Meeting Title: Brainforge Legal Sync Date: 2026-04-03 Meeting participants: Rico Rejoso, Holly Condos, Uttam Kumaran
WEBVTT
1 00:01:06.370 ⇒ 00:01:07.720 Holly Condos: Hey, Rico.
2 00:01:09.970 ⇒ 00:01:11.749 Rico Rejoso: Hey, Ollie, how are you?
3 00:01:11.750 ⇒ 00:01:12.569 Holly Condos: Where are you?
4 00:01:14.740 ⇒ 00:01:16.080 Rico Rejoso: I’m good, good.
5 00:01:16.530 ⇒ 00:01:19.470 Holly Condos: Good. How have things been? Are you busy?
6 00:01:20.810 ⇒ 00:01:26.719 Rico Rejoso: Beth, for, I mean, we’ve been ramping up hiring so far, so, yeah, contract has.
7 00:01:26.720 ⇒ 00:01:27.580 Holly Condos: Yeah.
8 00:01:28.170 ⇒ 00:01:30.120 Holly Condos: Yeah, that’s great. I’m glad.
9 00:01:31.770 ⇒ 00:01:33.960 Uttam Kumaran: Hey, everyone. Hi, Holly.
10 00:01:34.150 ⇒ 00:01:35.559 Holly Condos: Hey, how are you?
11 00:01:36.750 ⇒ 00:01:39.820 Uttam Kumaran: Hey, I’m good, I’m actually in, in the Bay Area.
12 00:01:41.280 ⇒ 00:01:43.359 Holly Condos: Oh, nice, I’m in Philadelphia.
13 00:01:44.230 ⇒ 00:01:49.960 Uttam Kumaran: Oh, nice! What brings you to Philly?
14 00:01:51.090 ⇒ 00:01:59.900 Holly Condos: My… my older son’s wife is Jewish by birth, and so we,
15 00:02:00.750 ⇒ 00:02:05.699 Holly Condos: We’ve been invited to celebrate Passover with them the past few years.
16 00:02:06.060 ⇒ 00:02:08.490 Holly Condos: Okay. So, we came for that.
17 00:02:09.440 ⇒ 00:02:10.300 Uttam Kumaran: Nice!
18 00:02:11.170 ⇒ 00:02:12.250 Uttam Kumaran: Amazing.
19 00:02:13.000 ⇒ 00:02:19.439 Holly Condos: I’m in the hotel, and I’m supposed to be out here shortly, so that’s why I don’t have my camera on, but…
20 00:02:19.550 ⇒ 00:02:28.059 Holly Condos: Okay, so per your agenda, do you guys, for processing the credit card payments.
21 00:02:28.190 ⇒ 00:02:34.700 Holly Condos: Do you have Stripe already, or have you thought about, you know, how you would go about it?
22 00:02:35.600 ⇒ 00:02:40.530 Uttam Kumaran: Yeah, I mean, like, I… I feel like my default is to just put
23 00:02:40.920 ⇒ 00:02:48.470 Uttam Kumaran: like, put it like a terms of service kind of thing, where you can just click in, pay through Stripe. You can also do through…
24 00:02:48.860 ⇒ 00:03:00.560 Uttam Kumaran: QuickBooks, if QuickBooks has, like, an alternative product, because I assume that because we’re already a QuickBooks customer, maybe there’s lower fees, or, like, it hits the account faster.
25 00:03:01.680 ⇒ 00:03:03.070 Uttam Kumaran: But whether.
26 00:03:03.070 ⇒ 00:03:03.520 Holly Condos: Yeah, it’s quick.
27 00:03:03.520 ⇒ 00:03:08.620 Uttam Kumaran: or Stripe, yeah, that’s sort of, like, what I was thinking. It’s almost like… subscribing…
28 00:03:08.910 ⇒ 00:03:13.830 Uttam Kumaran: to, like, a software, you know? It’s some of… it’s just advice we got from…
29 00:03:14.040 ⇒ 00:03:17.340 Uttam Kumaran: Another agency owner who was like, yeah, it just helped…
30 00:03:17.840 ⇒ 00:03:20.479 Uttam Kumaran: Like, get cash in the door faster, you know?
31 00:03:20.910 ⇒ 00:03:28.630 Holly Condos: Yeah, yeah, and you’re on the right track. So, I did a little bit of… well, I have one customer who recently
32 00:03:29.700 ⇒ 00:03:44.319 Holly Condos: added credit card payments. So, I wasn’t on the ground with them, but they… just in the last year, they did it. So, I talked to him, and then I did a little bit of research, but yeah, to your point.
33 00:03:44.480 ⇒ 00:03:50.329 Holly Condos: So you need a payment processor, and QuickBooks does have a payments
34 00:03:50.660 ⇒ 00:04:04.270 Holly Condos: module, so yeah, I think if you’ve already got that subscription, then they probably have that that you could add on to your subscription, right? That would make sense, and it would integrate with QuickBooks, and that would all be easy.
35 00:04:05.600 ⇒ 00:04:08.610 Uttam Kumaran: Yeah. I guess I’m curious, like, from…
36 00:04:08.850 ⇒ 00:04:09.620 Holly Condos: yours?
37 00:04:09.620 ⇒ 00:04:12.640 Uttam Kumaran: Like, basically, we wanted to say, like.
38 00:04:12.810 ⇒ 00:04:20.389 Uttam Kumaran: anything less, like, for example, we’re starting to do these, like, quick, like, 2-week audits, or one-month audits, 5K or 10K.
39 00:04:20.529 ⇒ 00:04:25.689 Uttam Kumaran: That we don’t really redline, or maybe, like, we’re just gonna say we don’t redline these.
40 00:04:26.020 ⇒ 00:04:27.230 Uttam Kumaran: Wow.
41 00:04:28.070 ⇒ 00:04:29.580 Uttam Kumaran: Right? So, like, any…
42 00:04:29.580 ⇒ 00:04:30.710 Holly Condos: Right?
43 00:04:31.520 ⇒ 00:04:32.569 Uttam Kumaran: Sorry, say it again?
44 00:04:33.160 ⇒ 00:04:34.250 Holly Condos: Like a packet.
45 00:04:34.830 ⇒ 00:04:36.060 Uttam Kumaran: Yeah, exactly, exactly.
46 00:04:36.060 ⇒ 00:04:37.490 Holly Condos: Yeah.
47 00:04:39.160 ⇒ 00:04:40.049 Holly Condos: Yeah, so you don’.
48 00:04:40.050 ⇒ 00:04:40.899 Uttam Kumaran: No, you don’t.
49 00:04:40.900 ⇒ 00:04:47.540 Holly Condos: line it, you want to just say, here, we can do this assessment. You want them to pay up front, you want a customer to pay up front?
50 00:04:48.120 ⇒ 00:04:49.110 Uttam Kumaran: Yes.
51 00:04:49.840 ⇒ 00:04:50.500 Holly Condos: Yeah.
52 00:04:51.540 ⇒ 00:05:01.680 Uttam Kumaran: The deliverables are pretty… the deliverables are pretty clear. I mean, I… I think, like, additionally, we’re starting to do half up front, half at the end. I think all of those are…
53 00:05:02.140 ⇒ 00:05:07.010 Uttam Kumaran: you know, possible. I think the other piece is just, even with the half up front, though.
54 00:05:07.200 ⇒ 00:05:12.969 Uttam Kumaran: it’s like, we invoice, and then it’s, like, net 30, and so I kind of just want to skip that, you know?
55 00:05:14.370 ⇒ 00:05:15.850 Holly Condos: Yeah, absolutely.
56 00:05:16.510 ⇒ 00:05:21.800 Holly Condos: Right. So, okay, so maybe we need a…
57 00:05:22.400 ⇒ 00:05:30.420 Holly Condos: A sow just for the assessments that have the standard boilerplate language.
58 00:05:30.620 ⇒ 00:05:38.459 Holly Condos: with the payment upfront language, is that what you’re thinking? Just, like, a separate standard style? Okay.
59 00:05:38.460 ⇒ 00:05:42.710 Uttam Kumaran: Kind of, yeah, and then I’ll just put that behind, like, a embedded…
60 00:05:43.000 ⇒ 00:05:45.999 Uttam Kumaran: Like, link where you can just review and then you sign.
61 00:05:46.990 ⇒ 00:05:57.370 Holly Condos: Perfect. Do you want me to draft the sow, or what would you like me to do there? Give you a shell, and you can dump in your assessment stuff? What would you like?
62 00:05:57.370 ⇒ 00:06:02.879 Uttam Kumaran: Yeah, if you can give me, like, a shell, and then, I can just…
63 00:06:03.360 ⇒ 00:06:07.680 Uttam Kumaran: We can take, like, one of our assessments, yeah, as an example.
64 00:06:07.680 ⇒ 00:06:08.210 Holly Condos: sample.
65 00:06:08.210 ⇒ 00:06:09.140 Uttam Kumaran: And then…
66 00:06:09.250 ⇒ 00:06:16.040 Uttam Kumaran: I can hand that to Robert to start to use, and then we’ll… I think once we agree on that, I can then work on the,
67 00:06:16.390 ⇒ 00:06:21.349 Uttam Kumaran: like, how does it work? Like, is it a Stripe link or a QuickBooks link? You know, things like that.
68 00:06:21.930 ⇒ 00:06:24.419 Holly Condos: Sounds good. Okay, so that’s… that’s easy.
69 00:06:24.590 ⇒ 00:06:30.320 Holly Condos: And then on your AI services terms.
70 00:06:30.730 ⇒ 00:06:34.989 Holly Condos: So, I know we had this come up with Ellie, right?
71 00:06:36.240 ⇒ 00:06:37.000 Uttam Kumaran: Yes.
72 00:06:37.440 ⇒ 00:06:41.939 Holly Condos: what I’m thinking, just… Based on the…
73 00:06:42.370 ⇒ 00:06:55.090 Holly Condos: type of work that Brainforge is doing and the number of customers, rather than tailor… taking time to tailor every single cell, what I thought was that
74 00:06:55.280 ⇒ 00:06:57.500 Holly Condos: We could just have an addendum
75 00:06:57.860 ⇒ 00:07:06.320 Holly Condos: with terms that apply to AI services, right? So if you’re doing…
76 00:07:06.420 ⇒ 00:07:09.300 Holly Condos: And maybe that’s wrong, I mean, you tell me, like.
77 00:07:10.170 ⇒ 00:07:17.309 Holly Condos: If you want it to be embedded in the MSA as is, we can also do that. I just wasn’t sure what you were thinking.
78 00:07:19.200 ⇒ 00:07:29.270 Uttam Kumaran: Yeah, I guess, can you brief me on what is going on on the Ellie side? I just have, like, some cursory understanding that they have some requirements, like…
79 00:07:29.420 ⇒ 00:07:33.419 Uttam Kumaran: I’d just love, like, just a quick, like, reminder of, like, what’s going on there.
80 00:07:33.600 ⇒ 00:07:44.080 Holly Condos: Sure, sure. So, so my question to Rob… or my series of questions to Robert was, is the information
81 00:07:44.550 ⇒ 00:07:50.810 Holly Condos: In the… in the call logs, that we would be… Exposed to…
82 00:07:51.720 ⇒ 00:07:57.879 Holly Condos: PHI or PPI, right? Private Health Information, Private,
83 00:07:58.530 ⇒ 00:08:06.810 Holly Condos: private personal information that falls under HIPAA, and if it does, Right? Because they, Ellie.
84 00:08:07.030 ⇒ 00:08:11.909 Holly Condos: Our, our governing… governing… governor…
85 00:08:12.330 ⇒ 00:08:15.389 Holly Condos: Governor Ning there. I can’t even say that word, why is that?
86 00:08:15.390 ⇒ 00:08:17.400 Uttam Kumaran: Wait, gover… governing. Governing.
87 00:08:17.400 ⇒ 00:08:18.100 Holly Condos: Perfect.
88 00:08:20.560 ⇒ 00:08:25.380 Holly Condos: They… their… their whole environment
89 00:08:25.540 ⇒ 00:08:40.589 Holly Condos: under HIPAA compliance, right? So, as the service provider, we need to either say, we’re not dealing with PHI and PPI, and so we’re not going to sign up for being in compliance with every
90 00:08:40.590 ⇒ 00:08:48.910 Holly Condos: secure… health… HIPAA and health security regulation and law there is in the world, because that’s the language that they put in the MSA.
91 00:08:49.090 ⇒ 00:09:02.200 Holly Condos: Right? And we carve that out, or we say, okay, yes, we are… we are going to be dealing with it, but the call log data is metadata, and so we’re not going to see anything
92 00:09:02.680 ⇒ 00:09:03.770 Holly Condos: That’s personal.
93 00:09:03.770 ⇒ 00:09:04.600 Uttam Kumaran: Hmm.
94 00:09:04.800 ⇒ 00:09:10.529 Holly Condos: Or, third, it’s not metadata, we are gonna see it, and here’s…
95 00:09:10.780 ⇒ 00:09:21.550 Holly Condos: how we meet the standard for securing that information under HIPAA. So it’s different than the general confidentiality umbrella.
96 00:09:21.660 ⇒ 00:09:24.369 Holly Condos: It’s more,
97 00:09:24.650 ⇒ 00:09:30.510 Holly Condos: Now, I wouldn’t say it’s more detailed, it’s just in compliance with the way that the HIPAA regulation is written.
98 00:09:30.640 ⇒ 00:09:32.850 Holly Condos: And as the service provider.
99 00:09:33.000 ⇒ 00:09:47.460 Holly Condos: there are a couple things that the service provider would just say, yes, you know, we are in comp… we’re handling your data, Ellie, this way, and it’s in compliance with HIPAA. And that’s what I was trying to get to with Robert.
100 00:09:47.670 ⇒ 00:09:48.810 Uttam Kumaran: Yeah.
101 00:09:49.330 ⇒ 00:09:54.420 Holly Condos: what I glean, and I haven’t spoken to him, but what I glean from our exchange in Slack
102 00:09:54.540 ⇒ 00:10:01.030 Holly Condos: is that either the data that we’re gonna see is not PHI,
103 00:10:01.420 ⇒ 00:10:04.729 Holly Condos: And or it’s meta, so we should be okay.
104 00:10:05.380 ⇒ 00:10:06.309 Holly Condos: That, that, but that’s.
105 00:10:06.310 ⇒ 00:10:19.769 Uttam Kumaran: Yeah, that’s… that’s… that’s also my case, but what I… what we can do is just confirm that. I… I… I think, knowing that, can you, maybe we can go back to your original suggestion, was just, like, just to have…
106 00:10:20.030 ⇒ 00:10:24.450 Uttam Kumaran: like, a document in case this happens, or… yeah, I guess walk me through that.
107 00:10:25.280 ⇒ 00:10:32.770 Holly Condos: Yeah, so I kind of just put together a little bit of an overview, so…
108 00:10:33.400 ⇒ 00:10:35.300 Holly Condos: Really, the big push
109 00:10:35.950 ⇒ 00:10:45.879 Holly Condos: for terms where AI, and that’s very general, but where… where a service provider is either using AI tools.
110 00:10:46.040 ⇒ 00:10:47.470 Uttam Kumaran: Oh, okay, okay, great, great.
111 00:10:47.470 ⇒ 00:10:48.660 Holly Condos: the output.
112 00:10:48.860 ⇒ 00:10:55.310 Holly Condos: Right? So, so if we’re using Claude for the output,
113 00:10:55.640 ⇒ 00:11:07.490 Holly Condos: and we’re giving that to the customer, but then the customer is doing something with that output for an eventual outcome. That’s one scenario, right? Are you with me?
114 00:11:07.490 ⇒ 00:11:09.220 Uttam Kumaran: Yeah, yes, yes, yes.
115 00:11:09.380 ⇒ 00:11:13.990 Holly Condos: But if we are not only using the AI tool.
116 00:11:15.070 ⇒ 00:11:27.080 Holly Condos: for the output, and then utilizing the output for the solution and the outcome, then it’s a slightly different liability lane.
117 00:11:27.800 ⇒ 00:11:36.350 Holly Condos: Okay? That’s basically the big feature from a legal perspective. And then you want to carve out
118 00:11:36.430 ⇒ 00:11:51.959 Holly Condos: within… within those scenarios, we want the terms to say, okay, based on the scope and what Brainforge is doing, this… these items are our responsibility, customer, and these are yours, customer, right? So we… we carve out
119 00:11:52.110 ⇒ 00:11:56.589 Holly Condos: What… based on what we’re doing, who’s responsible for what?
120 00:11:56.590 ⇒ 00:12:17.979 Holly Condos: So that if there is a question, or if there is an issue, we can always point back to, well, customer, we told you, right, that we were giving you an assessment. It… the assessment was built on, you know, just for an example, built on our use of cursor and various tools, but that’s only to enhance our…
121 00:12:17.980 ⇒ 00:12:19.849 Holly Condos: Provision of the assessment.
122 00:12:20.090 ⇒ 00:12:27.380 Holly Condos: And what you do with the results of the assessment is your thing, right? If you’re determining some outcome
123 00:12:27.560 ⇒ 00:12:39.249 Holly Condos: based on the results of our assessment, that’s your baby, not ours. All we did was use cursor, etc, to put together the framework of our assessment, and we disclosed that.
124 00:12:40.270 ⇒ 00:12:42.319 Holly Condos: Okay? Are you with me?
125 00:12:42.850 ⇒ 00:12:43.700 Uttam Kumaran: Yes.
126 00:12:44.240 ⇒ 00:12:52.050 Holly Condos: And then, you know, overlaying it is the data use and privacy, so…
127 00:12:52.290 ⇒ 00:12:55.560 Holly Condos: If we say… let’s just use Ellie.
128 00:12:55.670 ⇒ 00:13:05.130 Holly Condos: well, let’s use an assessment customer. So, again, if we say, here’s our report from our assessment, But,
129 00:13:05.920 ⇒ 00:13:11.400 Holly Condos: But it doesn’t involve any personal information of the customer, which it won’t, right?
130 00:13:11.560 ⇒ 00:13:15.869 Holly Condos: Or, I guess, could it? Mmm, I guess it could if it’s Ellie.
131 00:13:17.140 ⇒ 00:13:18.689 Uttam Kumaran: It’s Ellie, yes.
132 00:13:24.120 ⇒ 00:13:34.540 Holly Condos: Well, if it’s Ellie, okay? So then we would say, hey, Ellie, we’re… we’re… we use… remember I asked you, I think, before, maybe on,
133 00:13:35.220 ⇒ 00:13:39.840 Holly Condos: Maybe it was Ellie, or was it, kaylint?
134 00:13:40.560 ⇒ 00:13:45.190 Holly Condos: There was another one I asked you about, Utam, where…
135 00:13:45.490 ⇒ 00:13:51.780 Holly Condos: I asked about the internal tools that we use, right, to protect a customer’s data.
136 00:13:52.120 ⇒ 00:13:53.070 Holly Condos: And you’re like.
137 00:13:53.070 ⇒ 00:14:04.680 Uttam Kumaran: Oh, yeah, yeah, yeah, I mentioned, yeah, we’re doing a lot of the… yeah, exactly, like, all the different governance policies. So we can… so on our side, yes, I think we can make it clear in all the ways that we…
138 00:14:04.890 ⇒ 00:14:10.630 Uttam Kumaran: basically do the… I think you mentioned there was a best practice like…
139 00:14:10.750 ⇒ 00:14:21.590 Uttam Kumaran: governance thing, and you list it out, I’m like, we can enforce that we’re doing all of that, and make sure that’s clear. And that’s just industry standard, like, data governance.
140 00:14:21.590 ⇒ 00:14:22.050 Holly Condos: Yes.
141 00:14:22.050 ⇒ 00:14:33.399 Uttam Kumaran: You know, for companies that are dealing with, like, more advanced PII, right, because, you know, we can start to layer on, like, as we get to enterprise, I expect that some people will be like.
142 00:14:33.460 ⇒ 00:14:46.419 Uttam Kumaran: there’s also gonna be things, like, you need to run… your… all your team needs to run on our machines, or all… your team needs to run on our… on our VPS, and, like, I’m assuming that’s gonna… that’s gonna…
143 00:14:47.140 ⇒ 00:15:00.260 Uttam Kumaran: come pretty soon. So, that’s definitely one piece, which is just on, like, broadly, how does Brainforge internally handle data, handle client data, handle client-customer data, right?
144 00:15:00.310 ⇒ 00:15:12.280 Uttam Kumaran: Right. So that’s, like, one piece. And then I think the AI piece is kind of as you described, so I think that’s the first time I sort of heard it laid out that way. So, yes, we are…
145 00:15:12.750 ⇒ 00:15:20.439 Uttam Kumaran: We are, we are, in some situations, passing data
146 00:15:20.890 ⇒ 00:15:26.350 Uttam Kumaran: Into language models for conducting the service.
147 00:15:26.570 ⇒ 00:15:28.249 Uttam Kumaran: And yeah…
148 00:15:28.250 ⇒ 00:15:28.650 Holly Condos: This is the.
149 00:15:28.650 ⇒ 00:15:29.000 Uttam Kumaran: times it.
150 00:15:29.000 ⇒ 00:15:29.849 Holly Condos: of data?
151 00:15:30.470 ⇒ 00:15:32.109 Uttam Kumaran: It is the customer’s data.
152 00:15:33.630 ⇒ 00:15:44.640 Holly Condos: And when you say we’re… sorry to interrupt, but this is important. When you say you’re passing it into an LLM, is it the Brainforge LLM, or what?
153 00:15:45.360 ⇒ 00:15:56.139 Uttam Kumaran: No, like, let me give you, like, let me give you an example, like, someone on our analy… one of our analysts is using, like, Claude to help them conduct, like, data analysis for… for a client.
154 00:15:57.560 ⇒ 00:15:58.320 Holly Condos: A.
155 00:15:59.320 ⇒ 00:16:04.470 Holly Condos: But it’s Brainforge’s premium subscription to Claude, where we can say, do not.
156 00:16:04.470 ⇒ 00:16:04.960 Uttam Kumaran: Yes.
157 00:16:04.960 ⇒ 00:16:11.669 Holly Condos: Do not train yourself on this data. Do not disclose this data. 100%, yes. This data’s encrypted.
158 00:16:12.060 ⇒ 00:16:21.800 Uttam Kumaran: Exactly. So, number one, it’s, like, all… it’s not like their… it’s not the… it’s not like their, like, free ChatGPT plan. It’s… it’s all through our,
159 00:16:22.120 ⇒ 00:16:31.869 Uttam Kumaran: SSO, it’s all through our enterprise plan, where we hit the don’t train button, we sign whatever we signed with them.
160 00:16:32.250 ⇒ 00:16:32.830 Holly Condos: Yeah.
161 00:16:32.830 ⇒ 00:16:38.199 Uttam Kumaran: But I just… even with that, I don’t think… That… that’s enough.
162 00:16:38.310 ⇒ 00:16:47.120 Uttam Kumaran: Like, we all… we’ve done that the whole… we’ve done that the whole time, and so there’s two… there’s two kind of paths I see, like, one,
163 00:16:47.940 ⇒ 00:16:55.969 Uttam Kumaran: Like, let me just lay out a few options, I just… probably… it just may seem, like, word salad, it’s the first time I’ve just, like, had a conversation about this, so…
164 00:16:55.970 ⇒ 00:16:56.970 Holly Condos: Yeah, no, it’s all good.
165 00:16:56.970 ⇒ 00:16:58.760 Uttam Kumaran: So,
166 00:16:59.340 ⇒ 00:17:09.169 Uttam Kumaran: in some situations, I could see cli… like, in all of our client work, and most of our clients are doing this themselves.
167 00:17:09.369 ⇒ 00:17:22.090 Uttam Kumaran: Right? So internally, they are throwing stuff into their enterprise clods or chat GPTs, and in many situations, we find that their employees are doing this on their personal stuff, too.
168 00:17:22.410 ⇒ 00:17:23.810 Holly Condos: Which is…
169 00:17:23.810 ⇒ 00:17:28.680 Uttam Kumaran: pretty insane, right? And so, like, so we’re seeing that. So… so second,
170 00:17:29.160 ⇒ 00:17:47.369 Uttam Kumaran: on our side, yes, we use AI to develop and to deliver our services, whether it’s, like, on the project management side, whether it’s on the data side or engineering side, it’s basically, like, one of our competitive advantages. And yes, as you put it, it is contributing to us
171 00:17:47.740 ⇒ 00:17:56.679 Uttam Kumaran: developing and actually delivering the final output, or final outcome, that they’re paying for.
172 00:17:56.880 ⇒ 00:18:06.969 Uttam Kumaran: Within that, yes, there… we are passing data to cloud-hosted LLMs, where we are… we’ve hit every button that they offer.
173 00:18:07.230 ⇒ 00:18:14.380 Uttam Kumaran: To maintain privacy. A couple of options, right? One, We could…
174 00:18:14.590 ⇒ 00:18:22.100 Uttam Kumaran: moved to, like, Brainforge, locally hosted, completely, like, off-the-internet language model.
175 00:18:22.100 ⇒ 00:18:22.890 Holly Condos: Excellent.
176 00:18:23.220 ⇒ 00:18:29.070 Uttam Kumaran: I completely moved to, like, Azure’s most super secure… like…
177 00:18:29.710 ⇒ 00:18:42.580 Uttam Kumaran: whatever set of, like, endpoints, right? So migrate off of using, like, ChatGBT, Claude, or any of, like, the, like, cursor, and just moving everything to…
178 00:18:42.650 ⇒ 00:18:58.519 Uttam Kumaran: Azure, or we kind of keep the status quo, which is, like, Brainforge provides all of our employees with access to the latest models to conduct services on behalf of clients. All of those models and the way they access data are all governed by our typical
179 00:18:58.780 ⇒ 00:19:10.160 Uttam Kumaran: governance policies, which is, like, it’s just that everybody’s, like, kind of on NDA, everybody’s, just accessing the clients they have access to, we have audit logs, right? So those are sort of, like.
180 00:19:11.000 ⇒ 00:19:14.439 Uttam Kumaran: The… and that’s, like, kind of… the third thing is where we are.
181 00:19:14.600 ⇒ 00:19:15.780 Uttam Kumaran: You know, right now.
182 00:19:17.660 ⇒ 00:19:18.530 Holly Condos: Okay.
183 00:19:19.660 ⇒ 00:19:20.780 Holly Condos: So…
184 00:19:21.540 ⇒ 00:19:28.459 Holly Condos: Have any customers… well, I guess I would have seen it, so I guess I know the answer. Nobody’s really pushed back.
185 00:19:28.460 ⇒ 00:19:31.309 Uttam Kumaran: No, no. Except for Ellie.
186 00:19:32.220 ⇒ 00:19:32.730 Holly Condos: But as…
187 00:19:32.730 ⇒ 00:19:45.539 Uttam Kumaran: Kelly is pushing back a little bit, it seems like, just on health and PII, because they’re a health company. No company we’ve talked to has been pissed off that we’re using AI to deliver.
188 00:19:45.540 ⇒ 00:19:46.040 Holly Condos: I don’t.
189 00:19:46.040 ⇒ 00:19:55.430 Uttam Kumaran: faster. In fact, we’re… we are actively helping them convert their businesses into that. I’m just trying to… I’m kind of just trying to get ahead
190 00:19:55.660 ⇒ 00:19:56.750 Uttam Kumaran: of it…
191 00:19:56.980 ⇒ 00:20:03.270 Uttam Kumaran: Because I want us to have a clear perspective. I don’t want a client to come and be like, hey, like.
192 00:20:03.630 ⇒ 00:20:14.660 Uttam Kumaran: I just, like, for example, I could see a client read some Doomsday article, come to us, and be like, hey, are you guys using AI? And this sort of gets us down some rabbit hole, right? So…
193 00:20:14.660 ⇒ 00:20:15.840 Holly Condos: Exactly, right.
194 00:20:16.620 ⇒ 00:20:32.760 Uttam Kumaran: I think baking it… baking the language into… into something seems reasonable. Maybe it protects us from, like, hey, we’re using this, it’s under our similar data governance policies, we are sending data to…
195 00:20:33.060 ⇒ 00:20:38.279 Uttam Kumaran: These providers, and we… these are the precautions, and here are their terms.
196 00:20:38.530 ⇒ 00:20:42.850 Uttam Kumaran: Like, because we don’t… we also turn off all, like, the model training stuff.
197 00:20:43.100 ⇒ 00:20:50.409 Uttam Kumaran: We can… I would prefer us to actually go towards the world that I mentioned, which is, like, it’s a Brain Forge models, or…
198 00:20:50.410 ⇒ 00:20:50.760 Holly Condos: Yeah.
199 00:20:50.760 ⇒ 00:20:57.019 Uttam Kumaran: it’s like a super secure Azure. It’s… it’s just more work. We just… there’s not… it’s on the roadmap right now, so…
200 00:20:57.140 ⇒ 00:20:58.170 Uttam Kumaran: Yeah.
201 00:20:58.610 ⇒ 00:21:02.679 Holly Condos: Well, certainly, it’s a business consideration and a cost, right?
202 00:21:03.010 ⇒ 00:21:05.029 Holly Condos: So, so here’s what I.
203 00:21:05.030 ⇒ 00:21:08.570 Uttam Kumaran: And I guess, Mike, I guess, Mike, I guess maybe to ask you, like, is it…
204 00:21:09.150 ⇒ 00:21:21.390 Uttam Kumaran: you know, how did this… how does this work with, like, hey, you know, team members are using, like, MacBooks, and Mac collects system diagnostics, like, is that an issue? Versus…
205 00:21:21.510 ⇒ 00:21:25.830 Uttam Kumaran: Is it more of an issue with, like, the outputs are…
206 00:21:26.430 ⇒ 00:21:30.730 Uttam Kumaran: Is actually a problem with, like, the output is, like, not deterministic, and, like…
207 00:21:30.950 ⇒ 00:21:35.280 Uttam Kumaran: it may not be 100%… like, I just… I’m not trying to… I’m trying to understand, like, where…
208 00:21:36.250 ⇒ 00:21:39.180 Uttam Kumaran: What, like, what could be the reason, you know?
209 00:21:39.690 ⇒ 00:21:40.110 Holly Condos: Yeah.
210 00:21:40.110 ⇒ 00:21:40.580 Uttam Kumaran: Well, yeah.
211 00:21:40.580 ⇒ 00:21:50.949 Holly Condos: So… So, so, let’s couch it in terms of the enterprise world. So, what’s happening with enterprise… enterprises…
212 00:21:51.250 ⇒ 00:21:58.850 Holly Condos: is, that they’re saying in their MSAs, Hey, service provider.
213 00:21:59.920 ⇒ 00:22:08.919 Holly Condos: In the course of providing the solution to… for us, If… you… Want to use the
214 00:22:09.380 ⇒ 00:22:19.090 Holly Condos: a generative AI tool, Which, you know, they don’t typically identify, they just use that high.
215 00:22:19.870 ⇒ 00:22:20.230 Uttam Kumaran: Yeah.
216 00:22:21.310 ⇒ 00:22:27.520 Holly Condos: We want you, service provider, to ask us… to tell us, and ask us…
217 00:22:27.920 ⇒ 00:22:42.230 Holly Condos: If it’s okay, and explain to us how you’re going to do it. So the way I’ve gotten around that, because, you know, that’s pretty invasive. I mean, you don’t want to have the customer in your knickers about, like, using clot, right?
218 00:22:42.230 ⇒ 00:22:45.040 Uttam Kumaran: Yeah, I mean, I don’t… yeah, it’s sort… and it also…
219 00:22:45.180 ⇒ 00:22:47.419 Uttam Kumaran: I, I, I think it’s, it’s like…
220 00:22:48.290 ⇒ 00:22:53.400 Uttam Kumaran: maybe 2 years ago, or a year ago, people used to ask me. Now, it’s like…
221 00:22:53.860 ⇒ 00:23:02.059 Uttam Kumaran: I don’t… who in their right mind would be like, don’t use Claude to do this stuff? Like, I don’t… we won’t work with them, I guess, frankly, because I think.
222 00:23:02.060 ⇒ 00:23:02.430 Holly Condos: Yeah.
223 00:23:02.430 ⇒ 00:23:04.630 Uttam Kumaran: Because that’s how our business operates, you know?
224 00:23:04.630 ⇒ 00:23:10.369 Holly Condos: I think that out of enterprise, it’s just kind of implicitly assumed that…
225 00:23:10.370 ⇒ 00:23:11.849 Uttam Kumaran: Yes, yeah, fair, fair.
226 00:23:12.130 ⇒ 00:23:17.690 Holly Condos: But to… but to your point, as a business owner, I think.
227 00:23:17.690 ⇒ 00:23:22.100 Uttam Kumaran: listen, it’s not enough for me. I’m like, I don’t want to get jammed, you know?
228 00:23:22.100 ⇒ 00:23:27.969 Holly Condos: Exactly. So, as a business owner, I think it’s prudent to get ahead of
229 00:23:28.810 ⇒ 00:23:35.350 Holly Condos: of it. Having said that, I think we also should discuss, explore.
230 00:23:35.730 ⇒ 00:23:41.489 Holly Condos: Whether and to what extent… not whether… to what extent we…
231 00:23:41.790 ⇒ 00:23:45.369 Holly Condos: Advise or disclose to the client
232 00:23:45.750 ⇒ 00:23:48.479 Holly Condos: Right? So, in this case, I think less is more.
233 00:23:49.640 ⇒ 00:23:53.049 Holly Condos: Because more is more… is like being…
234 00:23:53.310 ⇒ 00:23:58.100 Holly Condos: A defendant in a court case, where if you say.
235 00:23:58.520 ⇒ 00:24:01.769 Holly Condos: more than yes, it just opens the club door.
236 00:24:02.250 ⇒ 00:24:05.100 Uttam Kumaran: Exactly. So, like, I think if… if it’s…
237 00:24:05.890 ⇒ 00:24:20.620 Uttam Kumaran: if it’s fair just to say, yeah, we use generative AI tools to conduct this… I mean, the… for me, the business reason is, like, this… this is, like, what… why we’re in business, you know? But the… I want us to also… I want it to be, like…
238 00:24:20.770 ⇒ 00:24:27.830 Uttam Kumaran: I want there to be no doubt about, like, are we using AI, and did we sign something that said that they could use AI?
239 00:24:27.990 ⇒ 00:24:39.649 Uttam Kumaran: I mean, for me, the big… a lot of the worry, and we are also thinking about how to capitalize this, you know, on the business… on the commercial side, is, like, people are like, oh, is OpenAI stealing my, like, company secrets?
240 00:24:39.930 ⇒ 00:24:40.540 Holly Condos: Right.
241 00:24:40.540 ⇒ 00:24:41.010 Uttam Kumaran: Fair.
242 00:24:41.010 ⇒ 00:24:41.330 Holly Condos: Right.
243 00:24:41.330 ⇒ 00:24:42.640 Uttam Kumaran: Fair.
244 00:24:43.810 ⇒ 00:24:51.659 Uttam Kumaran: Yeah, I understand that, and we’re gonna develop some services that’ll say, hey, if you feel like that could happen, we have another…
245 00:24:51.720 ⇒ 00:25:07.829 Uttam Kumaran: service for you, so, like, it’s… I’m not… it’s… in that situation, right, it’s going to be… like, there are some people who are like, I need completely private, local LLMs, it’s gonna just be a different price tag, and we’re gonna take that work as well. So I’m actually not worried about…
246 00:25:08.270 ⇒ 00:25:17.859 Uttam Kumaran: I’m not worried about people being like, I’d love to work with you, we just need to have this constraint. Perfect. Like, we have another service offering where you can run
247 00:25:18.000 ⇒ 00:25:21.909 Uttam Kumaran: This completely locally, it won’t hit any other servers, blah blah blah.
248 00:25:22.470 ⇒ 00:25:23.060 Holly Condos: No.
249 00:25:23.430 ⇒ 00:25:32.389 Uttam Kumaran: But outside… but that, again, that’s, like, just commercial terms. I feel like this is more just, like, make it really clear that, yeah, we’re using AI,
250 00:25:32.870 ⇒ 00:25:37.960 Uttam Kumaran: Everybody, and here’s kind of, like, our standard data governance policies.
251 00:25:38.280 ⇒ 00:25:38.650 Holly Condos: Yup.
252 00:25:38.650 ⇒ 00:25:40.030 Uttam Kumaran: And, yeah.
253 00:25:41.000 ⇒ 00:25:43.520 Holly Condos: I think that that is a good approach.
254 00:25:43.790 ⇒ 00:25:56.569 Holly Condos: I think logistically, and I, I mentioned it earlier, but logistically, do you want to do that in the context of each sow, so that we can kind of…
255 00:25:56.750 ⇒ 00:26:05.420 Holly Condos: you know, monitor who we’ve set it to, or do you want to just be transparent and, you know, lay one sentence into the MSA?
256 00:26:06.100 ⇒ 00:26:11.670 Uttam Kumaran: I think we just lay it in, and there’s not… there’s not a single… there’s not a single thing we’re gonna do…
257 00:26:11.840 ⇒ 00:26:16.860 Uttam Kumaran: like, I think the amount of things we do outside of AI in this company is…
258 00:26:17.010 ⇒ 00:26:19.050 Uttam Kumaran: Very small these days.
259 00:26:19.050 ⇒ 00:26:19.530 Holly Condos: Okay.
260 00:26:19.530 ⇒ 00:26:28.389 Uttam Kumaran: So, I… and, like, I don’t want… I want there to be, like… I don’t want there to be, oh, we didn’t put this in this one statement of work, and we put in the other.
261 00:26:28.390 ⇒ 00:26:30.199 Holly Condos: Understood, yep, that’s fair.
262 00:26:30.200 ⇒ 00:26:34.019 Uttam Kumaran: Yeah, so I think… I think it’s just… I think we throw it into the MSA.
263 00:26:34.310 ⇒ 00:26:34.740 Holly Condos: Okay.
264 00:26:34.740 ⇒ 00:26:40.459 Uttam Kumaran: I’m… I will… Yeah, I’m not expecting… I don’t know, this may be the first time people have…
265 00:26:40.710 ⇒ 00:26:42.170 Uttam Kumaran: kind of scene…
266 00:26:42.770 ⇒ 00:26:49.100 Uttam Kumaran: this in a contract. I should look at some of the stuff we’ve signed. I haven’t seen it really mentioned in anything I’ve signed for any vendors.
267 00:26:49.100 ⇒ 00:26:57.799 Holly Condos: I mean, like I said, most… most of the… if you’re not Enterprise, it’s… it’s not there yet, right? So…
268 00:26:57.800 ⇒ 00:26:58.170 Uttam Kumaran: Yeah.
269 00:26:58.170 ⇒ 00:27:00.910 Holly Condos: I mean, at the risk of giving you names.
270 00:27:01.320 ⇒ 00:27:12.290 Holly Condos: So, the companies that I’ve dealt with in the last 6 months who want that kind of Gen AI language and approval chain, if you will.
271 00:27:12.460 ⇒ 00:27:15.590 Holly Condos: Our Disney, Sumitomo Bank.
272 00:27:15.590 ⇒ 00:27:15.990 Uttam Kumaran: Yeah.
273 00:27:15.990 ⇒ 00:27:21.529 Holly Condos: PNC, right? They’re big hitters. Bank of America is another one, right?
274 00:27:21.530 ⇒ 00:27:22.010 Uttam Kumaran: Yeah.
275 00:27:22.010 ⇒ 00:27:27.550 Holly Condos: They want… but they want even more than that. They want SOC 2, and they want to look at your logs every.
276 00:27:27.550 ⇒ 00:27:30.570 Uttam Kumaran: But, but, like, when, when, when they,
277 00:27:30.790 ⇒ 00:27:41.379 Uttam Kumaran: like, are… was there… did you find that their worry was what I mentioned, which is, like, they’re worried about stuff going OpenAI? Was their worry about the employees? Like, what was the fear? Do you know?
278 00:27:41.380 ⇒ 00:27:46.390 Holly Condos: The, the, the major, the major theme of the concern is
279 00:27:46.610 ⇒ 00:27:56.040 Holly Condos: twofold. They are worried that their data, their information, is going into the AI ether.
280 00:27:56.480 ⇒ 00:28:10.799 Holly Condos: And might be used by agents, could be, you know, sitting in an LLM that pops into somebody’s query or chat request.
281 00:28:11.140 ⇒ 00:28:11.820 Holly Condos: And…
282 00:28:11.820 ⇒ 00:28:17.639 Uttam Kumaran: But this is, like, You’re saying, but outside of the context of the services that they’re purchasing, like…
283 00:28:18.160 ⇒ 00:28:33.170 Uttam Kumaran: put another way, they’re like, hey, if your employees are using this, and then someone, like, for personal use asks, and this data pulls up, so then, isn’t that the same thing as, like, okay, OpenAI is taking our data and then training, like.
284 00:28:34.030 ⇒ 00:28:37.200 Uttam Kumaran: making it more available? Is that… is that, like…
285 00:28:37.500 ⇒ 00:28:38.060 Holly Condos: Yes?
286 00:28:38.060 ⇒ 00:28:38.660 Uttam Kumaran: Is that accurate?
287 00:28:39.410 ⇒ 00:28:45.770 Holly Condos: Yes, but for these CIOs and their VPs, right?
288 00:28:46.040 ⇒ 00:28:47.480 Holly Condos: and legal.
289 00:28:47.820 ⇒ 00:29:00.950 Holly Condos: legal is saying, you can’t give… you, CIO, can’t give the service provider any of our data… you have to make sure they don’t use any of our data in an LLM. And that’s unrealistic, right?
290 00:29:00.950 ⇒ 00:29:01.360 Uttam Kumaran: Yeah, yeah.
291 00:29:01.360 ⇒ 00:29:02.690 Holly Condos: is unrealistic.
292 00:29:02.870 ⇒ 00:29:08.800 Holly Condos: So, the middle ground is, okay, service provider, we’ll… we’ll…
293 00:29:08.980 ⇒ 00:29:13.319 Holly Condos: will jump on the bandwagon of you using AI tools.
294 00:29:13.590 ⇒ 00:29:24.189 Holly Condos: and putting our data into an LLM, but we want to have some level of control. We want to know what the tool is, what the LLM is, what you’re doing.
295 00:29:24.190 ⇒ 00:29:24.560 Uttam Kumaran: Okay.
296 00:29:24.560 ⇒ 00:29:28.339 Holly Condos: What the output is? What are you using the output for?
297 00:29:28.870 ⇒ 00:29:31.620 Uttam Kumaran: And then what about… what is the mechanism of…
298 00:29:32.000 ⇒ 00:29:37.230 Uttam Kumaran: collecting that? Like, are they… is the service provider putting together, like, a report?
299 00:29:37.390 ⇒ 00:29:40.070 Uttam Kumaran: Yes. Is it like… okay, okay.
300 00:29:40.070 ⇒ 00:29:44.200 Holly Condos: Which is… which is arguably onerous, right? So, I’ve all.
301 00:29:44.200 ⇒ 00:29:51.089 Uttam Kumaran: Well, I mean, I don’t… yeah, I mean, I guess on one side… yeah, I guess also, I’m just like…
302 00:29:51.510 ⇒ 00:30:03.380 Uttam Kumaran: what it doesn’t matter what I put into that report? Like, what are you gonna do? I mean, it’s… I’m gonna say I’m using this LLM, and I’m delivering you this work, and here’s who used it, and here’s how good the work was.
303 00:30:03.380 ⇒ 00:30:09.520 Holly Condos: And, and understand, understand that this, this… Area?
304 00:30:09.760 ⇒ 00:30:13.879 Holly Condos: in the law is very new, right?
305 00:30:13.880 ⇒ 00:30:14.310 Uttam Kumaran: Yes.
306 00:30:14.310 ⇒ 00:30:24.839 Holly Condos: because there’s very little governance and regulation about the use of AI at a federal level, right? There aren’t… there aren’t…
307 00:30:24.940 ⇒ 00:30:42.140 Holly Condos: regulations like HIPAA for AI usage. I mean, that’s just… there are at the EU level, right, the European Union level and some other worldwide levels, but there aren’t from a U.S. perspective. And so, this is really an untouched
308 00:30:42.580 ⇒ 00:30:51.960 Holly Condos: you know, new greenfield area, as far as enforcement goes, and to your point, yeah, what’s gonna happen? You’re gonna give the customer a report, and then what?
309 00:30:52.740 ⇒ 00:30:53.959 Holly Condos: Well, then I guess my.
310 00:30:53.960 ⇒ 00:30:54.639 Uttam Kumaran: The question to you is, like
311 00:30:55.780 ⇒ 00:31:04.569 Uttam Kumaran: Yeah. I guess my question is, like, in the last time, like, a technology like this came out, and pre-having, like, precedent.
312 00:31:05.540 ⇒ 00:31:18.339 Uttam Kumaran: do you know, like, what kind of, like, how people handled it, and, like, did they handle it well? Like, I guess taking the LLM out in particular, like, okay, let’s just say, like, the internet, or, like, phones, or whatever.
313 00:31:18.470 ⇒ 00:31:19.920 Uttam Kumaran: Like, how…
314 00:31:20.400 ⇒ 00:31:26.810 Uttam Kumaran: Do you recall, like, what the… how people structured stuff, and then, like, did they just roll with the punches?
315 00:31:27.320 ⇒ 00:31:29.190 Uttam Kumaran: yeah.
316 00:31:30.110 ⇒ 00:31:34.380 Holly Condos: Well, so, I, I haven’t seen, like.
317 00:31:34.510 ⇒ 00:31:40.080 Holly Condos: Bank of America hasn’t sued anybody yet, right? At least that we know of.
318 00:31:40.270 ⇒ 00:31:46.270 Holly Condos: For this particular issue, but… 2 years ago, there was a lawyer
319 00:31:46.450 ⇒ 00:31:48.140 Holly Condos: I think he… think he was in town.
320 00:31:48.140 ⇒ 00:31:50.810 Uttam Kumaran: Yes, I think I know the case, yeah, where he used.
321 00:31:50.810 ⇒ 00:31:57.900 Holly Condos: Right, so he basically asked Claude to prepare his, closing statement for a case.
322 00:31:58.250 ⇒ 00:32:07.399 Holly Condos: And this was early on, when Claude wasn’t as smart as it is now, and there were a lot of misreferenced cases
323 00:32:07.600 ⇒ 00:32:13.140 Holly Condos: Right? The case law was wrong, and the judge, judge’s clerk was like, this is bogus.
324 00:32:13.860 ⇒ 00:32:20.400 Holly Condos: It might have been a motion, not a closing argument. And that lawyer got disbarred.
325 00:32:21.230 ⇒ 00:32:22.209 Holly Condos: Right? So they…
326 00:32:22.210 ⇒ 00:32:22.620 Uttam Kumaran: Yeah.
327 00:32:22.620 ⇒ 00:32:27.639 Holly Condos: I mean, that’s not a gov… or that’s not a business perspective.
328 00:32:28.710 ⇒ 00:32:35.140 Holly Condos: I don’t… I can’t think of any cases right off the top of my head, because again, I think it’s just too new.
329 00:32:36.480 ⇒ 00:32:41.740 Uttam Kumaran: Like, I guess what… I guess… Yeah, put another way… what, like.
330 00:32:42.090 ⇒ 00:32:44.290 Uttam Kumaran: I guess my question is, like.
331 00:32:44.570 ⇒ 00:32:50.500 Uttam Kumaran: where do you think it’ll arrive? Like, everybody’s gonna be using this for knowledge work.
332 00:32:50.930 ⇒ 00:33:04.829 Uttam Kumaran: I think what I can tell you, though, is there’s gonna be a mix of people using, like, these cons… like, just, like, the random start… like, for example, this… one of the reasons I don’t use, like, random startups…
333 00:33:05.100 ⇒ 00:33:21.809 Uttam Kumaran: around here for work, and we use sort of, like, these bigger names. They’re still innovative. You could call them startups, they all came in the last two years, but they’re sort of, like, super big enterprise-style applications versus, like, randomstartup.io is because of, like.
334 00:33:22.110 ⇒ 00:33:33.880 Uttam Kumaran: those… I know some of those companies will start looking at all the data, and so we try to use those, but I can see this going a couple ways. One is, like, they say, okay, only if you use the major cloud providers, and you turn on these settings.
335 00:33:34.050 ⇒ 00:33:46.599 Uttam Kumaran: The other thing, and this is what we’re betting on, kind of, commercially, is that people are gonna try to run their local language models, right? They’re gonna say, I want to run, like, this… the language model locally on our hardware.
336 00:33:46.770 ⇒ 00:33:52.699 Uttam Kumaran: And I’m also totally fine with that. That is actually the best option, and I actually would…
337 00:33:52.860 ⇒ 00:33:59.859 Uttam Kumaran: like, I think people should… I think those companies should do that. Bank of America should say, we have our own language models in order.
338 00:33:59.860 ⇒ 00:34:00.220 Holly Condos: Yes.
339 00:34:00.220 ⇒ 00:34:15.050 Uttam Kumaran: any of our… in order to, like, use language models to produce work, you should use… you have to use those. The reason they don’t do that is they’re… they’re, like… they’re, like, 5 or 10 years from ever… I don’t think they have the capability
340 00:34:15.510 ⇒ 00:34:17.519 Uttam Kumaran: To, to, like, set that up, but, like…
341 00:34:17.719 ⇒ 00:34:22.750 Uttam Kumaran: that’s where I think the market should head, is ultimately really private, local.
342 00:34:23.139 ⇒ 00:34:25.889 Uttam Kumaran: Language model stuff that sits on your device.
343 00:34:26.449 ⇒ 00:34:26.829 Holly Condos: Yeah.
344 00:34:26.830 ⇒ 00:34:32.610 Uttam Kumaran: But do you… do you think it’ll… in the short term, it’ll arrive at any of those options, or… yeah.
345 00:34:33.610 ⇒ 00:34:42.700 Holly Condos: Well, so I think from… from the legal, you know, ear to the ground, I think that that is the preferred
346 00:34:43.130 ⇒ 00:34:46.490 Holly Condos: Outcome, if you will, is… is that…
347 00:34:46.760 ⇒ 00:34:56.219 Holly Condos: each service provider and each company would have its own LLM, so you treat it like it’s, you know, it’s a new form of IP, if you will.
348 00:34:56.320 ⇒ 00:34:57.130 Holly Condos: Right?
349 00:34:57.670 ⇒ 00:35:06.099 Holly Condos: But… Practically, and again, I think you’re touching on it, what does it take to do that? So…
350 00:35:06.100 ⇒ 00:35:09.250 Uttam Kumaran: No, it’s really, it’s really complicated. Yeah, it’s really complicated.
351 00:35:09.250 ⇒ 00:35:17.920 Holly Condos: So, does Bank of America really have budget, or have they allocated on their roadmap? And they’re probably in 2027 or.
352 00:35:17.920 ⇒ 00:35:20.350 Uttam Kumaran: They should call, they gotta call us soon, hopefully.
353 00:35:20.350 ⇒ 00:35:20.820 Holly Condos: Right?
354 00:35:21.370 ⇒ 00:35:21.860 Holly Condos: Right?
355 00:35:21.860 ⇒ 00:35:22.340 Uttam Kumaran: Like…
356 00:35:22.340 ⇒ 00:35:22.879 Holly Condos: Do they eat?
357 00:35:22.880 ⇒ 00:35:26.380 Uttam Kumaran: No, no, I’m telling you no. It’s not happened yet. Yeah, it’s not happening.
358 00:35:26.380 ⇒ 00:35:31.249 Holly Condos: Right? And honestly, a lot of enterprise are still stuck with…
359 00:35:31.350 ⇒ 00:35:36.610 Holly Condos: what to do. So the conundrum for them is, we bought all this hardware.
360 00:35:36.630 ⇒ 00:35:55.440 Holly Condos: and software, and we have a mix of on-prem and off-prem, and now we’re supposed to, like, jump over… you know, what do we do with all the investment that we’ve already made? How do we… you know, how do we retire this stuff? How do we sunset it without it just, like, looking like we lost everything?
361 00:35:55.620 ⇒ 00:36:02.460 Uttam Kumaran: Yeah, they’re totally jammed. I mean, you want… a lot of them are gonna spend a lot on cloud providers.
362 00:36:03.600 ⇒ 00:36:19.870 Uttam Kumaran: and get very little ROI in return. And then now they’re sort of figuring out, like, okay, we need to build these AI agents and things like that. Similarly, you’re gonna see that, like, okay, if Anthropic is, like, a national security…
363 00:36:20.060 ⇒ 00:36:21.560 Uttam Kumaran: pipeline risk.
364 00:36:21.930 ⇒ 00:36:34.289 Uttam Kumaran: It’s like, what do we do, right? But the local language model stuff is… is quite complicated, I don’t… and it’s not only, like, understanding how to do it, the talent needed to do it is nowhere.
365 00:36:36.090 ⇒ 00:36:40.839 Holly Condos: And it’s not like you’re training an agent in a week to do this kind of work, right?
366 00:36:40.840 ⇒ 00:36:48.450 Uttam Kumaran: No, no, no, this is also, like, it took… yeah, I mean, it’s sort of what I’ve spent 3 years thinking about every day.
367 00:36:48.630 ⇒ 00:36:57.540 Uttam Kumaran: I’m not, like, you know, the best, but it takes a lot of work, like, I… and we’ve had absolutely no politics, no friction, right?
368 00:36:58.020 ⇒ 00:37:12.580 Uttam Kumaran: So, I don’t know. I mean, on one side, I think of it as, like, a huge opportunity for us, but on the other side, I am, like, I want to hedge both sides. If an enterprise or a company says, we want to go completely local.
369 00:37:12.660 ⇒ 00:37:24.120 Uttam Kumaran: Or if they ask me for my recommendation, I want to be able to give them, like, hey, you can run these on your device, it never goes to the cloud, here’s sort of some of the trade-offs, and here’s how we can support that.
370 00:37:24.300 ⇒ 00:37:25.480 Uttam Kumaran: Yeah. Yep.
371 00:37:25.860 ⇒ 00:37:42.439 Uttam Kumaran: And then, if other people are like, hey, the opportunity is now for us to gain market share, while other companies are, like, debating local versus not local, and this is the most… this is the craziest technology to ever exist, I want to be there for… for those people as well, you know?
372 00:37:43.090 ⇒ 00:37:50.389 Holly Condos: Yeah, I hear you. So, I just had a thought, but from the enterprise… again, from the enterprise perspective, typically.
373 00:37:50.870 ⇒ 00:37:55.190 Holly Condos: Do you have to go? What time is that? No, no, no. Do you have Retro today?
374 00:37:55.600 ⇒ 00:37:56.099 Holly Condos: We got.
375 00:37:56.100 ⇒ 00:37:58.529 Uttam Kumaran: Yes. Yeah, we have time, we have time.
376 00:38:00.620 ⇒ 00:38:14.649 Holly Condos: So, what the enterprise company, the IT departments do, either you’ve won the RFP, or you have an existing relationship, and then the IT department decided that they want to do
377 00:38:14.770 ⇒ 00:38:28.599 Holly Condos: you know, some new project. So, that’s why you’re in there. But, having said that, you still have to get qualified as a vendor, so they send you a security questionnaire. Have you seen these?
378 00:38:28.600 ⇒ 00:38:29.070 Uttam Kumaran: No.
379 00:38:29.070 ⇒ 00:38:36.820 Holly Condos: It’s an Excel spreadsheet with 4 or 5 tabs, and basically it’s, you know, it’s how do you protect your data?
380 00:38:36.820 ⇒ 00:38:37.660 Uttam Kumaran: I see, I see.
381 00:38:37.660 ⇒ 00:38:43.679 Holly Condos: lockdown? Do you use SSL? Do you use MFA? Right?
382 00:38:43.680 ⇒ 00:38:45.869 Uttam Kumaran: Yeah, I’ve seen these before, yeah.
383 00:38:45.870 ⇒ 00:38:51.039 Holly Condos: Yeah. Some of… some of them, some of the enterprise want…
384 00:38:51.330 ⇒ 00:39:07.369 Holly Condos: you to have a SOC 2, right? Sure. Are you familiar, right? Or, if you don’t have it, that you have a plan to get it. And so, I’ve got one customer who’s been saying they have a plan to get it for 5 years, but they really
385 00:39:08.300 ⇒ 00:39:11.570 Holly Condos: They’re still trying to convince the CEO that it matters.
386 00:39:12.040 ⇒ 00:39:15.479 Holly Condos: Yeah. I mean, look, that’s what we’re seeing…
387 00:39:15.910 ⇒ 00:39:18.079 Uttam Kumaran: Yeah, I mean, we’re seeing that, like.
388 00:39:18.810 ⇒ 00:39:27.759 Uttam Kumaran: when… the thing is real right now, and the ROI is very real, and what were rules before, like…
389 00:39:28.050 ⇒ 00:39:33.520 Uttam Kumaran: no longer seem like that, you know? Because… like…
390 00:39:33.730 ⇒ 00:39:40.289 Uttam Kumaran: You can really change your entire, like, you know, working culture with these technologies, and…
391 00:39:40.290 ⇒ 00:39:41.020 Holly Condos: Exactly.
392 00:39:41.020 ⇒ 00:39:45.699 Uttam Kumaran: what is the cost of waiting 5 years, or 10 years? Like, I don’t know.
393 00:39:46.310 ⇒ 00:39:47.359 Holly Condos: Well, and I think.
394 00:39:47.360 ⇒ 00:39:47.750 Uttam Kumaran: Yeah.
395 00:39:47.750 ⇒ 00:39:56.000 Holly Condos: Companies that realize what you’re saying, or that adopt and act, Are the ones that…
396 00:39:56.520 ⇒ 00:40:02.420 Holly Condos: will be differentiated. I don’t want to say be successful. I mean, they will be the different.
397 00:40:02.420 ⇒ 00:40:06.690 Uttam Kumaran: Well, they will, they will at least try, they will at least, yeah, they’ll be trying. Yeah.
398 00:40:07.050 ⇒ 00:40:07.470 Holly Condos: Now.
399 00:40:07.470 ⇒ 00:40:09.920 Uttam Kumaran: Other people will be stuck in the mud a little bit, yeah.
400 00:40:09.920 ⇒ 00:40:14.390 Holly Condos: Exactly, exactly. Second point, I think your, your thought about…
401 00:40:15.200 ⇒ 00:40:29.360 Holly Condos: having your own LLM, I mean, is that a business opportunity for Brainforge, or whoever, right? To say, hey, company, you know, here’s all the buzz and the noise today.
402 00:40:29.360 ⇒ 00:40:44.789 Uttam Kumaran: So we want to take advantage of, sort of, both tropes, like, both plotlines. Like, there is a lot of… there are enterprise CIOs and people in technology. I am… I tend to be one of, like… this is where I’m so conflicted, because.
403 00:40:44.790 ⇒ 00:40:45.260 Holly Condos: I’ve.
404 00:40:45.260 ⇒ 00:41:00.230 Uttam Kumaran: I’m very biased, but I see both sides. Like, I do believe that, given the chance, OpenAI, Anthropic, these guys will take your stuff, and there’s nothing you’re gonna be able to do. Like, the leverage is really incredible, there’s nothing you’re gonna be able to do.
405 00:41:00.230 ⇒ 00:41:00.600 Holly Condos: Goodbye!
406 00:41:00.600 ⇒ 00:41:01.450 Uttam Kumaran: So, I am…
407 00:41:01.450 ⇒ 00:41:02.820 Holly Condos: for years.
408 00:41:03.620 ⇒ 00:41:15.630 Uttam Kumaran: Yeah, I agree, I agree. Yeah, I mean, there’s so much precedent that all they do is just look at how things are working, and then they just take your business. The second thing, so I’m actively trying to move Brainforge from…
409 00:41:15.630 ⇒ 00:41:26.649 Uttam Kumaran: using these cloud providers to then using, like, a fixed LLM, and then I will, sometime this year, move us to, like, our own stuff. Like, that’s how I’m… that’s how I’m gonna operate.
410 00:41:26.650 ⇒ 00:41:45.439 Uttam Kumaran: But I’m also… but I’m doing that because I want to kind of explore what that’s like, but then, yes, when we talk to customers, we are playing both sides. Like, there’s some customers who are like, I need the latest, I’ve seen it working, like, we need to use the technology now, just give me whatever you can use. Fine.
411 00:41:45.580 ⇒ 00:41:49.209 Uttam Kumaran: Then there’s some customers who I can tell are very, like.
412 00:41:49.330 ⇒ 00:42:00.439 Uttam Kumaran: whether they’ve read something recently, they’ve always been like this, they’re really, really concerned, and for them, I wanna… I wanna basically say, you’re totally right to be concerned. Here’s our, like, private, local first.
413 00:42:01.100 ⇒ 00:42:06.730 Uttam Kumaran: Language model solution that we want to pitch you on, and that’s what we’re going to… you know, that’s what we’re going to do.
414 00:42:07.970 ⇒ 00:42:08.530 Holly Condos: Yeah.
415 00:42:08.530 ⇒ 00:42:20.720 Uttam Kumaran: And we’re gonna go to market with both. We’re actually starting to… Yeah, we’re trying… we’re actually… this is, like, I mean, in my, in whatever my free time’s… free time I’m working on.
416 00:42:20.720 ⇒ 00:42:21.480 Holly Condos: Oh, no.
417 00:42:21.480 ⇒ 00:42:29.960 Uttam Kumaran: sort of, like, local language model, like, a service line around that, and so it’s kind of some of the stuff that I’ll probably be spending a lot of time talking to people
418 00:42:30.040 ⇒ 00:42:48.379 Uttam Kumaran: about VixelCon is, like, how we’re developing, sort of this, like, enterprise work platform. I mean, you’re… you’ve seen a little bit of a brain forge, but what… right now, it’s just been, like, our edge has been the fact that we have all the information organized, and that we sort of figured out the training, and we figured out getting people
419 00:42:48.380 ⇒ 00:43:00.099 Uttam Kumaran: Like, they can use one platform to do all types of work. We’re thinking about expanding on that, but a big piece of that is gonna be, like, okay, can we also get the language models
420 00:43:00.360 ⇒ 00:43:03.440 Uttam Kumaran: Themselves running locally.
421 00:43:03.840 ⇒ 00:43:04.370 Uttam Kumaran: Right.
422 00:43:04.670 ⇒ 00:43:05.360 Holly Condos: Right.
423 00:43:05.360 ⇒ 00:43:05.780 Uttam Kumaran: Yeah.
424 00:43:06.410 ⇒ 00:43:15.090 Holly Condos: Yeah, and I think… I think it’s all good. So, a couple things there. I will see you at VixelCon, by the way.
425 00:43:15.090 ⇒ 00:43:15.780 Uttam Kumaran: Perfect.
426 00:43:16.190 ⇒ 00:43:20.310 Holly Condos: Yeah, and we can, you know, spend some time talking more about this, but I know…
427 00:43:20.310 ⇒ 00:43:22.959 Uttam Kumaran: Are you just in Austin for that… for that day?
428 00:43:24.010 ⇒ 00:43:26.429 Holly Condos: I’ll be there for a couple days.
429 00:43:27.040 ⇒ 00:43:27.939 Uttam Kumaran: Okay, okay.
430 00:43:27.940 ⇒ 00:43:31.379 Holly Condos: Through… through Tuesday night or Wednesday, I think.
431 00:43:31.990 ⇒ 00:43:37.310 Uttam Kumaran: Okay, okay, great. Yeah, I think we’re gonna see if some people want to get lunch or dinner, but that would be amazing, yeah.
432 00:43:37.710 ⇒ 00:43:43.280 Holly Condos: Okay, yeah, I would love to. But I know that, that they…
433 00:43:43.650 ⇒ 00:43:56.840 Holly Condos: are pushing that idea. That’s part of, you know, what you’re gonna talk about. But I… but I will share with you that I was talking to someone, he… he works for a manufacturer.
434 00:43:57.240 ⇒ 00:44:00.600 Holly Condos: And he’s in the IT department, and so I was just…
435 00:44:00.810 ⇒ 00:44:08.479 Holly Condos: telling him about my business, and I said, yeah, I have… I have two really interesting customers, you being one of them.
436 00:44:09.350 ⇒ 00:44:12.449 Uttam Kumaran: I appreciate it. Interesting in a good way.
437 00:44:12.450 ⇒ 00:44:17.619 Holly Condos: And, you know, I don’t mention names unless it looks like there’s a potential deal opportunity.
438 00:44:17.620 ⇒ 00:44:18.310 Uttam Kumaran: Sure, sure.
439 00:44:18.310 ⇒ 00:44:37.659 Holly Condos: I didn’t… I didn’t say who you were, but, I said it’s really cool because these guys have taken the bull by the horns and gone internally to figure out how to utilize tools to help their employees be really productive for the benefit of the customers.
440 00:44:37.850 ⇒ 00:44:46.570 Holly Condos: And conversely, I have another client who’s a Google Premier partner who can’t get out of his own way with Slack.
441 00:44:46.800 ⇒ 00:45:01.429 Holly Condos: Like, it’s just… they’re so diametrically opposed, and yeah, so, you know, you describe it as you’re just getting it together, but really, in the big scheme, compared to other companies, Utam, I mean, you guys are way ahead.
442 00:45:01.430 ⇒ 00:45:04.800 Uttam Kumaran: I appreciate that, I mean, but I… I see that as, like.
443 00:45:04.800 ⇒ 00:45:05.250 Holly Condos: It’s cool.
444 00:45:05.250 ⇒ 00:45:05.990 Uttam Kumaran: hot.
445 00:45:06.320 ⇒ 00:45:18.219 Uttam Kumaran: Yeah, I see that, like, we have to stay ahead. Like, I have not pushed… I have not released on the gas at all. In fact, I think we are moving faster.
446 00:45:18.300 ⇒ 00:45:30.910 Uttam Kumaran: People are adopting… we are getting people to use it on day one. We are… and then, you know, I just think it’s such a lovely, like, situation we’re in, where we’re selling the thing that we are dogfooding.
447 00:45:30.910 ⇒ 00:45:31.260 Holly Condos: Cleep.
448 00:45:31.340 ⇒ 00:45:34.949 Uttam Kumaran: And it’s, it’s like, could not be a better situation.
449 00:45:35.120 ⇒ 00:45:36.849 Uttam Kumaran: And that was sort of the hope.
450 00:45:36.850 ⇒ 00:45:38.270 Holly Condos: You do it.
451 00:45:38.440 ⇒ 00:45:38.980 Holly Condos: Right?
452 00:45:38.980 ⇒ 00:45:54.460 Uttam Kumaran: Exactly, and it’s all practical, too. Like, I… I like these conversations, but I’m like… I… I… this… this… the legal side has just been something on my mind, and I’m like, okay, it’s… as we’re going bigger, and something changed in the last 6 months, where I’ve been…
453 00:45:54.570 ⇒ 00:46:01.059 Uttam Kumaran: a lot of enterprises are thinking about this, and it’s worth us doing, but yeah, I mean, we are…
454 00:46:01.320 ⇒ 00:46:11.559 Uttam Kumaran: We are trying really hard, and in our… in us finding out that it’s working internally, it makes the customer’s story, us telling the story about us and them.
455 00:46:11.920 ⇒ 00:46:16.149 Uttam Kumaran: Way easier, and then our people are also promoting it, and so…
456 00:46:16.590 ⇒ 00:46:17.170 Holly Condos: Right.
457 00:46:17.170 ⇒ 00:46:27.580 Uttam Kumaran: Yeah, I mean, but it is, again, like, we got… I just think, as a business, we’re lucky because there’s no politics, I’m very forward on it, there’s no, like…
458 00:46:29.420 ⇒ 00:46:36.789 Uttam Kumaran: Yeah, we’re just kind of running up against waking hours in terms of how fast we can go right now, so… Right, right.
459 00:46:36.790 ⇒ 00:46:39.680 Holly Condos: And, I mean, that’s, you know, that’s good and bad, but…
460 00:46:40.230 ⇒ 00:46:40.600 Uttam Kumaran: Yeah.
461 00:46:40.600 ⇒ 00:46:43.060 Holly Condos: tutor credit. Right now, it’s, it’s working.
462 00:46:43.950 ⇒ 00:46:44.280 Uttam Kumaran: Yeah.
463 00:46:44.280 ⇒ 00:46:46.929 Holly Condos: So, to recap, I think…
464 00:46:49.080 ⇒ 00:47:01.640 Holly Condos: Let me, I’ll come up with a couple angles, but I think, to your point, we want to just add a little bit of language to the MSA in a very general, high-level sense.
465 00:47:01.820 ⇒ 00:47:17.620 Holly Condos: And then I think what we need to be prepared for in the back… on the back end is the Ellie situation, where, you know, there’s HIPAA compliance, or… or there’s gonna be something like that, where we need to just be able to deal with it quickly.
466 00:47:17.820 ⇒ 00:47:23.740 Holly Condos: And… maybe, you know, some other…
467 00:47:24.560 ⇒ 00:47:40.480 Holly Condos: maybe the internal LLM language, like, what that would look like, you know, what we do, what we’re responsible for, what the customer’s responsible for, what the liability is, because we don’t want to be liable for
468 00:47:40.670 ⇒ 00:47:47.629 Holly Condos: if… if you convince a customer that you can set up an LLM for them that they use internally, right?
469 00:47:48.720 ⇒ 00:47:58.800 Holly Condos: You basically want to be in and out. You don’t want to be, like, lingering in there and have… if something happened, or if they got, you know, one of their employees did something stupid.
470 00:47:59.400 ⇒ 00:48:02.640 Uttam Kumaran: Yeah, that’s the thing, I don’t want any liability on, like, they used…
471 00:48:02.640 ⇒ 00:48:03.799 Holly Condos: I don’t want Brainforge
472 00:48:04.140 ⇒ 00:48:11.959 Holly Condos: them coming after you for that, so… Yeah. If that makes sense, then I’ll put that together, and we can talk about it next week.
473 00:48:12.490 ⇒ 00:48:21.070 Uttam Kumaran: That makes sense, and then, yeah, for the Ellie situation, really, it’s just gonna be Robert making a call, and like, if the juice is worth a squeeze, on that. And so…
474 00:48:21.070 ⇒ 00:48:29.620 Holly Condos: Well, and that’s why I asked the… Rico, just for your understanding, that’s why I asked about the sows and the value, right? Because a lot of times.
475 00:48:30.050 ⇒ 00:48:37.799 Holly Condos: and I try to help people understand this, contract terms shouldn’t be negotiated in a vacuum. They should be…
476 00:48:38.610 ⇒ 00:48:52.159 Holly Condos: part and parcel, in parallel with the deal, with the value, right? Is this a new partnership? Like, there are components that should be considered. And to your point, is it worth the squeeze?
477 00:48:56.880 ⇒ 00:48:57.550 Holly Condos: Still there?
478 00:48:57.550 ⇒ 00:48:59.760 Uttam Kumaran: Yeah, I’m totally with you, yeah.
479 00:49:00.400 ⇒ 00:49:06.499 Holly Condos: So, well, and in the… in the Ellie context, the other thing is that wherever
480 00:49:06.610 ⇒ 00:49:10.970 Holly Condos: there’s HIPAA information, or the customer thinks that
481 00:49:11.130 ⇒ 00:49:22.990 Holly Condos: HIPAA compliance is important, then you’re going to get the business associate agreement, and everybody who works on the project has to sign it. So even though all your people are
482 00:49:23.380 ⇒ 00:49:26.500 Holly Condos: Well, consultants aren’t your agent, but…
483 00:49:26.780 ⇒ 00:49:37.749 Holly Condos: your employees are your agent, right? Yeah. So, they… they… you can… you can explain to them that this is just a requirement because of the HIPAA flag.
484 00:49:37.750 ⇒ 00:49:43.320 Uttam Kumaran: That’s… that’s also… that’s… I think that’s fine, like, for me, that question is, like, I’m… I’m gonna go to our…
485 00:49:43.630 ⇒ 00:49:45.890 Uttam Kumaran: service lead for data and be like.
486 00:49:46.000 ⇒ 00:49:49.810 Uttam Kumaran: Hey, like, what would it cost you to, like, enforce all these audit things?
487 00:49:50.760 ⇒ 00:49:54.920 Uttam Kumaran: people are gonna sign, whatever, I’m not worried about that, but then, again, that’s what…
488 00:49:55.160 ⇒ 00:50:03.580 Uttam Kumaran: Roberty to say, okay, like, what’s the premium we put on this contract? Is this even worth it taking on, you know? And so…
489 00:50:03.580 ⇒ 00:50:06.059 Holly Condos: Bubble it back up for him, like.
490 00:50:06.060 ⇒ 00:50:06.950 Uttam Kumaran: Yeah, yeah.
491 00:50:06.950 ⇒ 00:50:09.179 Holly Condos: Here’s some things we discussed.
492 00:50:09.570 ⇒ 00:50:13.360 Uttam Kumaran: Yeah, perfect, yeah, and I’ll catch up with him later today, too, I can mention it.
493 00:50:13.910 ⇒ 00:50:34.240 Holly Condos: Okay, that sounds great. Alright, so I’ll put some stuff together, and then, like I said on the pay… on the credit card thing, you should just check if you have that payment module in your subscription with QuickBooks, because that’ll be your easiest. And then, Rico, I think I need to give you, or I can just put it in the MSA about
494 00:50:34.300 ⇒ 00:50:40.409 Holly Condos: Eft, the electronic funds transfer, and credit card payments.
495 00:50:40.640 ⇒ 00:50:43.590 Holly Condos: We’ll just tweak the payment language a little bit there.
496 00:50:44.200 ⇒ 00:50:44.780 Uttam Kumaran: Okay.
497 00:50:44.990 ⇒ 00:50:45.600 Rico Rejoso: Okay.
498 00:50:46.760 ⇒ 00:50:51.400 Holly Condos: Alright guys, well, good to talk to you, and Utam, I’ll be excited to see you soon.
499 00:50:51.570 ⇒ 00:50:54.360 Uttam Kumaran: Yes, definitely, yeah, see you soon, really excited.
500 00:50:55.160 ⇒ 00:50:56.830 Holly Condos: Alright, I’ll talk to you guys later.
501 00:50:57.210 ⇒ 00:50:57.900 Uttam Kumaran: Thank you. See ya.
502 00:50:58.170 ⇒ 00:50:59.110 Holly Condos: Thank you. Bye.