Call —
Transcript batch
0:00 Caller: This call is being routed by Google.
0:00 You: This call will be recorded.
0:08 Caller: The cost of this call is 1 cents per minute.
0:20 Caller: Hello? Yeah, speaking to this. Great, hi, Jess. Yeah, yeah, yeah, yeah. Sorry, I was just in the kitchen. Let me just get to my laptop real quick. Yeah, all right, how are you?
0:20 You: hello gram please i'm dread i just responding to your email two hours ago so now a good time to talk
0:40 Caller: No worries, I'm actually shocked.
0:40 You: Well, forgive me, it's a bit earlier here in New York, but as soon as I.
0:45 Caller: Yeah, you basically just woke up and we're like, okay, let's get this call out the way.
0:49 You: Well, I have other things in process, so I want to try to give Elliot a chance to catch up.
0:56 Caller: Yeah, great.
0:57 Caller: Yeah, so I guess from
1:00 Caller: From my standpoint of things, so just to give you a bit of a heads up, I'm a quantitative research recruitment specialist, primarily by side.
1:00 You: Thank you.
1:10 Caller: Elliott management pretty much at my new firm, we cover a lot of tech, AI research and stuff like that, so they pretty much placed a lot of...
1:20 Caller: Elliot management's team and this is a new role for them and my director pretty
1:20 You: I'm
1:26 Caller: much is swamped with roles for them so he passed this on to me so there might be some
1:31 Caller: technical questions that you're going to ask me that I might not have a direct
1:34 Caller: answer for but I can give you a bit of a lowdown on how I found you and and how I've
1:40 Caller: found your profile. So we have a AI model here called RECOS that pretty much is built
1:40 You: Thank you.
1:46 Caller: behind deplexity. Our director pretty much has built the whole work-to-code model that looks
1:53 Caller: for the right candidates that fit our client's needs. And your profile popped out as a 9 out
1:59 Caller: for 10.
2:00 Caller: for this role so just to give you that Craig yeah Craig have you spoken to
2:00 You: Oh, you mean Craig or someone else?
2:05 You: Okay.
2:07 You: Briefly and then he had something and then that lead died.
2:07 Caller: Craig before okay no worse yeah um
2:13 You: So good to know that the tool Craig built found me again.
2:19 You: Thank you.
2:20 Caller: So yeah, pretty much for this role, you're one of the highest candidates, like highest rated candidates on the market.
2:20 You: We're going to be.
2:26 Caller: What the director, Ben, who pretty much got this role, said for me to ask you, if I can just get that up quickly,
2:36 Caller: your profile is very good for the role.
2:40 Caller: but he had one reservation, which was,
2:40 You: Thank you.
2:45 Caller: we need to understand how you adopt AI, specifically,
2:52 Caller: where is it?
2:56 Caller: If AI adoption directly affected the SDLC that you talk about.
3:00 Caller: in your or wide adoption of AI and that was it he pretty much said you're a perfect
3:00 You: So first of all
3:04 Caller: candidate and it's just that point that he wanted to know more of so yeah whilst
3:10 Caller: answering the kind of next question that that's just going to be one of the focuses
3:14 Caller: that we're going to talk about but other than that yeah great to get you on the
3:18 Caller: phone
3:20 Caller: And I guess, yeah.
3:20 You: what do you know about the STLC?
3:23 Caller: Well, yeah, yeah, yeah.
3:24 Caller: So what do I know about SDLC?
3:26 Caller: Very little.
3:27 Caller: Very, very little.
3:28 You: Okay, so STLC stands for software development lifecycle.
3:32 You: It's a series, it's a process that was formalized within the software development world.
3:32 Caller: Mm-hmm.
3:34 Caller: Okay.
3:37 You: I want to say the 70s or 80s.
3:40 Caller: Mm. Mm.
3:40 You: and iterated on since then.
3:42 You: And the premise of the software development
3:43 Caller: Mm.
3:44 You: lifecycle is how you need to have certain steps
3:47 You: and certain gates between steps
3:49 You: to ensure the quality of software.
3:51 You: Now, a lot of that involves intermediate testing
3:56 You: from step to step, and the reason why
3:58 You: that intermediate testing was important.
4:00 Caller: Uh-huh.
4:00 You: important back then was because the cost of writing code, because it was artisanal, was very expensive. Now that writing code is cheap, what's important today is ensuring that the orchestration of that code from the LLMs is correct, is validated. So a lot of what I do...
4:20 Caller: All right.
4:20 You: is I put all of the testing, all of the validation, all of the checking between the agent and what I call the staging environment.
4:30 You: So the dev environment, where the agent is manipulating the code, is free for the agent to do as they please.
4:38 You: There is no...
4:40 Caller: Thank you.
4:40 You: to production. The staging environment reads data from the production environment but writes data to the staging environment. So that way you can mirror or better guess what the impact on production would be with again no risk. And then the other major change
5:00 Caller: Thank you.
5:00 You: I have in the software development life cycle is I rely very heavily on something called continuous integration,
5:06 You: so the CI part of CICD. And within that continuous integration, I have on the order of hundreds or thousands of tests,
5:15 You: depending on the software being built, that represent...
5:20 Caller: Uh-huh.
5:20 You: not only the happy path, the desirable outcome,
5:22 You: but also all of the edge cases that I've accumulated
5:26 You: over the course of testing.
5:28 You: So a practical example is option trading.
5:32 You: I do trade options,
5:34 You: and I rely very heavily on correctly calculating
5:38 You: things like the wall surface or
5:40 Caller: Um.
5:40 You: the different Greeks. Now, the formulas are established, and I have plenty of examples of
5:45 You: inputs that produce a specific output against which the LLMs can have, can build the necessary
5:53 You: formulas and improvements on those formulas to speed up the calculation. And that...
6:00 Caller: Thank you.
6:00 You: combination of input and output allows me to maintain the quality of the calculation,
6:04 You: verify its accuracy, and most importantly, do it autonomously.
6:09 You: So where in the past, prior to LLMs, I maybe had 8 to 10 hours of productive coding work per day,
6:17 You: I now consistently have 20...
6:20 Caller: Yeah, yeah, that helps. So pretty much in terms of where you place yourself, you'd say there is a
6:20 You: because the LLMs are working overnight within a framework that allows them to self-correct.
6:28 You: Does that all help?
6:40 Caller: clear AI SDLC strategy that you kind of follow because part of it is automated and
6:40 You: It's not code review, to be clear.
6:45 Caller: has been made easier by the AI assisted code review or test generation as you call it.
6:51 Caller: And yeah.
6:53 You: There is a specific mold, for lack of a better term, that the software the AI produces must fit.
7:00 Caller: Yes.
7:00 You: any deviation from that mold is returned as an error which the software must correct well like like what churchill said i trust agents to do the right thing after trying everything else or more or less
7:07 Caller: And it will automatically do that because you pretty much set it up to do that.
7:13 Caller: Thank you.
7:20 Caller: Yeah.
7:20 You: I give them the room to fail and most importantly the room to self-correct.
7:27 Caller: Yeah.
7:28 Caller: All right.
7:29 Caller: So there is practical.
7:31 Caller: Okay.
7:32 Caller: So I guess that pretty much does answer Ben's question.
7:35 Caller: And if you don't mind me phrasing it this way, there is significant practical experience.
7:40 Caller: It's it.
7:40 Caller: integrating AI and other than tuning into like your engineering workflow and this is the specific
7:40 You: The biggest distinction between what I do and what other self-proclaimed LLM engineers do,
7:46 Caller: part of it um so um mm-hmm mm-hmm
7:54 You: for me, the human is on the loop, meaning the human is responsible for the
8:00 Caller: Yeah.
8:00 You: error state or what might be known as the dead letter queue. For others, the human is in the loop.
8:03 Caller: Yeah.
8:05 Caller: Yeah.
8:07 Caller: Okay.
8:07 You: The LLM cannot move without human approval, and I'm of the opinion in the loop is soon becoming
8:12 Caller: All right, perfect.
8:12 You: antiquated versus on the loop. That's an important distinction I can clarify for the hiring
8:16 Caller: Okay, all right, perfect.
8:20 Caller: Yeah, perfect.
8:20 You: if they have their doubts.
8:22 You: Separately, my CV, which you have, includes my GitHub,
8:24 Caller: Mm-hmm.
8:26 Caller: Mm-hmm.
8:26 You: as well as my personal website where I do publish a few notes
8:28 Caller: Mm.
8:29 You: about my stance on certain elements of AI development.
8:32 Caller: Mm.
8:35 Caller: Mm.
8:35 You: And then to further assuage any concerns or
8:36 Caller: Mm.
8:37 Caller: Thank you.
8:40 Caller: Yeah.
8:40 You: risk, I was a senior engineer at Meta before the advent of LLMs, so I'm very capable of
8:46 You: handwriting code that works at planet scale.
8:50 Caller: Perfect.
8:51 Caller: All right.
8:52 Caller: So these are more, so my next set of questions are less technical and just more about you.
8:57 Caller: So as I now know that you're actively on the market.
9:00 Caller: and I guess the question does come down to what does good look like for you.
9:00 You: Do you have an example?
9:03 Caller: Should Elliot be interested, what are the kind of considerations that we need to have
9:07 Caller: when presenting your profile for them to be aware of?
9:15 Caller: So for some people, it's a matter of the reason I'm leaving my shop is
9:20 Caller: I need green space, everything is a bit too slow where I'm at, that kind of stuff.
9:20 You: I see. So there are three things I look understood. There are three things I look for in any role. The first is a place where data is both present and necessary. Instinct and insight are good, but I am of the opinion they must always be backed by data. The second
9:40 Caller: Mm-hmm.
9:40 You: is a place where I have the relative freedom to pursue what I genuinely think is the best solution to a problem.
9:42 Caller: Mm-hmm.
9:47 You: My career is quite diverse, spanning multiple industries, and it's getting longer, and I want to be able to bring forth the skills, knowledge, and experience into the next role.
9:52 Caller: Mm-hmm.
9:54 Caller: Mm-hmm.
9:56 Caller: Thank you.
9:58 You: And the last is a place where the...
10:00 Caller: Yeah. Perfect. So there is a large emphasis on ownership of this role. So specifically, I'll focus on the AI STOC because chances are they'll speak to you about both roles was presented and you might end up.
10:00 You: phrase, that's not my job, doesn't exist. I see that as a sign of bureaucracy and I try to avoid it.
10:20 Caller: having your lay of the land and how you pick it out.
10:20 You: Thank you.
10:23 Caller: So one of the things that they have mentioned to us
10:26 Caller: with the AISDLC role is that this person pretty much
10:29 Caller: will have ownership throughout the tech stack.
10:32 Caller: The idea is that they want somebody who can act
10:36 Caller: as an individual contributor, but at the same time from day one,
10:40 Caller: One, you won't be managing anybody from the day you join.
10:40 You: Thank you.
10:44 Caller: What they want to see is evidence of leading without authority, running pilots, coordinating
10:48 Caller: across teams, owning outcomes beyond their own work.
10:52 Caller: Formal people management experience is a positive signal, but the interview questions
10:57 Caller: very likely to ask is tell me.
11:00 Caller: about a time where you were accountable for a team outcome without direct authority.
11:00 You: Thank you.
11:05 Caller: So I feel like that does fall into the kind of bracket that you're dating in that sense,
11:10 Caller: whereby that's not my job, is not the point.
11:13 Caller: The point is you'd be able to do everything with direct authority,
11:17 Caller: specifically within the realm of AI.
11:20 Caller: ELC. So if that is okay with you, what we can do is, and finally, in terms of your kind of
11:20 You: So.
11:27 Caller: compensation numbers, do you have a direct number that you're looking for? Because we've been
11:30 Caller: told it's uncapped, but at the end of the day, it would probably be a good idea to present
11:34 Caller: them a range that they should be well aware of before either wasting your time or jumping into a
11:36 You: So.
11:40 Caller: without understanding what to essentially yeah so the total comp would be between the
11:40 You: Understood. So the range you shared about the, I think that is a total comp or the bonus?
11:53 Caller: total comp would be between the 400 and 700k the base is just 250 capped pretty much
12:00 Caller: Mm-hmm.
12:00 You: as is, based on the lower end of where I'm interviewing at the moment, the median role is paying the order of 320,000 base, and that's assuming I get them. So it's in the neighborhood, but other places are close.
12:05 Caller: Mm-hmm.
12:07 Caller: Mm-hmm.
12:10 Caller: Yeah.
12:13 Caller: Thank you.
12:20 Caller: Yeah.
12:20 You: as it were unclear how flexible Elliott is going to be in that sense if it makes a
12:26 You: difference of the 510,000 Claude code users that I track I am comfortably
12:34 You: within the top 1% in terms of pro
12:40 Caller: Mm-hmm.
12:40 You: what do you call it, in terms of productivity, and I am defensively within the top 50 worldwide.
12:45 Caller: Yeah.
12:47 You: So I recognize that this is, I recognize there are plenty of people who claim to be experts with AI-powered coding.
12:59 You: I at least...
13:00 Caller: yeah all right um and i guess just to give you a highlight of the process so it would be four
13:00 You: have plenty of evidence to demonstrate it independently.
13:14 Caller: interviews the first one with jim in recruiting who pretty much will just cover a behavioral
13:20 Caller: sense. The second one would be with their lead. One of the AI leads don't actually know his name.
13:20 You: We'll find that soon enough.
13:29 Caller: Sorry about that. I was just given this rule last morning. The third is head of DevOps. Yeah.
13:31 You: My biggest concern...
13:35 You: Oh, DevOps.
13:36 You: Okay, happy to nerd out with them on this.
13:40 Caller: Yeah, and then the final, which would be one of the senior development development managers,
13:40 You: I'm
13:47 Caller: an independent read on an organizational influence, how do they come across to someone who will be receiving the end of their recommendations and stuff like that.
13:55 Caller: DevOps is pretty much, this is the president you'll be working closely with from day one.
14:00 Caller: And then the second round would be just a gather of your experience pretty much a similar conversation on building a dedicated AISOC function that makes every engineer more productive, stuff like that.
14:00 You: We're going to be.
14:12 Caller: So that's a clear outline of the process. If you're happy with all of that, what I'll do is I'll send over a rise up to Ben.
14:20 Caller: posting what we've talked about and your points on and from there Ben
14:20 You: excellent uh one caveat i have other processes in flight some of them are now at the outsider super day
14:25 Caller: we'll pretty much make an intro and we'll find out within the day or two
14:29 Caller: whether they'd be interested in having you along for a first shot right
14:33 Caller: mm-hmm yeah
14:40 Caller: Yeah. So, it's four rounds, which is typically easier for me to understand than the normal trade and quite research around. So I think if the availability matches up.
14:40 You: how quickly can elliot run through their interview process i anticipate this is functionally a run
15:00 Caller: you can pretty much get this done fairly quickly once we do let them know that you have other
15:00 You: is there any chance all four is there any chance all four rounds can be booked up front and then if i don't make it past nth round the remaining get cancelled splendid all right
15:20 Caller: and I'll let you know as soon as possible.
15:20 You: Anything to get this process completed promptly.
15:22 Caller: Thanks for calling me so early in the morning.
15:24 Caller: I appreciate that.
15:27 You: Thank you very much, Graham.
15:30 Caller: Perfect. Have a good one.
15:31 You: You too.
15:31 Caller: Cut up soon.