that’s pretty good.
206 00:19:41.320 ⇒ 00:19:42.070 Uttam Kumaran: Yeah.
207 00:19:42.200 ⇒ 00:19:46.189 Uttam Kumaran: And we… I think we just… we have about, like, 13 clients.
208 00:19:46.300 ⇒ 00:19:48.079 Uttam Kumaran: And it’s just been, like, crazy.
209 00:19:48.080 ⇒ 00:19:49.819 Neil Cohen: Do you have an office, or you work at home?
210 00:19:49.820 ⇒ 00:19:56.849 Uttam Kumaran: I work at… everybody’s remote. Everybody’s remote. I would like to hire here, but it’s tough, like.
211 00:19:57.020 ⇒ 00:20:06.200 Uttam Kumaran: it’s just hard to hire in CS, like, I feel like… for us, like, I can’t bring on people that are gonna mess up client work, so you have to find people that kind of
212 00:20:06.320 ⇒ 00:20:11.449 Uttam Kumaran: Know enough, but… but most of our team was from product background, like, not from consulting background.
213 00:20:11.790 ⇒ 00:20:12.220 Neil Cohen: Whoa.
214 00:20:12.220 ⇒ 00:20:16.680 Uttam Kumaran: Like, for consultants, they get a… for the… for the clients, they get a way better experience, because…
215 00:20:16.830 ⇒ 00:20:19.659 Uttam Kumaran: We move at, like, the pace of, like, a product company.
216 00:20:19.880 ⇒ 00:20:23.509 Uttam Kumaran: You know, but it’s still in, like, kind of in consulting.
217 00:20:23.510 ⇒ 00:20:25.249 Neil Cohen: How’d you get the customers?
218 00:20:25.690 ⇒ 00:20:27.150 Neil Cohen: Just word of mouth?
219 00:20:27.150 ⇒ 00:20:29.810 Uttam Kumaran: But just soul, just learn how to sell.
220 00:20:29.810 ⇒ 00:20:31.230 Neil Cohen: You just call cold call, you do all the stuff.
221 00:20:31.230 ⇒ 00:20:35.589 Uttam Kumaran: We did a lot of that, and then referrals, we did partnerships.
222 00:20:37.200 ⇒ 00:20:50.269 Uttam Kumaran: And then you just… you just do that every… try… just try everything every day for a few years, and then now it’s starting to grow because someone will leave a company, they’ll bring us to the next one, someone will ask a friend, and they…
223 00:20:50.270 ⇒ 00:20:59.419 Uttam Kumaran: And then also, we do… we still do outbound, and… but we just have a lot of case studies now, and so we’re able to use that to just win deals, and then we’re always, like, faster, and…
224 00:20:59.450 ⇒ 00:21:03.100 Uttam Kumaran: Like, more honest than most of the other consultants, like.
225 00:21:03.250 ⇒ 00:21:03.710 Neil Cohen: No.
226 00:21:03.710 ⇒ 00:21:13.029 Uttam Kumaran: These aren’t, like, freelance, this is… we’re competing on, like, like, larger consultancies, right? But they’re all, like, kind of slow, they do, like, decks, they’re talking to, like, a salesperson.
227 00:21:13.500 ⇒ 00:21:16.349 Uttam Kumaran: You know, so that’s our advantage, I feel like.
228 00:21:17.830 ⇒ 00:21:18.400 Uttam Kumaran: Yeah.
229 00:21:18.400 ⇒ 00:21:23.299 Neil Cohen: Yeah, I mean, the big cons… the big companies are like that for consulting, I remember them.
230 00:21:23.300 ⇒ 00:21:33.220 Uttam Kumaran: You know, and they just… they suck, like, it’s… you know, they’re not, like… So… Yeah.
231 00:21:33.460 ⇒ 00:21:35.070 Neil Cohen: Have you heard anything about Flowcode?
232 00:21:35.510 ⇒ 00:21:46.539 Uttam Kumaran: Not really. I just… I still talk to Patrick and, like, Ivana really often. Me and Patrick maybe talk once a month, and we’re trying to implement their product for some clients.
233 00:21:46.720 ⇒ 00:21:52.690 Uttam Kumaran: But I haven’t heard anything. I called David the other day, but…
234 00:21:52.690 ⇒ 00:21:53.920 Neil Cohen: Oh, he’s still there, yeah.
235 00:21:53.920 ⇒ 00:22:00.339 Uttam Kumaran: He’s still there, but I was asking about, like, something else. We were working with a client, you remember Prey.com?
236 00:22:01.040 ⇒ 00:22:13.369 Neil Cohen: Oh, yeah, it was a mobile app or something, right? Yeah, and we were helping them drive conversions to the app store. We were pitching a client that’s, like, similar to them, and I called them, I was like, what did we do for them again? I went, I want to use it in our pitch.
237 00:22:13.370 ⇒ 00:22:17.160 Uttam Kumaran: But I don’t know, I feel like he’s… he told me that they’re, like.
238 00:22:17.790 ⇒ 00:22:20.709 Uttam Kumaran: They’re growing and stuff like that, but I don’t know.
239 00:22:20.980 ⇒ 00:22:29.949 Neil Cohen: Cassie contacted me, like, I don’t know, several months back, because they were going to do a buyback, share buyback, so you could sell your stock back to the company.
240 00:22:29.950 ⇒ 00:22:30.990 Uttam Kumaran: Why?
241 00:22:31.310 ⇒ 00:22:46.749 Neil Cohen: I guess, I don’t know why, but I said, okay, I’m interested, but it would only be for the, whatever, the fair market value, so you get, like, whatever, 35 cents a stock, 35 cents a share, something like that, but I figured the stock’s probably not going to be worth anything anyway, so maybe I should sell some of it.
242 00:22:46.750 ⇒ 00:22:57.639 Neil Cohen: But then… and she said, oh, you gotta… you gotta tell me right away, because we’re gonna have a board meeting and try to get it approved, and then that was, like, 4 months ago, so I said, yeah, sure, I’ll sell a certain chunk of it, but then I never heard back, so I guess it didn’t get approved.
243 00:22:57.750 ⇒ 00:23:16.399 Neil Cohen: I don’t know why they… but I asked, you know, why are you doing that? She said, are you bringing in a new investor? It’s like, no, no, we’re not bringing in a new investor. Tim just wants… I guess he wants to give some money back to some of the people who were there earlier or something. I don’t know, you know, Tim’s always… Tim’s always coming up with some crazy shit. But I guess the board must have denied it or something.
244 00:23:16.670 ⇒ 00:23:19.330 Uttam Kumaran: When did you… when did you get out of there?
245 00:23:19.840 ⇒ 00:23:34.930 Neil Cohen: It’s been, like, 3 years now, around 3 years, because they brought in this new CTO, and he just wanted to rewrite everything, he just wanted to redo everything. Even though everything was working fine, you know, he comes in with his own architecture patterns, and he wanted to hire everyone from his old company.
246 00:23:35.230 ⇒ 00:23:38.889 Neil Cohen: I forget what it was, Blue something, I don’t remember the name of the company, so…
247 00:23:39.070 ⇒ 00:23:49.449 Neil Cohen: But he only lasted, so I left. He joined, and I left a few months later, and then he brought in more of his guys. But then he got kicked out, like, in two years, so I don’t know who the CTO is now.
248 00:23:49.830 ⇒ 00:23:50.440 Uttam Kumaran: Yeah, I don’t know.
249 00:23:50.440 ⇒ 00:23:52.479 Neil Cohen: Yeah, Tim going through different people.
250 00:23:52.830 ⇒ 00:23:55.699 Uttam Kumaran: I don’t know if they ended up raising any more money.
251 00:23:55.700 ⇒ 00:24:00.789 Neil Cohen: Yeah, I mean, Cassie said there was, I think their revenue was, like, 15 million.
252 00:24:01.130 ⇒ 00:24:05.110 Uttam Kumaran: Okay. But the growth wasn’t… they weren’t hitting their growth target, so…
253 00:24:05.600 ⇒ 00:24:12.840 Neil Cohen: I think that that’s the big problem, is that they’re not growing the way that they need to in order to justify the valuation.
254 00:24:13.400 ⇒ 00:24:19.279 Uttam Kumaran: I mean, but still, like, I wonder how they’re… like, Tim must have put in way more money, because the last time they raised…
255 00:24:19.690 ⇒ 00:24:24.389 Uttam Kumaran: I thought was when I… when I… in 2020, like, I don’t think they’ve raised since then.
256 00:24:24.390 ⇒ 00:24:41.389 Neil Cohen: No, there were two rounds. I mean, there was a round when I was there, then when I left, they… because I think when I was there, they raised a round, and the valuation was, like, 300-something million. Then I left, and I think Alex Markowitz told me that they had raised another amount, then the valuation was, like, $225 million, so it went down.
257 00:24:41.520 ⇒ 00:24:47.839 Neil Cohen: But there were two rounds. But yeah, still, I think, I mean, if you do the calculation, Tim must be funneling money into the company.
258 00:24:50.630 ⇒ 00:24:53.519 Uttam Kumaran: I learned a lot from there. I learned a lot how to not run a company.
259 00:24:53.520 ⇒ 00:24:54.450 Neil Cohen: Like, I think that’s…
260 00:24:54.450 ⇒ 00:25:00.150 Uttam Kumaran: Company, there’s no… there’s no yelling, there’s no, like… Weirdness at all.
261 00:25:00.150 ⇒ 00:25:02.759 Neil Cohen: Riddling people at meetings, and yeah.
262 00:25:04.090 ⇒ 00:25:08.839 Uttam Kumaran: Yeah, there’s none of that. There’s, like, no suspect stuff at all.
263 00:25:08.940 ⇒ 00:25:12.649 Uttam Kumaran: It’s just crazy that… how these guys, like, are… are…
264 00:25:13.460 ⇒ 00:25:15.230 Uttam Kumaran: Still in business, I don’t know.
265 00:25:16.670 ⇒ 00:25:22.820 Neil Cohen: I think it’s just all Tim. I mean, he’s about, I guess, a billion dollars or whatever, and I think he likes to go into the office in New York, so…
266 00:25:23.780 ⇒ 00:25:24.310 Neil Cohen: And…
267 00:25:24.310 ⇒ 00:25:27.660 Uttam Kumaran: But it’s just… it’s just, like, it’s like people’s lives, you know?
268 00:25:29.850 ⇒ 00:25:32.550 Uttam Kumaran: So… Yeah.
269 00:25:33.160 ⇒ 00:25:36.119 Uttam Kumaran: How can I… can I help with anything? How can I be helpful?
270 00:25:37.360 ⇒ 00:25:38.170 Neil Cohen: With what?
271 00:25:38.170 ⇒ 00:25:39.180 Uttam Kumaran: With anything.
272 00:25:39.720 ⇒ 00:25:40.970 Neil Cohen: I’m not sure.
273 00:25:42.890 ⇒ 00:25:46.349 Uttam Kumaran: Yeah, I’ll let you know if we’re ending up doing any cool AI projects.
274 00:25:46.470 ⇒ 00:25:48.410 Uttam Kumaran: Right now, we’re just building some applications.
275 00:25:48.920 ⇒ 00:25:49.720 Uttam Kumaran: So…
276 00:25:53.410 ⇒ 00:26:01.149 Uttam Kumaran: But yeah, let me know how… if… let me know if you… if the… if the company ends up folding, and you need any more. You want to work on some projects.
277 00:26:01.370 ⇒ 00:26:19.659 Neil Cohen: Yeah, I mean, I don’t know, I mean, I would guess by summer it may be over, we’ll see. We may get something, but I mean, the problem is we have zero, zero customers, we just have these, we’re doing stuff with NATO, but they haven’t agreed to anything. They give us, you know, we won some innovation awards, you get a 100,000 there, so…
278 00:26:19.700 ⇒ 00:26:32.569 Neil Cohen: The problem is that the founder, he’s like the visionary type, where he doesn’t like to bend or anything like that, or, hey, let’s change this for… and try to get some customers. You know, he wants to essentially boil the ocean and conquer the world, so…
279 00:26:32.800 ⇒ 00:26:38.450 Neil Cohen: He wants everyone to use the product, not to actually solve particular problems that people have.
280 00:26:38.450 ⇒ 00:26:38.920 Uttam Kumaran: Yeah.
281 00:26:38.920 ⇒ 00:26:47.509 Neil Cohen: That’s the problem. It’s kind of like a… it’s… I mean, it was an… it’s been an interesting thing to build, but it’s really more of, like, a solution looking for a problem.
282 00:26:47.980 ⇒ 00:26:50.130 Neil Cohen: Than what most people need.
283 00:26:51.070 ⇒ 00:26:58.510 Uttam Kumaran: Yeah, I feel like… I don’t know, I couldn’t care less about, Like, winning the industry.
284 00:26:58.510 ⇒ 00:26:59.100 Neil Cohen: Yeah, honey.
285 00:26:59.100 ⇒ 00:27:02.010 Uttam Kumaran: Just, like, doing a good job with some clients, we’re making money.
286 00:27:02.010 ⇒ 00:27:06.399 Neil Cohen: Yeah, that’s good enough. You don’t need to change the world. He wants to change the world.
287 00:27:07.030 ⇒ 00:27:13.439 Uttam Kumaran: Yeah, I’m not interested in that at all. I mean, like, I like… we’re doing well for our clients, and, like, that’s what I can control, you know?
288 00:27:14.080 ⇒ 00:27:21.600 Uttam Kumaran: So… Okay. Alright, I’m gonna jump to another call, but yeah, it was great catching up.
289 00:27:21.600 ⇒ 00:27:22.700 Neil Cohen: get gratinum Bier.
290 00:27:22.700 ⇒ 00:27:27.549 Uttam Kumaran: Yeah, I’ll tell Patrick you said hi. You should definitely say hi to him. I think his product’s pretty cool, too, so…
291 00:27:28.190 ⇒ 00:27:30.089 Neil Cohen: Yeah, yeah, I’ll reach out to him sometime.
292 00:27:30.090 ⇒ 00:27:32.549 Uttam Kumaran: Yeah, totally. Okay, thanks, Neil.
293 00:27:32.550 ⇒ 00:27:33.370 Neil Cohen: Alright, bye.
294 00:27:33.370 ⇒ 00:27:34.180 Uttam Kumaran: Okay, bye.