Deep Dive Issue: Gauntlet AI

Welcome to a special deep-dive issue of The Austin Business Review.

In last week’s newsletter, I told you I was working on this piece about Gauntlet, the intense AI bootcamp that’s guarantees every graduate a $200k job offer.

I’m fascinated by this company. I’ve been down there twice so far, and it’s just electric.

This thing was way too long to put in Thursday’s email, and it took me longer than I thought to get it done. But I wanted to get it to you asap, because they’re hosting their demo day tomorrow (Aug. 21), where the entire graduating cohort will be showing off stuff they’ve built, including real businesses they’ve launched and gotten paying clients for.

Let me know what you think of this, and if you want to see more like it.

Back tomorrow with our typical programming,

-Ethan

Running The Gauntlet: How AI’s Toughest Bootcamp Is Shaping Your Next Dev Hires

There’s a man guarding the entrance to 416 Congress Ave here in Austin, and if you’re there between 6PM and 8PM, he will likely be the first person you meet.

“Drop off?” he’ll ask, eyeing you through the cracked-open door. And you will have no idea what he means.

But if you look lost enough, long enough, he will take pity on you, and tell you that he works for DoorDash, and that his job is to meet the wave of food deliveries that come here every night around this time. 

He meets them, accepts their cargo, and carries the parcels ten steps across the lobby to a small shelf against the wall without ever letting dashers set foot inside the building.

An unexpected job, perhaps. But spend ten minutes in that lobby at the dinner rush, and you’ll agree this man is providing a critical service; the deliveries are non-stop.

For this building is the headquarters of Gauntlet AI, and by 6PM, her two floors of inhabitants have been here all day long, are hungry, and still have nearly a full day’s work to go before they’ll leave.

Started by Austen Allred, Gauntlet is an intensive ten-week training program designed to take smart, driven young people, and turn them into the kinds of high-leverage, AI-first engineers that will power companies of the future.

It evolved from Allred’s experience building the Bloom Institute of Technology – formerly Lambda School – one of the better-known coding bootcamps to arise over the last ten years.

“Technically, they’re the same company,” he told me as we stepped off the elevator onto the second floor.

In April of 2024, Bloom had announced a course called, AI for Developer Productivity, a fully remote program designed to help companies upskill their existing engineering teams.

Classes quickly grew to 100 or more, and one client – Trilogy, here in Austin – realized that the new AI skill stack required them to rethink their entire talent pipeline. They urged Allred and his team to go all-in with three key changes:

First, the program should focus on developing new talent. Second, it should be extremely intense – no dabbling. And third, it should be in-person.

From those concepts, and with Trilogy helping backstop the program financially, Gauntlet was born.

It’s free for any person who’s accepted, and guarantees a $200k job offer to all who graduate. In between those two is ten weeks of intense, 80- to 100-hour weeks, seven of which take place here in this building.

“Demos, demos. Who’s got demos?” Allred said as we rounded the corner into the bull pen.

The space looked almost exactly as I had imagined. A series of three large, open rooms, filled with rows and rows of sit-stand desks, desktops snaked with cables, and the walls in many places lined with whiteboards.

Throughout the room, heads clustered around monitors, and here and there you could spot the typical Bric-à-brac of smart, ambitious people who work long hours – chess boards sitting in various states of play, cartoons drawn next to technical schema on the whiteboards, a soldering iron perched next to someone’s keyboard. 

“Who’s got demos,” he said again as we strolled the floor. And you could tell from the way he said it that for the people here – challengers, they’re called – this is a daily occurrence. 

He hasn’t told them who I am, and yet no one seems the least bit phased by a stranger walking around, hoping to see their work.

Indeed, it’s a major part of the Gauntlet experience.

The first cohort was largely a recruiting effort for Trilogy and her sister companies (part of how Gauntlet can guarantee job offers to graduates). 

Now in the midst of their second cohort, several more companies have signed on as hiring partners, so challengers are used to visitors of all types.

A young man locked eyes with us, grinning, and shifted in his chair so that we could see both of his screens.

“Got a demo?” Austen said as we stepped around his desk.

His name was Abdurrahman Mirza, and I will try to describe what happened next. But first, you should know, I’m not what you’d call a deeply technical person. I know a little. Enough to be dangerous. But no one has ever accused me of being an early adopter of anything, and when my grandfather’s computer stops working, it’s my dad he calls, not me.

So I will try to describe this accurately…

Mirza blazed through screens, scrolling here, clicking there, chatting amicably the entire time, as though he was born straight into an Aeron chair.

He was building, I gathered, a tool that would digest data from unsolved missing person’s reports, and compare it to other unrelated databases of unidentified bodies in the hopes of finding a likely match and helping lay cases to rest.

Austen peppered him with questions as he zoomed from one screen to the next.

One looked like a profile page, showing details of a potential match. It listed characteristics of the person in question, like their age, where they went missing, a few other traits, and a confidence score that the algorithm had assigned to the match.

Another page was a map with missing persons and found-bodies visibly marked, and golden threads connecting dozens of potential matches, the thickness of each line varying based on that confidence score I mentioned earlier.

This wasn’t a random personal project. It was a challenge submitted by one of the companies hoping to hire from the cohort. And here we come to one of the main differences between Gauntlet and its predecessor, Bloom: The business model.

At Bloom, education was paid for by students. Gauntlet works more like a recruiter.

If a company wants to hire from the cohort, they join as a hiring partner and pre-pay for hiring credits. 

Each credit is currently $50k, and buying more gets you perks, like earlier recruiting access to the group, the ability to host custom events, or to submit challenge projects, like the one Mirza was working on.

Gauntlet charges 25% of a hire’s total year-one comp, so with a $200k offer, each credit is good for one hire. If the comp package is over $200k, the partner simply settles the difference upon hiring.

As I write this, the median salary for those coming out of the program is reportedly $200k (that’s base salary only), though at least one was said to be north of $900k.

The challenge projects have two main purposes.

First, they’re a great way for Gauntlet participants to get the attention of hiring partners they like and might want to work for.

There’s a two-week period near the end of the program during which challengers must complete at least four of these partner projects, and members of this cohort were able to choose from more than fifty that were available.

As they build, the hiring partner is the one grading or giving feedback on the work, so it helps test what the working relationship might be like with a hire.

But the hiring partners can also get value from the deliverables themselves.

They often represent some utility that the partner needs – a problem they’re trying to solve, or some part of the business they wish could be more efficient. 

Sometimes, it’s a service they’re currently providing manually, and they want to see different ways AI might be used to automate it. Other times, it’s an app they need built but don’t have bandwidth to do themselves.

For this cohort, one partner submitted about twenty-four apps they needed on a very short deadline.

To sweeten the deal, they offered challengers a $10k bounty for each they delivered. 

The Gauntlet crew attacked the work, knocking out about twenty of the apps, and on the night of my first visit, staff was preparing for an armored car that was coming the following day to dole out $200k in cash.

Mirza’s missing-persons matcher had started as a challenge from a startup in the forensics space, and had taken him a couple of days to build.

It was one of two projects he’d done for that company, the other one being a program that used AI to visually tag and label crime scene photos, which had taken only one day and was slotted for use in the real world.

“Wild,” was all I could say; All I had said for whole minutes now. “Had you ever built anything like these before?”

“Nah,” he said matter-of-factly. “I just graduated. We all did.” He motioned to the group of young men working nearest him.

They were computer science grads from a mix of prestigious universities, and to a man, they said they learned more about building in the first few weeks here than they had throughout much of their time at school.

These were themes I found repeated over and over as we moved among the challengers that evening. 

Many of them had much more of a technical background than I’d been expecting – typically a blend of either CS degrees, or a few years spent working at mag-seven companies.

“We tried without that,” Allred told me. “But the learning curve is really steep.”

Still, there are exceptions. 

Ash Tilawat, Gauntlet’s Head of Product, calls them “the wonder kids” – the small percentage of the cohort that doesn’t come from a traditional CS background, but is able to thrive here in spite of that.

They tend to still be top-performers in their previous fields – former military, that sort of thing.

“The biggest thing I’m looking for is work ethic,” he told me, then offered up the example of one challenger who used to be a writer for ABC, and only started coding in recent years. He worked hard at it, and had just signed a job offer earlier that week.

“Gauntlet is not just for people who are experienced,” Ash said. “We love to have people who are experienced. But I think there’s at least a small percentage – maybe ten, fifteen percent – who will be the wonder kids. We love it.”

As Head of Product, he oversees most of the curriculum development here and has been working with Allred since the early days of Lambda School.

When we met, I asked him about another theme I noticed among the challengers I’d spoken to – virtually all of them, when asked, said that weeks one or two were the ones that almost broke them. 

Why’s that?

“At that point, they’re still coming up with their AI framework,” he told me.

The first three weeks are entirely remote, and in that time, participants build mobile, web, and desktop apps, all with very little guidance, and on extremely tight deadlines.

They don’t yet have a repeatable process – or favorite AI tools – for researching the problem, engineering a solution, building it, and verifying that it works, so the whole thing is incredibly difficult.

It’s an important stage though. Challengers are learning how to think, rather than a concrete list of tools to use, and that serves them at every stage beyond that.

After week three, anyone who’s still in the program is flown to Austin, where they have seven more weeks of training, each with a specific focus.

First, there’s a team project, which is designed to see how people work together and sift out any slackers.

Then, they design a video game. That one’s important because it typically exposes students to a tech stack they’ve never used, and forces them to learn and build quickly.

Then, the “Enterprise” project – students are required to pick an old, unmaintained code base of at least a million lines and use AI to streamline it and add modern features.

“It’s like, okay, you’ve been living with rainbows and unicorns thinking you’re gonna green-field everything so far,” he told me. But real companies need people who can tackle large existing codebases, and that’s what that step prepares them for.

It also helps to chip away at the popular myth that AI can’t be used on old school code.

During this cohort, Gauntlet offered a thousand-dollar bounty to any challenger that did their enterprise project on a code base using fortran, cobalt, or assembly.

“AI has not been trained on these,” Tilawat told me. “There’s no training data… And like, all of our infrastructure in this nation runs on these languages.”

“We were like, we think AI can do this. We have to understand how to context engineer, but we do think it’s possible. And so Austen said, $1,000 for anybody who works on any of those types of repo’s, and if you’re able to make significant process on them, you win.”

They ended up paying out bounties on four projects, including the one in this video, which took an old math library written in Fortran 77 and updated it to Fortran 90 exclusively using AI.

After that, the remainder of the program is broken into two parts.

First, the partner challenges, in which participants tackle those four hiring partner projects I mentioned earlier.

And finally, their capstone project – challengers work either solo or in groups to dream up their own business idea, then have two weeks to build, launch, and get paying customers.

That last one is graded on a matrix: 0.5 points for each member of the cohort who signs up for your product, 1.0 point for any stranger outside the cohort, and 100 points if you get an SMB or enterprise company to sign up.

You need a total of at least 100 points to pass, and you need to present ad Demo Day.

It’s not just about technical experience. Throughout the program, challengers are also pushed to hone the soft skills needed to be successful in the field. 

For example, communication. Everyone is required to share each of their projects on X, and challengers frequently go live in front of thousands to stream about the projects they’re working on.

Indeed, there’s some great technical content to be had, watching Gauntlet challengers live-test Grok 4, showcase their projects, or teach about Claude Code.

“I make them come out of their comfort zone and try do that,” Tilawat told me, “which forces them to have agency.”

It also helps land jobs.

When Nataly Smith and her team of fellow challengers built an open-source cancer risk genetic screening tool (in 24 hours), all four of them posted about it on X, where Allred also boosted it, and I’m told one of them signed a job offer shortly thereafter.

All of this plays into two key concepts that Tilawat says lay at the foundation of Gauntlet’s philosophy.

First, that product manager and engineer are becoming the same profession.

“You need to understand product, users, feedback, iteration,” he told me. “Because one day we believe that there’s gonna be digital teammates that you’re in charge of, and they’re gonna be coding for you. So the only thing that really matters is what you’re building and how you spread awareness about it.”

Several challengers are already dabbling in a version of this, deploying different versions of Claude Code in cloud containers, and having them work off different git work trees, tackling different tasks, and being managed autonomously.

“Is it perfect right now? No,” he said. “Will we get to a version that’s perfect in the next year? Probably.”

So the future of engineering, as he sees it, is small teams of perhaps two or three humans. Maybe less.

And the second major philosophy here: Agency is the skill.

“It’s probably the only skill that matters,” he said. And it’s why they push so hard on demo-ing.

“We want Gauntlet engineers to not only be technically very good, but I want them to go to the company and be the AI evangelist. [To say,] here’s how you use AI to code… Here’s how I would think about AI within the engineering team.”

The curriculum changes with every cohort, because of how fast AI tools are evolving. But Tilawat is at pains to point out that the stack is less consequential than most people think.

“It’s not the stack,” he says. “It’s the hours.”

Each challenger develops their own approach to problem solving, he said, thanks to the sheer volume of work and projects they’re forced to do. They vary from person to person, but it’d be a mistake to place too much weight on the tools they’re using.

“I could tell you the stack right now,” he continued. “Thirty percent are probably Cursor users… Probably sixty percent are Claude Code users, and another ten percent are Claude Swarm users. Then they have these auxiliary tools – RepoPrompt, GitIngest, DeepWiki, Perplexity, so on and so forth.”

“But it’s not the tool. It’s knowing when to use which tool. When, and how. And knowing when to use which model inside the tool, depending on what task you wanna do.”

Outsiders appear to be looking for the easy solution when they ask him for the Gauntlet tool stack, he says. But the brute force hours are the thing they’re typically missing.

“The reason they’re so good is because they’re not dabbling,” he said of the cohort. “They’re like, hey, I’m gonna do this for a thousand hours. Let’s see what happens.”

That’s all for this issue! If you enjoyed the story, connect with me here, and keep an eye out Thursday for the next edition.

Hit reply to share any feedback, or let me know what you think.

Until next week,

-Ethan