Gauntlet AI's Austen Allred: junior engineering is 'Armageddon' and the AI coding market will fundamentally restructure software teams

next week. Uh for now, we have Austin Allred uh coming in to the studio. How you doing, Austin? Good to hear. Hey, good. How are you guys doing? What's going on? How are you? Yeah. Yeah. Hanging out. Well, yeah. I wanted to have you on the show for a few reasons.

Obviously, there's a lot of stuff happening with um AI coding agents. What's what's this whole AI thing? Yeah. What's this whole AI thing? Whole AI thing. Haven't heard about it.

Um yeah, I mean maybe start with like introduction on yourself and and and and what uh what you're up to now, what you've done in the past, and then we can kind of go into uh the structure of uh of AI in the workplace, how people are learning to use AI, and then kind of some of the disruption that's potentially happening with AI tools actually replacing jobs.

Yeah, for sure. Um, so, uh, a little bit of background, co-founder of Bloomtech, which used to be known as Lambda School before a trademark lawsuit. Um, it's technically still the same company.

Um, but about 18 months ago, we started talking to people about this new fangled AI thing and, you know, if it had any impact within engineering teams and frankly, I was pretty skeptical in the early days. Uh, but we sent out a research team. They looked at what everybody was doing.

And at the time it's not quite that way anymore, but there were a handful of people who had, you know, sat in the basement playing with LLMs, seeing if they could make them write real code for a couple of years. And when we saw that, we were like, "Oh, wow. This is actually working. It's not mainstream at all yet.

" And at the time, even less so. Um, and then working with a couple of those companies that we were, so we started training engineers to, you know, use AI really well. Um, and it's one of those things where there are no best practices. Everybody's learning more every day.

The game completely changes every day, let alone every week.

Um, and then one of the companies we were working with brought us in to do a, you know, fully intensive 100h hour a week, find all the people with 98th percentile IQ and above, fly them into Austin, train them over the course of 10 weeks to be expert at using AI, um, and kind of staying on that that cutting edge, riding that wave up.

Um, and then they hire them on the other side and pay us to do that. Now we're expanding that to more companies.

But I basically spent the past several months holed up in a little office with a bunch of sweaty people trying to figure out how to max out AI and figure out how to stay on the the cutting edge of using AI to build software.

Is the shape of that more like PM who's now able to write code or instantiate their ideas in code or is it software engineer who understands some basic programming that can then write more functional code or just be more performant like which which angle is it both or is one is one narrative driving more of like AI adoption?

Yeah, I'd say um it's more the latter. So the more experience you have as an engineer, the more it sucks to start using AI because you know you're going from here to here as far as codew writing ability goes.

So in the early days of Gauntlet, we force everybody, you know, and the more senior you are, the more you hate it. We say you can't write code manually. We put software on everybody's computer that, you know, watches them and AI has to do all of the code writing for you.

And everybody thinks that that's the dumbest thing they've ever heard. And they fight against you for a week, week and a half. And then, you know, the more senior you are, the longer it takes for you to actually get to parody. Um, but it's you, you'll never go back once you get there.

Um, it will, it has fundamentally changed the software industry forever in ways that people don't fully appreciate yet. entire companies are going companies are going to have to rethink the way they do everything. Um, and we're only seeing the very very very early stages of that, right?

Do you guys have your own kind of messaging internally with the students around uh the framing of like vibe coding because it maybe is it's like this cool viral phrase people. Yeah. But it's kind of different but it's it's it's also different if you're doing it in this hyperintentional way.

trying to effectively get to the same quality and consistency as regular software engineering but just doing it in a super AI native way. Yeah, I feel like that's muddied the waters quite a bit.

I mean the when I say vibe coding, most people imagine somebody who doesn't understand software kind of blindly trying to oneshot applications. Uh which it's incredible that that works at all.

Um but when we think of you know whether you call it vibe coding or some people call started calling it superb building internally I don't know if that ever um made it out but using AI to write performant enterprisegrade bulletproof software um there's not a great distinction for that out in the market but we do view it very very differently.

There's a time and a place for both. you know, if you're writing your own personal software or little internal tools like yeah, just blindly telling an LLM to do something might work.

But most of what our engineers are doing would work in, you know, very enterprise, very legacy, very big, large scale applications, you know, security is taken care of, the the formatting's done correctly. It it's it's scalable and secure and all the things that software needs to be.

What are you seeing uh the most success with in terms of tools? There's so many different options. Every week there's a new tool, right? You know, maybe they were using Devon at some point, then they're using Claude and then they're using uh now Codeex. What what what's your kind of how how are people prioritizing?

Are they using things like Replet? You know, what does that actually what does the stack look like? Yeah, it it's not an exaggeration to say that it changes every week in a very fundamental way.

Um, so you know, I remember when Claude 37 came out, before Claude 37 came out, pretty much everybody was using Claude 35 as their daily driver. Um, and I feel like Claude 37 was the first kind of breaking point where that became no longer true anymore. Um, so before Claude had basically a monopoly.

If you were to walk around the gauntlet offices, um, and ask everybody what they were using, 95% was Claude 3. 5. Um, with 37, it worked better for some people's workflows and worse for other people's workflows. Um, Grock is gaining ground. Um, it doesn't, you know, most people don't have API access to it yet.

So that's a, uh, you know, you have to use repo prompt or something to get it into your IDE. So that's a you know separate thing. Um I would say on average if for people who are working on really big really enterprise applications cloud 35 still seems to be winning out even over cloud 37.

Um and Gemini I think kind of changed the game. Um when was that? Was that two weeks ago or a year ago? I kind of lose track at this point. Um but it has a much broader context window. So if you have a small code base, you can get most of the code base inside the context window as opposed to cloud 35.

Um claw 35 is has proven to be better for more surgical stuff. Cloud 37 is like cloud 35, but it kind of runs away and tries to do a lot. Um so some people like that, some people don't.

Um, every engineer ends up with their very ordinate, well orchestrated workflow at Gauntlet and they, you know, they do things their own way.

Whether it's, um, you know, hey, I build all the front-end stuff really, really well and make sure that's all working and then I try to build a back end behind that or I diagram everything and plan everything out with a model like 03 and then I'm creating a checklist and pulling pieces off the checklist.

Every everybody has slightly different workflows, but uh it's on Twitter. What What are you seeing on the other side on the on the on the hiring side? Are companies expressing preferences already to you to some degree around, hey, you know, this is the stack we're using internally.

We want people to kind of be ramped already on on this set of tools or is it more kind of, you know, bring your own uh toolkit? Yeah. So there's there's uh there's spectrum of course. Um on one side I would say the companies that I I call them the companies that don't really get it yet.

They're like okay give me somebody who has five years of experience with Typescript and they understand a little bit of AI and we'll drop them into our traditional engineering org. Um the companies that are on the other side of that spectrum are more actually this what was this entire team can now be one person.

Um and what language we use matters a whole lot less than how good they are at utilizing AI because anything that you don't know. AI is better at writing code academically than any engineer you're going to talk to.

Um knowing how to get it to write the right code at the right time in the right place is where all the magic happens. So on the other end of that spectrum, there are companies who are I mean the average company that we bring into Gauntlet and we show them what students are doing.

They walk away saying, "Oh my gosh, I need to rethink our hiring. I need to rethink our entire strategy. Our roadmap needs to change. " Uh it is a fundamentally different world when you're utilizing AI in the correct way.

And and then I guess even further on the other end of the spectrum, there are a lot of companies who tried it once and wholesale rejected it and aren't going to really use it or if an engineer does it at all, it's on their own time and against the will of the manager.

Um there's more of that than people would anticipate if you're spending all day on X like I am.

Um, but but yeah, there's a pretty broad spectrum and I go back and forth with frankly how much time I'm going to spend trying to convince people to do it all the right way or to experiment with new ways of building companies versus just okay, you know what, I'm going to go create a holding company and spin up 10 companies myself and show everybody what can happen.

Um, I I go back and forth on that, but but yeah, it's it's wildly wildly different. Mhm. What is the what what's your read on the on the hiring market right now? Obviously, there was big headlines. Maybe it was last last week, I believe, midweek.

Microsoft that again, our our read on that situation was Microsoft has always taken taken the approach of kind of cutting the bottom 3%. There still was a bunch of stories that came out from people being like, I've been here however many years working 12 hours a day. Florida says they're not going to hire anyone.

that kind of went back on that like yeah overall hiring market temperature what's it looking like um I mean I think so first of all I agree with you that when there's a 3% riff that's rarely actually a riff that's performance management being done in as kind of way as possible so you can give people severance um which if you're a company like Microsoft with as much cash as they have you probably should do it that way um that said generally say generally speaking what we're seeing is a if you're really really good at being an AI engineer.

I mean, we had there were times when we had multiple billionaires wandering around our office with printed out offer letters for people begging them to come work for them. Um, that's what it looks like if you're really good at AI. Most companies are not really understanding that. So, that's not everybody.

Um, it is interesting. I I think it's fair to say the junior engineering market is completely decimated. It's pretty well gone. Um, even I mean if you're at at Stanford, maybe you can get an okay internship somewhere, but it is it's Armageddon to be. What does that mean?

Does that mean that like junior engineers should be vibe coding companies and starting more apps and trying to just like build lifestyle businesses almost like what what is your advice for people entering the workforce in software engineering capacities? Yeah, that's a really good question. It's it's a tough one.

Um because the the way that I describe it is someone who's not very good at AI and AI engineering can still get roughly the output of a you know starting junior engineer out of AI. And that's what a lot of companies are doing. Instead of hiring a junior engineer, I'll just get you know the output that I need.

There's still and there always is crazy demand for senior and you know high-end and everything else. Um I think it'll take a while to figure out how that plays out because the flip side of that is you don't get super senior engineers if you don't have junior engineers.

Um my and then you talk to the universities and mo the average university that I talk to is closer to ex explicitly and permanently forbidding any type of AI usage in the classroom than they are to interweaving it or adopting it in any material way. Um because it's cheating.

It's so much better at writing code than you are that the professor has no idea if you know how to write code because the AI will write the code for you. Send him latest post. I was at the I was at the park a couple days ago and uh was, you know, at the swings, you know, pushing the little guy.

Uh and there was another dad there who was a teacher and he said that that a high school teacher. He said they basically now weight homework at effectively zero. Like it's like kind of check the box.

It it maybe contributes to like a few percentage points of your grade and then tests are now just 100% because it's the only environment in which you can actually prove. Would you advise people to learn to code now or has the idea of learning to code changed in some fundamental way?

So I still think I feel like I'm almost on an island on this. I still think you should learn to use AI and if you want to use AI in software environments, you should learn to code. Now, to me, that looks very different than it did five years ago.

Um, and our, you know, our learn to code uh business, we we've basically shut down. It no longer operates. Um, I'm not interested in teaching people to code in the traditional way.

Um, now the flip side of that is normally when you would train people to code, you would start kind of close to the metal and at the bottom of the stack. And so you would start with, okay, let's write binary and then we'll figure out how to write some, you know, low-level Java.

And eventually, a year and a half later, you start actually building programs that can do things that are interesting to you. The right way to learn to code is actually in my mind the inverse where you start by building applications and then you figure out what's broken and what's missing and what you don't understand.

And the thing that is different about that than you know just a creating a different curriculum is AI understands that AI understands um you know so we had software that would basically watch what you were doing on your computer and from that we could derive what CS principles you understood what CS principles you didn't understand we could have AI generate a custom curriculum that you know meets those guidelines specifically and we can fill any gaps in a way that I would have killed to a decade ago training people how to code.

So I think you can actually just start by building stuff and then figure out how it works down and you know start at the top of the stack and slowly work your way down. Um, and to what extent you get down the stack depends on what you want to do, right?

Not the average engineer today isn't, you know, playing with a Linux kernel or writing C. They're, you know, writing JavaScript and they're building applications. Um, to be at that level, you can probably get there relatively quickly. Um, but I think I'll credit Martin Casado uh from A16Z with this.

um who said basically for to be a good engineer you've always wanted to understand one layer of the stack beneath what you were actually working on and I think that remains true. Yeah. Uh give me your view on the AI software engineer market.

OpenAI now has like three offerings between you know 03 will just write code randomly for you. Codeex is a new product. They now own Windourf. Um there's top down enterprise companies like Cognition and Devon. There's bottoms up enterprise like Windsurf and uh Cursor. Um how does this all play out?

What do you look what are you seeing in the like new this entirely new market of AI tooling? Yeah, my my first point here is um we're in an incredibly lucky position because as a as a consumer, you don't have to predict who will win. you can just figure out what's gaining momentum and latch on to the best.

Um, and so, you know, I have a lot of empathy for the companies trying to build those products and spend a billion dollars building a model and if something is 1% better, I'm switching to it tomorrow. Um, so that's a great place to be in as a consumer. How is that playing out?

I think everybody's trying to fight similar battles. Um, Cursor and Windsurf I think are uniquely positioned because they're focused primarily on the UX of the experience as opposed to the data. Um, it's kind of in AI company's DNA to think that every problem is a data problem and they're generally speaking right.

Um, but where Windsor are getting better and better is okay. It's really cool that there's this model. I'd say on average that is the problem that AI companies have. people have no idea what they can do with the models or how they should be using them.

I think that gets simplified over time the same way every product becomes, you know, I shouldn't have to know whether I want to use 03 mini or 40. No normal human's ever going to care about the difference between those. Um, so I think they'll just get better and better. Um, who wins? I I don't have a clue.

Trillion dollar question. Yeah, it is. If you'd probably be a VC if you're in that game, but thanks. so much for stopping by. We'd love to have you back. This is uh this is great. Yeah. Keep us keep us posted. We'll talk to you soon. Cheers, Austin. Bye. Have a good one. Next up, we have Jeff Morris Jr.

Junior himself coming in