⚡️ 10x AI Engineers with $1m Salaries — Alex Lieberman & Arman Hezarkhani, Tenex

Okay, we're here in the remote studio with Alex Lieberman and Arman... Oh my God, I did not prep this.

Leave it in.

How do I, how do I-

Leave it in.

How do I say it? Uh, just say Arman.

Keep it rolling, Swenx. Keep it rolling.

If it makes you feel bad, for the first probably 20 times that I said Arman's name, I said it the wrong way, and he was very polite in guiding me to the right pronunciation. He used to say Armen, not Arman, so it's okay.

It's totally fine.

Point the finger at me too.

I don't think you have to... I mean, it's Hezarkhani, but we don't need to-

Hezarkhani.

Yeah. Yeah, yeah. Arman Hezarkhani is fine. Um-

Amazing.

Yeah, that's fine.

Honestly, it'd be even funnier now where as you're about to introduce him, you just dub Arman's saying his own name over your mouth.

Introducing-

So guys, play it

... Arman Hezarkhani.

Play it.

That's so funny.

It's like when you're, when you're, like, on a voicemail and you're, like, saying your name while, like, the automated machine thing.

Totally. Totally.

So you guys are the co-founders of Tenex and also, uh, MCs and speakers at AI, AIE, right? So I, I mean, uh, and I, I think for me, I have a little bit of extra context on Alex because I followed Morning Brew for a while.

You, you have been an inspiration on the newsletter business. But let's, let's talk about Tenex. You know, like, I, I think that I... My, my goal here is just to introduce people to you guys, maybe you individually and then you together.

So whoever wants to take it first.

Well, I can give you a little bit of the backstory, uh, behind the business and how Arman and I got to know each other, and then Arman I'm sure will fill in some gaps. You know, Arman and I met in 2020, uh, when I had invested in his previous business, uh, Parthian.

And Parthian, uh, was a AI financial tools business originally for consumers being AI tooling for financial advisors or RI- RIAs. And, um, you know, throughout Arman building that business, we had continued to talk about just our philosophy on product, how AI was influencing just product in general.

And I kind of think, especially for non-technical folks like myself, there's, like, a moment where you get smacked in the face by how profound this technology can be if harnessed in the right way, and I experienced that moment in conversation with Arman.

So it was probably this point nine months, nine, nine-ish months ago. Arman and I were talking, and he, he had shared a story about with Parthian, he unfortunately had to downsize his engineering org. And when he had downsized his engineering org, he had to decrease the size of his engineering team by 90%.

And when he did so, he had to rebuild. He had to basically re-architect the entire product and engineering process to be AI first because he just no longer had a human resource, and so he needed to, like, accelerate it with this technology.

And basically, what Arman had shared with me is that output of production-ready software had text after making this shift with the org. And I, I kinda didn't believe him at first 'cause I, I had never seen kind of that level of leverage.

Like, I'd use ChatGPT, I'd use Groq, I'd use all of these things. But... And yes, they've been life-changing for me, but I wouldn't, I wouldn't have explained them as Tenex experiences. So we basically talked through it, and he kind of shared with me why AI and specifically LLMs have made such a profound impact on engineering as a type of knowledge work.

And from there, the, the thought was around how the way in which engineers are compensated has to change materially. Because if you think about it, like, historically, people charge for their time by hour, and then all of a sudden, let's just say you're a new...

like, you're truly an AI engineer who's truly Tenex higher throughput. Imagine you're selling your work and someone's used to pa- spending $100 an hour for an engineer, and you go to them and you look them dead in the eyes and you're like, "Yeah, I'm 1,000 bucks an hour."

You're gonna get laughed out of the room even though you're, you're a better engineer than the engineer they would have hired. And also you're perversely incentivized because you leveraging AI in your work has you operating faster, but your incentive, just like a lawyer or just like any hour, uh, pay-based knowledge worker, is to rack up as many hours as possible.

And so, like, actually the, the kernel of insight that started all of Tenex was, how do we hire the best engineers in the world? How do we offer them unlimited upside by compensating them for output rather than hours?

And then how do we harness that in the right direction to help companies transform with AI in their business? So I know there's a lot there, but Arman, is there anything I missed?

I mean, basically, yeah. Like, like I think Alex covered it. Like, like I was writing code, and I was deeply incentivized to generate more output, high-quality output, but faster, more, right? And it's because it was my company. But the whole thought is, like, when you work at someone else's company, even if you have some equity, even if you deeply care about the mission, you're not deeply incentivized day in and day out to try new AI tools and push yourself to work better and faster and smarter.

And so the economic model behind our company is one that does drive that, and my talk is basically to show how we do that and how I think other companies might be able to adopt similar models.

Well, this is very tempting because every question I'm gonna ask might actually just leak your talk.

It, it's okay. The talk will, the, the talk will just reiterate very important points.

Well, it should be, I mean, it should stand on its own on YouTube, right? So it's whatever. Let people... Uh, I do like to encourage people to remix the content in different formats, so this is the podcast version.

Totally. Totally.

So I think, like, I think the, the classic thing is, well, what is a unit of output of a software engineer? Is it a PR? Is it a story point? It's extremely unclear, and it's very basically unsolved. Like, I mean, don't tell me you've solved it.

You know, like, what, what's... Maybe you have, I don't know. But, like, I'm, I'm default skeptical on the- Well, what gets measured gets gained.

Yeah, we do use story points, but you're right that it's, it's easy to game it, right? Like, like if we were to hire somebody who just... Like, if you think about a technical system, right, a smart hacker will find ways to exploit it, and the easy way to exploit the story point system is to deflate the concept of a story point and decide that, okay, any line of code-- Like, lines of code are gonna be dir-directly proportional and equal to story points.

Well, then, of course, you've hacked the system, right? But your clients will churn, and you'll probably get let go of, a-and it just won't work long term. And so what we found is that hiring two-- What we found is that this problem gets solved in the hiring process, and it gets solved by hiring people who fall into two buckets.

One is people who are selfish, but they're long-term selfish. Everybody's selfish, but we need to look for people who are long-term selfish, people who understand that these incentives are longer than just today's story points. They're forever, right? And we need to think about, how do we maintain the client relationship?

And that means that we're gonna give them very robust story points so that we can maintain the relationship and, and continue to make money. But the other is that we hire people who just like writing code and like working with really smart people and, and they're not sharp elbowed, and they just want to do great work.

And, and that sounds squishy, but that really is a part of it as well. I think both are, both are really important.

Just two other things I would quickly add is, one, when we work with clients at Tenex, there's basically two role players. There's the AI engineer, and then there's the technical strategist, and, and one of the best ways to fight perverse incentives is to incentivize two people at odds with each other in a healthy way.

And so our technical strategists are incentivized based on NRR. Are, are incentivized based on retention and account growth for a client, and they are the final one to sign off on the engineering plan for a client before we begin a sprint.

So, like, they are the last line of defense of quality before a client ever sees anything. So, so that's one thought. The interesting thing, and I don't know if Arman has thoughts on why this is, is we have not yet, and again, we're a young company, so this could change at some point, but we have not yet had any clients argue about how we assign story points or ever feel like we are sandbagging story points or any of these things, which is just interesting because I think to your point, Swix, like, I would have expected that to have already happened.

Yeah. It can be a political process when things go not well. But when things go well, no one, you know, everyone's just, like, steaming ahead. Okay, you hire great people. You, you work well with story points. I think one thing I'm trying to get my guests to do a, do a better job of is just brag.

Could you brag a bit? Like, just, like, some really impressive project that you accomplished just, just opens people's minds. Like, let's get, let's get specific without maybe naming the exact client, unless you can. Um, and then also, like, what's the highest hourly rate that one of your engineers has made since you're technically uncapped?

Yeah. So I'll answer the last one, uh, or the second one first. We will probably have more than one engineer make a million dollars cash next year based on this model, and that is just with story point compensation.

It's very likely that we will have more than a handful of folks make more than a million dollars next year. Um, the answer to the first question, like, for example, one project we built-- So we work with this company that's, uh, they, they build-- They work-- They partner with retailers to basically make cameras in the business more valuable.

And the way that they do that is they deploy what was historically like a gen four Raspberry Pi to the stores, and they would, they would run, like, one model on that device. We basically took some off-the-shelf models and trained some models ourselves and then quantized them down so they could actually run on that, um, on that four, but also on Jets and Nanos, and we got them to all run in parallel.

So now basically, what these models allow you to do is, as a store, you can get a heat map. You can see where the lines and the queues are forming in your store. You can even get pictures of shelves and understand what needs to be stocked, and you can do things like theft detection because we have body analysis, and we can understand things like b- things are crossing arms, right?

And this took our team two weeks to put together early prototypes, and now we're just refining accuracy and improving metrics from there. Um, and again, this was, like, one of the, the many examples. I think, of course, with that specific example, that's more of a research project, and it's going to take a while to improve accuracy and things like that.

Like, we're not claiming that we're, like, these magical beings. But previously, that alone, building a prototype of that would take several quarters for robust teams of engineers together, and we were able to prototype that out very quickly, and now we're, we're working together with that team for, for a year to build more and all that stuff.

Alex, anything? I mean, I guess another one, Snapback Sports. We built them a mobile app in a month that hit twentieth on the App Store globally, and there was no AI in this app. It was a really fun trivia app, but we built it together, deployed it, hit twentieth in the, in the world.

Yeah. I mean, one other example I would just add is, and this is just looking at things from a different angle, which is sales. I think the power of AI engineering and fast prototyping is incredibly powerful within sales motions now.

And so just one example is we had a big influencer who wanted to basically build basically ChatGPT, but specifically as if it is your fitness, like your health and your, your health coach and your nutritionist. So it has all this context-

Is a fitness influencer.

Yeah, exactly. And we originally reached out to work with him, and basically, he said no because he was like, "You guys are, like, too early. You don't have your, like, a design team built in yet." And so he said no, and it seemed like the conversation was done.

One of our engineers was like, "I'm just gonna build a working version of this app as soon as humanly possible." So basically within, I don't know, it probably took him four hours. He got-- He had just, like, a working version of the app that was in the hands of this influencer And that influencer hasn't launched the app yet, but we are number one on their list right now to actually do this build, and the only reason is, is the speed by which, like, working product could be in hands of someone is faster than it's ever been.

Yeah. That's amazing. Okay, so, uh, uh, the quick question on just the, uh, the stack that you guys have landed on. Like, is there a house stack? What are you guys finding in terms of, like, the various coding agents and, and all that?

Yeah. Um, we do work in a number of different stacks, a number of different languages and stuff, but we feel pretty strongly in, like, high structure allows for agents to work autonomously for longer. And so our default stack is TypeScript front end, TypeScript back end with a shared file where...

or a shared folder where all of our shared types and schemas and things like that live, and typically React front end or even something as simple as, like, Express on the back end. Like, we don't really care about the frameworks.

It's more just like TypeScript allows us to have that flexibility to, um, like the flexibility of JavaScript, but the, but the constraints of TypeScript. And then those error messages allow the Claude code or cursor agents or whatever to iterate on themselves and, and run things, see the errors, and, and continue.

In terms of the actual, like, AI engineering stack and what coding agents and things like that we're using, I always tell clients this, like, our team doesn't have a favorite coding agent of the year or of the month or even of the week.

Like, if I go over there to our team right now and I ask them what model is performing the best for coding right now, they'll say, "Today at four forty-two, we're noticing that Claude Code is actually performing better because of X, Y, Z reason.

But yesterday Codex was outperforming Claude Code on object, on, on activities like X, Y, and Z." Right? And, and we stay really, really deeply on top of all the different models, all the different applications, uh, of these agents, um, to make sure that we're getting...

they're really pushing the most out of this and so that we can advise teams on how they should best use these things.

I mean, well, so yeah, but they're, you're gonna... It's very anecdotal, right? Like, don't you need more comprehensive evals? Because otherwise it's like you are just behaving or believing things based on the luck of the draw.

I think at this state, did a samurai have a measurably better sword than the person to their left or right? No. Right? At, at a certain point, I think a, a, a, a warrior's weapon becomes something of a feel.

And I think that at this point, a lot of these, like, the coding agents are so good. Like, like, yes, you can have evals that, that provably show that one is better than the other, but for a lot of these things, it really is feel.

It's like, hey, this agent actually, like, it just, I can work better with it on a warm-blooded level, or it, it writes code more like I like to or whatever and, uh, at least that's what we've noticed.

Yeah, fair enough. And so I think, like, there's this-- You're, you have, like, kind of a SWAT team approach. You're paying... You're, you're very meritoc- meritocratic, I think is, is the, probably the right, right term in this. Are you human bound or are you agent bound?

Like, what is your limiting factor in Tenex becoming a bigger business than, you know, either of you have run before?

Today, it's human bound one hundred percent. Uh, that is-

You're, you're recruiting? Yeah.

Yeah. Yeah. We, we are... The thing that keeps us up at night is how can we hire enough good engineers fast enough? Um, and then the second thing that keeps us up is how do we match those, the great people within the business with the right process such that delivery doesn't suffer as we scale.

And I think more and more as we build this business, like technology is gonna be en- an enabler of the work we do, and I think long term, if, if we're to talk about the long term of the business, there are ambitions of this business beyond just acting as a transformation and engineering partner for companies.

There, there are ambitions to build our own technology. But today, and probably for the foreseeable future, we are human capital constrained.

How do you interview? You don't have to, like, give the exact interview questions, but, like, has interviewing changed for either of you guys pre-AI versus post?

Yeah. This is actually somewhat controversial. A lot of my friends stopped doing take-home interviews after AI. We still do take homes, but our take homes are immensely... They're like, our take homes are unreasonably difficult. And so when I, when I first wrote them up, I told Alex, I was like: "Hey, man, like, like, people might get mad at you, you know.

Like, you have a public persona. Like, we're sending these to people. Like, your reputation might take a hit if we send these to people because they are so unreasonable for us to ask this of people." And Alex, in, in classic Alex fashion, was like, "F it.

Let's just do it. You know? Like let's send it. If this is the bar, then-

Bring it. Yeah

... then we need to do it." Yeah, exactly. And what we found is that fifty percent of the people don't even respond to the take home interview. But because our take home is so difficult, our interview process is actually quite short.

We do two calls before the take home, then we send the take home, then we review the take home, and then if it goes well, we do maybe one or two meetings afterwards. So it can be done in the fastest in a week.

It's very, very quick if, if people can get through that take home.

Yeah. And just a few things to add. Like, I'm thinking about what are some of the, the most common questions we ask. A few that Arman asks that I really like are, one, he basically says, like: If you had infinite resources to build a AI senior e- software engineer, like truly one that could replace either of you on this call right now, what would be the first major bottleneck that you would have to, uh, figure out how to overcome to build that?

That, that's one question that he always asks. And Arman, out of, out of curiosity, I don't know if you wanna share it because then people will start giving the right answer on that, but is it... Oh, I guess just for Swix, like-

Yeah, I can-

Is there-

I can off-

Yeah.

I can offer one. I don't know.

Yeah, yeah, let's hear it.

I mean, so the, the, the classic answer is just model intelligence, right? Like, we think the models are good, but, like, actually, they have been really trained into a certain sort of local minima of, like, well, here's all the Python because Suitebench is all Python, all Django.

Um, and actually beyond that, we've maybe generalized like a little bit of front end, but hasn't really done like full back-end distributed s- services and all that. So model intelligence is gonna be like the main blocker. But I don't know if that's a good answer because it's kind of like, well, you just wait and then maybe the f- the Frontier Labs will solve it.

Yeah.

Arman?

I, I generally think that it has to do with context. I think it's not necessarily context length. I think that it's context engineering, in Andrej Karpathy's words, right? It's, it's the problem of how do you, how do you get the right context into, into the mo- like into the LLM and get the LLM to pay attention to the right parts of that context, all of which I would consider context engineering.

And then from there it's like, okay, there's a lot of ways that you could solve that, right? You can, on the model layer, do a lot of work to make sure that the ec- attention mechanisms are paying attention to the right stuff.

You could do work on the application layer to, to context engineering. You can extend context lengths. Like there's a lot of different work, and then it leads to a really interesting discussion. So, so yeah, that's, that's one thing.

One thing I was just gonna say is I feel like Dan, who's one of our engineers at 10X, he shared a different answer that I remember your reaction to. It was like it, it broke your brain a little bit.

Do you remember what his answer was?

I don't remember. No, I should ask him.

But I believe it had to do with entropy.

I should ask him what it was. There he is. Dan, come here. What was the question- ... that we asked you during the interview? Remember? You b- like I asked you if you had unlimited resources and you needed to build an AI engineer, what would you need to solve?

You gave me an answer that like blew my mind.

It would be like, what would be the, um... What would be like the limiting factor, the hardest thing?

Yes. Yeah. What was it? What'd you say?

I said it was like controlling entropy.

Oh, there it is. Yeah, controlling entropy. Controlling entropy from what?

Sw- Swyx does not-

Like tell me more

... does not agree with Dan.

You... If there is some error rate in... Y- your question was basically like-

Wait, come closer so they can hear you. Come closer to my headphones. This is Swyx. You're on a podcast.

Yeah, we can hear you.

Yeah.

We can hear you. It's good. It's good. We're rolling with it. Um, so basically like if there is some... Your question was about a fully autonomous like coding loop.

Yeah.

So like what would it take to get the human out of the loop? If in that loop you have some error rate, um, let's say it's 99% accurate with code, even that 1% error rate will just multiply and decay more and more, and that entropy will build and cr- like accumulate, and that's like kind of a compounding thing that will derail the agent more and more.

And so I think it's less of like a context engineering question per thing you're implementing and it's like, um, making sure that the agent can reduce the entropy for a given task such that it gets to 100% accuracy.

Um, and then you don't have this like accumulating error issue.

Cool, man. Thanks, brother.

No, that's, that's impressive. Oh, no, actually, so-

Yeah

... that's like, that's, that's the sophisticated version of context engineering, right? Like a lot of people are gonna answer context engineering.

Exactly.

We have one of the people that coined context engineering coming to speak, uh, I think in one of, I think one of the early sessions on, on Friday. And yeah, like but this is the actually, like the, the advanced like...

Yes, this is one of the four ways in which long context fail, and if you have enough experience, you know that this is the one that gets a lot of agents off track. And once they're off track-

Yeah

... it's really hard to get them back on track.

Exactly.

Exactly.

And going back to your question-

I wouldn't use the, the same words, but yeah. It's, yeah, I get it.

Yeah.

And going back to, going back to your question about constraints in the business, it's just how do we find more people like him is the thing that keeps us up at night.

Well, you know, I'm in the business of making more. You're helping to contribute by putting this conference together where we're just sharing knowledge, and the more people that watch, like are kind of like drawn to you, they, they might answer your call to action of like trying out one of your super hard tests or any, any just learning and just in advancing the state of the industry.

For sure.

So I'm excited to, to have you guys. Uh, do you have any questions for me? I mean, you know, it's like a whole like two, three-day affair. Uh, you know, I've, I've, I've done this a little bit now.

One question I have for you is like as Arman knows that I'm voraciously curious and I'm a lifelong lear- learner, but I'm also not an engineer by training, and my goal is to get as smart about this space as quickly as possible.

And so like, you know, I, I... One of the first things I did is... Was it Arman, did you send me the three blue, one brown lecture? Like you know the three w- blue, one brown-

Yeah

... like does the lecture on LM- LLMs. I took that. Then he's like, "If you wanna go super deep, do any of Andrej Karpathy's..." Like he does like the lecture series on how ChatGPT works, and he's like, "Actually like write out notes by hand and like you truly understand like the math behind these models."

And Arman did that and he was like, "It's just like that's how you understand things at the deepest level." So when, when I'm not either working or taking care of a four-month-old at home, that is next on the list.

But I guess like my question for you is like as a non-technical person who's always been like both enamored and intimidated by technical folks, what would you do if you were me to make the most of this conference where when I'm not like the core archetype of the person who's there?

Geez. Yeah, that's, that's a tough one 'cause I, I spend zero time thinking about that. Okay, so, so I think like latch onto the keywords and whatever people are excited about, like context engineering people were excited about maybe four or five months ago and now it's like entering the mainstream.

Typically the, the people at this kind of conference would be sort of stewing around those ideas. Like MC- Uh, the last time we were here in New York, MCP was kind of just taking off and we did the workshop and it, that really blew up, uh, MCP.

And I think like that is something that you will see a little bit of, like the, the, just like think like-

By the way, Arman, Arman grimmed, grins because he has very strong feelings about MCP.

Mm-hmm.

Very strong feelings.

Pro or anti? We're h- we're hosting a debate. He-

I just think that MCP is a three-letter word for API and like-

Yeah

... Alex always, every time he hears someone say the word, the, the letters MCP in, uh, in that order, he tells them that I hate MCP and starts a, a, a war, a religious debate.

No.

So Well, I will say though, I do think a few of our engineers have warmed you up to it more with specific use cases, Arman. Yeah, I mean, uh, like, like are MCPs useful? Like, of course. I use all the MCPs with Cloud Code.

I just think that there's... Like, what, what bothers me is when people create a new name for something and then use that to raise some inordinate amount of money because they know that three-letter acronyms get investors excited. That's, like, the thing that, like...

That's, that's why I giggle when I hear MCP, 'cause I'm like, a lot of people just say that.

Yeah.

And, like, the tweets that bother me are like, "MCP is coming for your job. Here's why you need to know about MCP." And it's like-

Yeah, it's horrible

... no, it's just like a useful thing , you know?

Maybe, maybe this is relevant to Alex's question. Uh, I do take a sociological and anthropological stance to tech in terms of, like, different groups of people coming in have different terminology to communicate with each other, and it's, it's just human behavior.

It's like I'm kind of non-judgmental about it. Like, people just gotta do what people do, and they, they always invent new language and they all... Like, there are only so many ideas doing, going around in the world. They're, they're gonna be recycled.

Yeah. Totally.

That said, like, I will defend MCP in a sense that, like, there actually are other parts of the spec that are not just API wrappers, but people just comparatively don't use them as much. But I, I think it's a little unfair to MCP, uh, the whole protocol.

But that's why we have a debate, where ac- we actually have, like, a podcast- ... booth and we're actually hosting, like, a, you know, pro and con debater and I, I think it's really fun.

Wow.

Yeah. Yeah, yeah.

That's awesome.

So I f- I actually really wanna get into this, 'cause I think I, we learn more by contrast than by agreement, right? Like, so in a, in a single talk, like, you know, you're the authority, you're out, you're up there on stage, you, you say whatever you, you wanna say and no, no one can really...

Like, people just fight in the comments, but they're never gonna rise to the same level. I think in a, in a real debate, you, you can learn from both sides and make up your own mind, and I think that's, uh, what we're gonna see.

That's awesome.

What we're trying for. Yeah, yeah.

I love that.

Well, uh, it's great to meet you guys. I'm looking forward to your talk, Arman. And Alex, you're, you're, you're opening the show for us, so-

Can't

... all power to you. I, I, I do think, like, I pers- I intentionally left that block that you're, Arman's in as the consulting block. Uh, we also have McKinsey speaking, but McKinsey's not in the consulting block. So I'm very curious because I think my theory is that a lot of our attendees will, will be from the enterprises that, like, might be looking to talk to you guys.

And I'm curious to, like, see how this sector grows. It's not something I'm personally that familiar with because I mostly just work in companies as, as an engineer, but, like, the sort of consulting digital transformation industry is kinda new, but it's also, like, very, very in demand, as you guys know very well.

And I'm just, like, excited to feature it for the first time.

Yeah, we're super excited to be there, and thank you for having us and pumped to learn a ton from, from you and from the, uh, the other speakers there and just the people who are attending.

Yeah. Yeah, yeah. I mean, like, everyone from the, the labs to, uh, the Fortune 500. It'll be, it'll be a whole party. All right, thank you.

Love it. Thanks, man.

Thanks.