# Why Your AI Agents Don’t Work with Dex Horthy of HumanLayer | In-Context Cooking

Latent Space · 2026-03-06

<https://addtry.com/2b94b1fe-2689-4fac-8845-eaa30eec133a>

Dex Horthy, founder of HumanLayer and creator of "context engineering," joins Alan to recreate Dan Dan noodles while explaining why most AI agent demos fail in production and how the top 1% of AI engineers build differently from the rest. Horthy describes how his NASA JPL internship at 17 taught him pathfinding with Dijkstra's algorithm, and later coining "context engineering" after observing that public hype around open-source frameworks doesn't match enterprise reliability needs. He defines AI "slop" as low-effort derivative content often produced without reading, and introduces the "dumb zone"—a rule of thumb to start thinking about context compaction at 40% usage unless you have strong intuition. Horthy argues that while smarter models narrow the gap, the frontier of reliably achievable tasks keeps shifting, so context engineering remains valuable. He predicts software engineering will evolve from writing code to producing working code, emphasizing shipping over coding, and notes HumanLayer helps enterprises standardize AI adoption with guardrails rather than vibe coding.

## Questions this episode answers

### What is the 'dumb zone' in AI context windows that Dex Horthy talks about?

Dex Horthy describes the “dumb zone” as around 40% of a model’s context window capacity. For beginners lacking intuition, it’s a rule of thumb to stop adding context and start “intentional steered compaction” to avoid diminishing returns. Experienced users may push to 60–70% when the trajectory is good, but the 40% guideline serves as training wheels for new AI engineers.

[15:23](https://addtry.com/2b94b1fe-2689-4fac-8845-eaa30eec133a?t=923000)

### Why do most AI demos fail to translate to production, according to Dex Horthy?

Dex Horthy explains that AI demos and production systems have different incentives: demos only need to look impressive and work ~80% of the time, while enterprise products demand near-perfect reliability. He observed that the top 1% of AI engineers build very differently from public frameworks and indie hackers, using techniques like context engineering to achieve consistent results rather than flashy one-off demos.

[5:09](https://addtry.com/2b94b1fe-2689-4fac-8845-eaa30eec133a?t=309000)

### How does Dex Horthy think AI will change the role of software engineers?

Dex Horthy predicts software engineers will shift from personally writing code to orchestrating AI that produces working code. He distinguishes “coding” (making software work locally) from “shipping” (deploying, maintaining, and improving in production). While AI excels at generating code, the reliability and maintenance of shipping are not yet automated, so engineers will focus on high-leverage tasks where human judgment is critical.

[18:46](https://addtry.com/2b94b1fe-2689-4fac-8845-eaa30eec133a?t=1126000)

### How does Dex Horthy recommend building intuition for engineering with AI models?

Dex Horthy advises building intuition by using AI (like Claude) extensively—up to 70 hours a week—and learning through trial and error. Practitioners will sometimes over-engineer simple tasks or under-engineer complex ones, but repeatedly putting in “reps” is the only way to gauge how much context engineering a problem needs. There’s no shortcut; even model labs agree this hands-on experience is essential.

[16:56](https://addtry.com/2b94b1fe-2689-4fac-8845-eaa30eec133a?t=1016000)

## Key moments

- **[0:00] Intro**
- **[1:22] NASA Internship**
  - [1:25] Dex Horthy landed a NASA Jet Propulsion Lab internship at 17 by building a naive Dijkstra's algorithm for rover pathfinding.
- **[3:35] Context Engineering**
  - [3:36] Dex Horthy coined 'context engineering' after discovering the top 1% of AI engineers build differently from the bottom 99%.
  - [5:08] The top 1% of AI engineers optimize for reliable enterprise software, while the bottom 99% build demos that only need to work 80% of the time.
- **[6:21] AI Slop**
  - [6:31] Q: How does Dex Horthy define AI 'slop'? A: Any low-effort, derivative, AI-generated content that is full of garbage and not actually read.
- **[8:06] Replicate & Founder**
  - [8:06] As Replicate's first customer-facing engineer, Dex Horthy closed 12 deals in 3 months after they were stuck for 9 months, saving the quarter.
- **[12:54] Context Engineering Evolves**
  - [13:23] Dex Horthy: Engineers who mastered context engineering got Opus 4.5 results from Opus 4.0, proving skill matters more than model upgrades.
- **[15:21] The Dumb Zone**
  - [15:27] Dex Horthy's 'dumb zone' rule of thumb: use 40% of the context window as a training wheel; experienced engineers can push to 60-70%.
- **[16:56] Building Intuition**
  - [17:21] 'Talk to Claude for 70 hours a week. That's how you do it. You gotta build intuition.' — Dex Horthy on developing AI engineering intuition.
- **[18:59] Future of SWE**
  - [18:59] Dex Horthy predicts the software engineer role will evolve from writing code to causing working code to be produced, not disappear.
  - [19:45] 'There's a difference between coding and shipping.' — Guillermo Rauch, cited by Dex Horthy.
  - [20:34] In fintech and enterprise, Dex Horthy says 'we can't afford to get it wrong' — no ship fast and break things.
- **[21:03] Plating**
- **[22:02] Taste Test**
  - [24:24] Dex Horthy wins the Dan Dan noodle cooking competition with a meatier protein ratio, as judged by Shawn Wang and the host.
- **[24:27] Verdict**

## Speakers

- **Alan** (host)
- **Swyx** (host)
- **Dex Horthy** (guest)

## Topics

Agent Platforms

## Mentioned

HumanLayer (company), NASA Jet Propulsion Lab (company), Replicate (company), Y Combinator (company), Claude (product), NotebookLM (product)

## Transcript

### Intro

**Dex Horthy** [0:00]
Look at that. Look at that sauce. Fully incorporated. I'm sure the model labs love that that's the answer. This is getting chaotic, dude. This is fun. Yeah, I wanna, I want, I want the verdict here.

**Alan** [0:12]
Hey guys, welcome to In-Context Cooking, a show where we take one dish, taste it, and try to recreate it with minimal help. My name's Alan, and today we have a very special guest, the man behind Context Engineering, Dex Horthy, founder and CEO of HumanLayer.

Welcome, Dex.

**Dex Horthy** [0:27]
Dude, I'm so stoked to be here.

**Alan** [0:29]
Yeah. I mean, I always start off with this. So on a scale of one to 10, how would you rate yourself as a cook? One being bad, 10 being amazing.

**Dex Horthy** [0:35]
I would put myself at, like, seven or eight. I don't cook that much anymore.

**Alan** [0:38]
Okay.

**Dex Horthy** [0:38]
And I do it 'cause it's kind of, like, fun. I li-... The idea of just following a process which is very deterministic, uh, but also kind of chaotic is, is, uh, is a lot of fun, so.

**Alan** [0:46]
Yeah, we like that. Chaotic, but with, um, certainty. Looking at these ingredients we have before us, do you have any guesses of the dish we're gonna try and make?

**Dex Horthy** [0:54]
I mean, clearly we're doing some sort of noodle. I see some bok choy.

**Alan** [0:58]
Okay. Yep.

**Dex Horthy** [0:58]
Um, I'm... But, uh, yeah, I, I, I don't know a lot about noodle dishes, so, uh, I'm excited to see what we got.

**Alan** [1:04]
Yeah. So the dish we'll be trying today is Dan Dan noodles.

**Dex Horthy** [1:08]
Oh, nice. I love Dan Dan noodles.

**Alan** [1:09]
Yeah. We have a lot of protein here, some flavors. Very simple to make at home, but yeah, we'll taste it and then try to recreate ourselves.

Great.

**Dex Horthy** [1:18]
Oh, that's fire. I'm gonna go for one more. That's awesome.

**Alan** [1:20]
Are you also good with spice or are you kind of-

**Dex Horthy** [1:21]
Oh, I love spice.

### NASA Internship

**Alan** [1:22]
Okay, great, 'cause we do have some chili oil here. I've read that you've started coding since you were 17, and you had an internship at NASA for the Jet Propulsion lab.

**Dex Horthy** [1:31]
Yeah.

**Alan** [1:31]
So, like, how does a high school kid end up getting an internship at this very prestigious lab?

**Dex Horthy** [1:36]
Uh, so they do have, they had this high school internship program where it was like-

**Alan** [1:39]
Mm-hmm

**Dex Horthy** [1:39]
... they would take for a summer. It was, like, summer after my junior year.

**Alan** [1:42]
Yeah.

**Dex Horthy** [1:42]
And they just, like, put us up in dorms at a local, like, university that was kind of empty. And so there's... The reason why people wanna go explore the South Pole of the moon is 'cause there's, like, craters there that are so deep-

**Alan** [1:54]
Mm

**Dex Horthy** [1:54]
... that there is, like, frozen ice there. The moon's very dry, but in, in these craters there's, there's, uh, water that has never been hit by sunlight. And so it's there for, it, like, has been frozen there since the formation of the moon, since it got, like, ejected from the Earth.

**Alan** [2:09]
Oh, wow.

**Dex Horthy** [2:10]
And so it's like, "Oh, we really wanna go see what's in that ice and collect samples and study it." Problem is, very, very deep craters.

**Alan** [2:16]
Yeah.

**Dex Horthy** [2:16]
And so, like, most of the rovers can't do it.

**Alan** [2:19]
Wow.

**Dex Horthy** [2:19]
And so we had to, like, we were basically, like, building some, like, pathfinding algorithms of, like, okay, given a rover and its current capabilities, and, like, what's the steepest hill it can go up? Like, I wanna go from here to here.

And I was 17. I didn't have a CS degree. Like-

**Alan** [2:31]
Mm-hmm

**Dex Horthy** [2:31]
... so we built a very naive implementation of Dijkstra's algorithm- ... to, like, find the shortest path according to some, like, constraints of what the rover could do.

**Alan** [2:39]
Great. Yeah. I mean, that's a great start. So we can now look at some of the spices so we have it in our context before we go off to cook.

**Dex Horthy** [2:46]
Here we go.

**Alan** [2:49]
I think we're ready to go. Are you ready? Do you feel comfortable and we get started?

**Dex Horthy** [2:53]
Let's ship it, dude. Let's go.

**Alan** [2:56]
Dex, are you ready?

**Dex Horthy** [2:57]
All right. Tell me what we're doing here, baby.

**Alan** [2:58]
Okay. Let's start with the meat first. So-

**Dex Horthy** [3:01]
Okay

**Alan** [3:01]
... you should probably see some-

**Dex Horthy** [3:04]
Pork?

**Alan** [3:04]
Yeah, ground pork. And so ta- let's take one of the bowls.

**Dex Horthy** [3:08]
Yep.

**Alan** [3:08]
Um, and just mix some of... Put the pork in there. Shoot.

**Dex Horthy** [3:13]
Okay.

**Alan** [3:13]
Uh, I don't know how much pork you have. I'm gonna try to put not too much to crowd it and then

yeah, I believe we use-

**Dex Horthy** [3:22]
Switching bowls here

**Alan** [3:23]
... the Shaoxing wine. Okay. Let's cut up the garlic.

**Dex Horthy** [3:27]
How much garlic you going for?

**Alan** [3:29]
Um, honestly, probably, like, three cloves. Two, three cloves.

**Dex Horthy** [3:32]
Three cloves?

**Alan** [3:33]
Yeah.

**Dex Horthy** [3:33]
All right, cool. I'm doing six.

**Alan** [3:35]
Okay. How did you get into context engineering? I feel like, you know-

### Context Engineering

**Dex Horthy** [3:39]
Oh

**Alan** [3:39]
... coining the term, now it's like-

**Dex Horthy** [3:41]
Incredible. Ac- according to Swix, there are- There are many people who, who, uh-

**Alan** [3:46]
Jamie Sorkin

**Dex Horthy** [3:46]
... Gemini thinks it was Andrej Karpathy.

**Alan** [3:48]
Okay. I mean, that's a very valid guess.

**Dex Horthy** [3:50]
Yeah. And then I, uh, and then it had a crash-out about Andrej Karpathy. I literally, I have a screenshot, it just, like, says the word An- Andrej Karpathy like 30 times in a row 'cause something gets, uh-

**Alan** [3:59]
Okay

**Dex Horthy** [4:00]
... yeah, you know. The standard-

**Alan** [4:01]
Yeah

**Dex Horthy** [4:01]
... Gemini crash-out. Uh, yeah, so basically I was building, like, dev tools for AI engineers.

**Alan** [4:08]
Okay.

**Dex Horthy** [4:08]
And, uh, what happened was I, like, talked to all the best founders and engineers and founding engineers that would, like, take my call, and I was like, "Hey, we're making this thing that helps you build better agents." I won't go into, like, the specifics of it-

**Alan** [4:22]
Yeah

**Dex Horthy** [4:22]
... but it was built of, like, I had this assumption about how, uh, people build agents.

**Alan** [4:27]
Mm-hmm.

**Dex Horthy** [4:28]
And, uh, so I built tools for, to, like, map onto that way of building agents, right?

**Alan** [4:34]
Mm-hmm.

**Dex Horthy** [4:35]
And the problem was, was that when I talked to all these people who were shipping, like, real AI to the enterprise-

**Alan** [4:41]
Yeah

**Dex Horthy** [4:41]
... like, reliable systems, it was like, I think, I think I saw this, uh... It was, like, an old tweet from, uh, from, like, Swix, actually. It was this idea of, like, the way the top 1% build is so different from the bottom 99%.

**Alan** [4:55]
Mm-hmm.

**Dex Horthy** [4:56]
You have all your, like, indie hackers and, like, open source frameworks that are very, very popular.

**Alan** [5:01]
Yeah.

**Dex Horthy** [5:01]
And, like, everyone uses, and that's what you see in public.

**Alan** [5:03]
Yeah.

**Dex Horthy** [5:03]
And then you go see how real people are building stuff and getting reliability, and it's completely different.

**Alan** [5:08]
Oh, okay.

**Dex Horthy** [5:08]
And so, like-

**Alan** [5:09]
Why do you think there's such a discrepancy?

**Dex Horthy** [5:11]
I think there's a difference between, like, people who wanna build, like, reliable software and people who wanna build a cool demo.

**Alan** [5:17]
Okay, yeah.

**Dex Horthy** [5:18]
I think that's the core of it.

**Alan** [5:18]
The incentives are different.

**Dex Horthy** [5:19]
The incentives are different. Yeah.

**Alan** [5:20]
Okay.

**Dex Horthy** [5:20]
The incentives of, like, okay, if this is right 80% of the time, that's fine. It'll look good on the demo.

**Alan** [5:25]
Yeah.

**Dex Horthy** [5:25]
But if you wanna go sell something to a real company and, like, charge them-

**Alan** [5:29]
An enterprise

**Dex Horthy** [5:29]
... you know, 100K a year-

**Alan** [5:30]
Yeah

**Dex Horthy** [5:30]
... it's gotta be much more, uh-

**Alan** [5:32]
Bulletproof and not as-

**Dex Horthy** [5:34]
Yeah.

**Alan** [5:35]
Okay, that makes sense.

**Dex Horthy** [5:36]
So I learned all this stuff and I was like, damn, I got a little bit, like, distracted by the, like, public hype machine.

**Alan** [5:42]
Mm-hmm.

**Dex Horthy** [5:42]
And so I said, "Okay, cool." Like, I don't want anyone else to go through this. So I wrote down everything that I learned and I called it, among other... Like, there was 12 chapters. There was only one of them was really about context engineering.

**Alan** [5:53]
Yeah.

**Dex Horthy** [5:53]
But that was the one that stuck.

**Alan** [5:55]
And then you went through Y Combinator, right? In 2024.

**Dex Horthy** [5:59]
Uh, yes.

**Alan** [6:00]
Okay.

**Dex Horthy** [6:00]
Yeah. So this is, this happened, we did Y Combinator in fall of 2024, and this whole, like, 12-factor agents context engineering journey happened in, like... Kind of like Q1, I wrote about it in Q1 of 2025-

**Alan** [6:12]
Okay

**Dex Horthy** [6:12]
... and then it kind of like went viral in April.

**Alan** [6:14]
Yeah, 'cause you had a couple AI engineering talks that, um, even recently, right, in New York. No Vibes Allowed, was it the title? That-

### AI Slop

**Dex Horthy** [6:21]
Yeah

**Alan** [6:22]
... you know, had, had some good reception. One thing that was very commonly said, even I think Swix had a talk about, was about slop. And I'm kind of curious, like-

**Dex Horthy** [6:30]
More slop

**Alan** [6:31]
... how do you define slop?

**Dex Horthy** [6:32]
I mean, it's kind of fun. The word has kind of evolved.

**Alan** [6:35]
Yeah.

**Dex Horthy** [6:35]
And like, you know, it used to be like that's just AI slop, but now-

**Alan** [6:39]
Okay

**Dex Horthy** [6:39]
... you can say slop about anything, I think.

**Alan** [6:41]
Yeah.

**Dex Horthy** [6:42]
Of just like any like low effort or like contrived or derivative thing that is like someone didn't put a lot of effort into.

**Alan** [6:48]
Yeah.

**Dex Horthy** [6:49]
It's obviously most common when someone just like uses AI to write a 10-page document that they didn't read and is like full of garbage.

**Alan** [6:56]
Yeah.

**Dex Horthy** [6:56]
But, um... All right, so all the gar- garlic's going in the meat?

**Alan** [7:00]
Yes.

**Dex Horthy** [7:00]
Okay.

**Alan** [7:00]
All the garlic, you just put it in there. We'll let the meat sit a little bit, and then we'll work on the sauce.

**Dex Horthy** [7:06]
All right.

**Alan** [7:07]
So, you should have a bowl in front of you. We can start off with a paste. I believe this is the... Yeah. Sesame paste-

**Dex Horthy** [7:13]
The sesame paste

**Alan** [7:13]
... you smelled earlier.

**Dex Horthy** [7:13]
Right?

**Alan** [7:14]
Yeah. Add like a spoonful.

**Dex Horthy** [7:17]
Small spoon or big spoon?

**Alan** [7:18]
Big spoon.

**Dex Horthy** [7:19]
Okay.

**Alan** [7:20]
Maybe like a s- big spoonful and a half.

**Swyx** [7:22]
Aren't you always big spooning?

**Alan** [7:26]
That's a crazy statement.

**Dex Horthy** [7:28]
Wow, Swix.

**Alan** [7:30]
Okay. Um, yeah, so sesame paste.

**Dex Horthy** [7:33]
Can we, can we quiet down in the back there, please?

**Alan** [7:36]
Okay. Um, we can add some sugar.

**Dex Horthy** [7:40]
Okay. Look at that. Look at that sauce.

**Alan** [7:43]
Okay.

**Dex Horthy** [7:43]
Fully incorporated.

**Alan** [7:45]
Chili paste. Now, do like a spoonful or two, depending on how spicy you like it. I'm doing like two.

**Dex Horthy** [7:51]
We're gonna go big here.

**Alan** [7:52]
Yeah.

**Dex Horthy** [7:53]
I didn't get my four cloves of garlic in, so...

**Alan** [7:56]
Okay. Let's move on to... We're on a time crunch. Let's move on to cooking the beef. Not the beef, the pork. So, turn on your induction stove.

**Dex Horthy** [8:04]
Yep.

**Alan** [8:05]
You with me? Okay.

### Replicate & Founder

**Dex Horthy** [8:06]
We're on.

**Alan** [8:07]
I saw that you were at Replicate, was it, for seven years?

**Dex Horthy** [8:10]
Replicate was dope. It was, um... We built... I started as an engineer, and we built a platform for, uh, how do we help... Uh, if you have like a SaaS application-

**Alan** [8:21]
Yeah

**Dex Horthy** [8:21]
... uh, like, you know, people subscribe, they wanna use it, and then you wanna sell it to like a big enterprise, like a bank or something.

**Alan** [8:27]
Mm-hmm.

**Dex Horthy** [8:28]
Um, they're pretty, like strict about data controls.

**Alan** [8:30]
Okay.

**Dex Horthy** [8:31]
And so instead of like using your SaaS, they'll almost always ask you to like send the app into their data center, into their AWS cloud.

**Alan** [8:39]
Gotcha.

**Dex Horthy** [8:40]
And that's a very hard and like expensive thing to build as like an engineering team.

**Alan** [8:44]
Yeah.

**Dex Horthy** [8:44]
And so we built a lot of tools to help solve that problem.

**Alan** [8:47]
Okay. Would you say there was a lot of learnings and, you know, experiences that you transfer even today with HumanLayer?

**Dex Horthy** [8:53]
Um, yeah. So I, I was an engineer there for like two years, and then I like... I kind of always knew I wanted to be a founder.

**Alan** [8:58]
What made you say that? Were there just moments that-

**Dex Horthy** [9:01]
There was just like... I just like... I don't know. I had a... I... One of the guys I did all my CS classes with in college, he would, um... Are you going veg in first?

**Alan** [9:09]
You put in the meat.

**Dex Horthy** [9:11]
Oh, the meat in. Okay. Yeah, yeah, yeah.

**Alan** [9:11]
Just the meat.

**Dex Horthy** [9:12]
Yeah.

**Alan** [9:12]
The veg we're saving for later.

**Dex Horthy** [9:13]
Okay. Like I said, I kind of like always wanted to be a founder.

**Alan** [9:16]
Mm-hmm.

**Dex Horthy** [9:17]
And so I wanted to do all the other things. So this opportunity came out to like basically do a more like salesy side of the job.

**Alan** [9:23]
Gotcha.

**Dex Horthy** [9:24]
Uh, and I like jumped at it. I was just like, "Yeah, I wanna learn sales, and I wanna learn product, and I wanna learn all..." So like I ran... I basically became the first customer-facing engineer.

**Alan** [9:32]
Okay.

**Dex Horthy** [9:33]
Or s- uh, yeah, basically. So, uh, and we had all these deals. We had like parted ways with our head of sales, so he had a bunch of deals on the like radar-

**Alan** [9:43]
Yeah

**Dex Horthy** [9:43]
... for like, "Hey, we're gonna close."

**Alan** [9:45]
Yeah.

**Dex Horthy** [9:45]
And then like they'd just been stuck for like nine months.

**Alan** [9:47]
Okay.

**Dex Horthy** [9:47]
And I went and I met with all of them, and we closed like... I think we closed like 12 deals in three months.

**Alan** [9:53]
Oh, wow. 12 deals in three months.

**Dex Horthy** [9:54]
And the CEO was losing his . He's like, "Holy . Like the investors are taking my calls again, and we just made the quarter for the first time in a while."

**Alan** [10:01]
Yeah.

**Dex Horthy** [10:01]
Uh, and it r- it was very exciting. He's like, "Dex, I know you wanna get back to coding, but like can you go hire three more people and like turn this into a team? 'Cause whatever, whatever you're doing of like going and onboarding into our, our stuff, it is, it is working."

**Alan** [10:14]
Gotcha.

**Dex Horthy** [10:15]
So that's how I got, that's how I got pulled out of like writing code all day and into like helping other people use this like Kubernetes Terraform like deployment s- deployment system.

**Alan** [10:26]
Gotcha. And even today, like would you say your work is kinda distributed or less engineering, more salesy?

**Dex Horthy** [10:34]
Uh, I still write a lot of code.

**Alan** [10:36]
Okay.

**Dex Horthy** [10:36]
Uh, I, I brought on, uh, like a technical co-founder over the summer.

**Alan** [10:40]
Yeah.

**Dex Horthy** [10:40]
Uh, and he's, he's awesome. Uh, he is a much better engineer than me. So I let him write most of the code.

**Alan** [10:47]
Gotcha.

**Dex Horthy** [10:47]
We like to joke. It was like, uh, yeah, we have the early... Oh, I was talking to some people last night, and I was like, "Yeah, we have an early product. It's got a couple bugs." And he's like...

And someone's like, "Yeah, all software has bugs." I'm like, "Well, this isn't really bugs. It's more like the features that Kyle let Dexter ship instead of building them himself."

**Alan** [11:02]
Mm-hmm.

**Dex Horthy** [11:02]
Yeah, so I'm, I'm probably, uh, you know, I'm spending most of my time, I'm on the like, uh, Dan Dan noodle CEO track- ... where I'm just hanging out making noodles with cool people now. And like we do a lot of workshops with customers.

**Alan** [11:14]
Yeah.

**Dex Horthy** [11:14]
Like what we're doing is like there's a product that helps you do it-

**Alan** [11:17]
Mm-hmm

**Dex Horthy** [11:17]
... but there's also just like a lot of like whiteboarding that goes into helping people like ship more code with AI-

**Alan** [11:22]
Gotcha

**Dex Horthy** [11:22]
... which is like we're focused on helping people ship like... How, how do we get coding agents to solve like hard problems in complex code bases?

**Alan** [11:29]
Yeah.

**Dex Horthy** [11:29]
And so that is, uh, you can automate a lot of that with a product, but there's still like, just like trainings and stuff. So I, I run those, and I own those, and like I travel around and like sit with our customers and help them like level up their engineering teams, in addition to like giving them a product that kind of like when I leave, they have kind of like guardrails and guidelines for how to do it.

**Alan** [11:49]
Gotcha. And are your customers mainly like more upstream, bigger enterprises, or are they also like startups, a bit smaller mix?

**Dex Horthy** [11:56]
Uh, we're doing a little bit of both.

**Alan** [11:57]
Okay.

**Dex Horthy** [11:57]
You could argue that we should pick one and focus a little bit.

**Alan** [12:00]
Yeah.

**Dex Horthy** [12:00]
Um, but yeah, at the moment it's really, it's a mix of like, uh, you know- 10 to 20 person engineering teams at, like, a bunch of, bunch of random, like, YC startups.

**Alan** [12:10]
Yeah.

**Dex Horthy** [12:11]
And then, like, a mix of, like, you know, 1,000 engineer, like, publicly traded companies that, like, have to figure out how to, like, standardize and adopt AI across their whole company.

**Alan** [12:21]
Yeah, okay, that makes sense. Your meat should also probably be cooked by this point. Um, you should take it off and turn off the stove when it's got a color, like a golden brown color.

**Dex Horthy** [12:30]
I need a little more color on this.

**Alan** [12:31]
Okay.

**Dex Horthy** [12:31]
I'm not, I'm not happy with it yet, but-

**Alan** [12:33]
Yeah. But what-

**Dex Horthy** [12:33]
... keep, keep, keep feeding me instructions, I will, uh, I will queue the messages.

**Alan** [12:38]
Okay, yeah, so once you have that just put it into your main plating bowl and then we'll-

**Dex Horthy** [12:43]
The main what?

**Alan** [12:44]
Plating bowl, the big white bowl.

**Dex Horthy** [12:46]
Okay.

**Swyx** [12:47]
Show the camera the plating bowl.

**Alan** [12:48]
Yeah, it should be behind.

**Dex Horthy** [12:49]
Yeah, see this? This is not, this is not colored enough. We're gonna get a little more color on here.

**Alan** [12:54]
We could talk more about your favorite, you know, context engineering. How have you seen the craft of context engineering change over the past, I guess, year? We had much smaller context, uh, windows, uh, but then now we have a lot larger ones, and so probably the needs change.

### Context Engineering Evolves

**Alan** [13:12]
Have there been, like, big things that you've noticed?

**Dex Horthy** [13:15]
Well, so I think there's two competing things here, right?

**Alan** [13:16]
Yeah.

**Dex Horthy** [13:17]
There's this idea of, like, as the models get better you don't have to do as much context engineering to get the same quality of results.

**Alan** [13:23]
Mm-hmm.

**Dex Horthy** [13:23]
Um, like I saw a lot, like, a lot of my... People who used to be, like, my favorite engineers that I looked up to, and I still look up to them, but they were just like I thought about them a lot more in, like, 2015.

**Alan** [13:32]
Yeah.

**Dex Horthy** [13:32]
People who were, like, OG engineers on Docker and people like Mitchell Hashimoto. Um, suddenly there's this arrive of, like, all of the OG, like, DevOps people when Opus 4.5 came out they're like, "Oh, this is good enough. This is a big change."

**Alan** [13:46]
Mm-hmm.

**Dex Horthy** [13:46]
And I saw the things that they were shipping and I was like, okay, but I know a bunch of engineers who got really good at context engineering, and they were getting the same results from Opus 4.0 and Opus 4.1.

**Alan** [13:56]
Gotcha.

**Dex Horthy** [13:56]
And so, like, even as the models get smarter, I wonder, like, those people who were getting, like, really good results back in the summer, what are they doing now with Opus 4.5? What have they unlocked now that it's not being like, "Oh my God, Opus 4.5 is AGI"?

It's like, Opus 4.5 is AGI, uh, for people who didn't want to put in the time to learn how to do it.

**Alan** [14:15]
Mm-hmm, mm-hmm.

**Dex Horthy** [14:16]
You know what I mean?

**Alan** [14:16]
Yeah. So would you say a lot of these models are kind of capped by the users or an enterprise's ability to kind of properly context engineer?

**Dex Horthy** [14:25]
Um-

**Alan** [14:25]
Or do you think we'll kind of get to a point where there won't really need to be thinking about what goes into the context of your model with such huge context windows?

**Dex Horthy** [14:33]
So there was a thing I said in my first AI engineer talk-

**Alan** [14:35]
Mm

**Dex Horthy** [14:36]
... that was based on, like, an episode where, like, the latent space guys, they interviewed, uh, the NotebookLM team.

**Alan** [14:42]
Yeah.

**Dex Horthy** [14:43]
And, uh, I think it was Usama said this thing that was like, basically, like, there will always be a thing that the model can, like, only kind of get right reliably. Like, you find a thing that's right on the boundary of the model's capabilities, and you figure out how to get it right over, and over, and over again.

**Alan** [14:58]
Yeah.

**Dex Horthy** [14:59]
And, like, that's how you build incredible AI experiences.

**Alan** [15:02]
Gotcha.

**Dex Horthy** [15:03]
And I think that's still true, it's just, like, that, that, that frontier of what's possible is constantly shifting outward.

**Alan** [15:08]
Mm-hmm.

**Dex Horthy** [15:08]
And so, like, it's still, it's still reasonable to know how to do these things.

**Alan** [15:13]
Yeah.

**Dex Horthy** [15:13]
Um, because sometimes, some- at some point you might hit a problem that the AI can't solve, and then you have to go figure out how to solve it reliably anyways.

### The Dumb Zone

**Alan** [15:21]
Yeah. No, that's fair. You mentioned, I think in the most recent AI engineering talk about, was it the dead zone? Where it's like after you've-

**Dex Horthy** [15:27]
Oh, the dumb zone.

**Alan** [15:29]
Yeah. With 40% right of context-

**Swyx** [15:31]
Bring me the dumb zone right here.

**Alan** [15:31]
The dumb zone.

**Dex Horthy** [15:31]
I'm the dumb... Dude, I, I know the dumb zone 'cause I've spent a lot of my life in the dumb zone. Let's put it, let's just, let's just be clear here.

**Alan** [15:38]
So I guess for, like, people who want to do simpler tasks or more advanced tasks, are there, like, heuristics that you advise either, like, through your work with HumanLayer or just in general on when they should kind of think about taking more time to engineer it versus just, you know, putting everything in context and not caring about, like, over fitting this whole space?

**Dex Horthy** [15:57]
Yeah, I mean, I, I love that question. Um, I think a thing, I tweeted about this a, a couple of weeks ago-

**Alan** [16:02]
Mm-hmm

**Dex Horthy** [16:03]
... is, like, the idea with the dumb zone is also, like, it's kind of meant to be a rule of thumb. Like-

**Alan** [16:07]
Gotcha

**Dex Horthy** [16:08]
... there are people I know who have been doing AI engineering for a long time who regularly push all the way to the end of the context window-

**Alan** [16:15]
Mm-hmm

**Dex Horthy** [16:15]
... because they kind of, like, know what they're doing and what they want and, like, sometimes a good trajectory is worth... Like, you'll actually keep getting better results 'cause your trajectory is really good.

**Alan** [16:24]
Yeah.

**Dex Horthy** [16:24]
So it's like if you don't know what you're doing and you don't really know what the AI model is capable of, and you don't have a lot of experience, like, you know, training wheels is like when you get to 40% start thinking about wrapping it up or like-

**Alan** [16:36]
Mm-hmm

**Dex Horthy** [16:36]
... like, doing a, like, you know, intentional steered compaction to where you want to be.

**Alan** [16:40]
Yeah, compaction is-

**Dex Horthy** [16:41]
But, like, depending on the model and what you're doing, and if it's really simple, like, I will regularly blow out to, like, 60, 70% of the context window used. Um, but it's like-

**Alan** [16:50]
Mm-hmm

**Dex Horthy** [16:50]
... if you don't know and you haven't developed the intuition, then, like, I tell people, like, shoot for 40% if you're learning.

**Alan** [16:56]
Gotcha. And so, like, how does someone even, like, get familiar and build this intuition? Like, if somebody came up to you and says they want to take this more seriously and don't really know where to start, and even when it comes to compaction, like, how to even assess, like, accuracy, or relevance, or what to even consider, and how to even judge whether compaction is useful, um, what would you advise them?

### Building Intuition

**Alan** [17:19]
Would you tell them to take specific steps or, you know?

**Dex Horthy** [17:21]
Uh, yeah, talk to Claude for 70 hours a week. That's, that's how you do it. You gotta build intuition. Uh-

**Alan** [17:27]
Just use AI as much as possible.

**Dex Horthy** [17:29]
Use as much AI as possible, make mista- I mean, like, the thing that people actually ask me is, like, we developed this framework called, like, research, plan, implement.

**Alan** [17:35]
Yeah.

**Dex Horthy** [17:35]
And I get this question a lot of, like, what is, like, how do I know when to use research, plan, implement versus, like, just talk to Claude, right?

**Alan** [17:44]
Mm-hmm.

**Dex Horthy** [17:44]
There's some issues that are really small and, like, it doesn't really warrant the whole, like, process of like, okay, we're making a one-line change. The only way you learn how hard the problem is for AI is, like, you're gonna do too much sometimes.

You're gonna be like, "Oh, that was way too much attention," and now, like, I, I wasted a bunch of time on a thing that I could've just talked to Claude on.

**Alan** [18:05]
Yeah.

**Dex Horthy** [18:06]
And then you'll learn and be like, "Okay, next time I'm gonna go to the other side." And sometimes you'll, like, go talk to Claude and Claude just, like, flails and can't fix it 'cause it's too big of a problem-

**Alan** [18:13]
Yeah

**Dex Horthy** [18:13]
... or the context wasn't curated well enough. And so it's like the only way to know how much context engineering to use on a given problem, I think, is- Um, you just have to get some reps.

**Alan** [18:24]
Yeah.

**Dex Horthy** [18:24]
I don't, I don't have a better answer than that. I'm, I'm sure the model labs love that that's the answer.

**Alan** [18:29]
Mm-hmm.

**Dex Horthy** [18:30]
Uh, but, uh, yeah, that's, that's my biggest advice, is like put in the reps and figure out what's possible, and like try to push the boundaries.

**Alan** [18:36]
Gotcha. Oh yeah, that's helpful. Definitely just using a lot more reps, kind of like if you're going to the gym and trying to work out a muscle group.

**Dex Horthy** [18:43]
Yeah, the, uh, what do we, what do we call it? Like, the mind-model connection.

**Alan** [18:46]
Yeah, the mind- ... mind-model connection. I guess like on that, do you have predictions going into like '26, '27, even the next, you know, couple of years about how this industry will change that are top of mind?

### Future of SWE

**Dex Horthy** [18:59]
Uh, I mean, I think I s- I mentioned this before is like the software engineering role is gonna keep evolving. I don't believe the like software engineering is dead and there will be more, no more, no more coders.

I think the, the way I would describe it is like the role of the software engineer will change from like write working code to like produce working code.

**Alan** [19:20]
Mm.

**Dex Horthy** [19:20]
Or like cause working code to be produced.

**Alan** [19:23]
Gotcha.

**Dex Horthy** [19:24]
And one of my biggest pet peeves is the people who run around Twitter talking about how much like Claude code they spent and how many tokens they did and how parallel their workflows are-

**Alan** [19:31]
Yeah

**Dex Horthy** [19:32]
... and don't talk about like, "Yeah, I shipped a reliable product to people that like it, and like maybe even are paying for it." But it's like stop. We need to talk less about how much like code we produ- like how much code we produce and how much we actu- Like Guillermo Rauch talks about this.

He's like-

**Alan** [19:45]
Yeah

**Dex Horthy** [19:46]
... "There's a difference between coding and shipping."

**Alan** [19:47]
Mm.

**Dex Horthy** [19:48]
Coding is like making the software work and maybe making it work on your la- on your workstation. Shipping is like getting it into prod, fixing the things that are broken, maintaining it over time, and like continuing to make it better.

**Alan** [20:00]
Gotcha.

**Dex Horthy** [20:00]
And like that part is not quite solved by AI yet.

**Alan** [20:03]
Yeah.

**Dex Horthy** [20:03]
Maybe we'll get closer, but like I'm always looking... I mean, our company's called HumanLayer-

**Alan** [20:07]
Mm-hmm

**Dex Horthy** [20:07]
... 'cause we used to do human-in-the-loop.

**Alan** [20:09]
Mm.

**Dex Horthy** [20:09]
But now it's a little bit more like I think about it as like what is the high leverage things for humans to do?

**Alan** [20:14]
Yeah.

**Dex Horthy** [20:14]
And like what is the things that we can like leverage AI for, and it's kind of a dance. There's like if you're making code, there's like parts of it that AI just can't do right now, and there's parts of it that it can do really well.

**Alan** [20:24]
Gotcha. Yeah, so to complement, sounds like quality over quantity, especially with all the, you know, tweets about-

**Dex Horthy** [20:30]
Yes

**Alan** [20:30]
... multiple Claude code instances and just shipping things.

**Dex Horthy** [20:34]
Well, it's like we're working with like big enterprises in like fintech and like other like spaces where it's like

we can't afford to get it wrong.

**Alan** [20:44]
Yeah.

**Dex Horthy** [20:44]
There's no ship fast and break things.

**Alan** [20:46]
The stakes are much higher. Mm-hmm.

**Dex Horthy** [20:46]
I mean, though, like it's great for all the indie hackers and vibe coders that they're shipping a bunch of Claude code stuff.

**Alan** [20:51]
Yeah.

**Dex Horthy** [20:51]
But it's like how do we make this stuff accessible and make it so that like real production-grade products can get access to the same gains?

**Alan** [20:58]
Okay. I'm in the assembling stage right now. I'll give you like another minute.

**Dex Horthy** [21:03]
I didn't realize we were on the clock, dude. All right.

### Plating

**Alan** [21:05]
Yeah. But no rush, of course, Dex.

**Dex Horthy** [21:07]
And then how are you, how are you straining these noodles out?

**Alan** [21:10]
I'm just taking a fork and just transferring it over.

**Dex Horthy** [21:12]
Oh, no. I tried to set this on 7 and it is still on 10. All right, we're gonna take this off. This is getting chaotic, dude. This is fun. Let's go.

**Alan** [21:20]
All right, whatever.

**Dex Horthy** [21:21]
All right, we're gonna just do a little sauce riff here.

**Alan** [21:23]
Great. All right.

Final plating.

**Dex Horthy** [21:30]
My plating is not gonna look as good as yours- ... but we're gonna figure it out.

**Alan** [21:34]
Yeah. Taste is all that matters.

**Dex Horthy** [21:36]
This is why I love cooking for myself- ... 'cause I like, I don't wanna have to care what it looks like. I'll eat, I'll eat the ugliest slop in the world if it's delicious.

**Alan** [21:45]
All right. Food's all plated?

**Dex Horthy** [21:46]
No.

**Alan** [21:47]
Nope?

**Dex Horthy** [21:48]
Let me just wipe this off. Let's get that fork out of there, and we are ready to rock.

**Alan** [21:55]
Great. Let's rock.

**Dex Horthy** [21:57]
Let me turn this off.

**Alan** [21:58]
Smell.

**Dex Horthy** [21:58]
It's beautiful. Look at that .

### Taste Test

**Alan** [22:02]
Great. How was that, Dex? Did you have fun?

**Dex Horthy** [22:04]
Uh, it was a little chaotic, but it was a good time. I'm excited to try these things.

**Alan** [22:07]
Yeah, it was definitely a rush during the end, but we both finished, so both of our Dan Dan noodles.

**Dex Horthy** [22:12]
All right. Let's have a look.

**Alan** [22:12]
Let's see. Okay. Wow, yours is a lot prettier than mine.

**Dex Horthy** [22:16]
I, yeah, I-

**Alan** [22:16]
Well, I would say least

**Dex Horthy** [22:16]
... at the last minute I decided I was gonna try to, try to present it a little bit.

**Alan** [22:20]
We should give it a try. Here's a fork for you.

**Dex Horthy** [22:22]
Let's give it a taste. Are you gonna, are you gonna try mine or we're gonna-

**Alan** [22:24]
Yeah, let's, let's try yours first.

**Dex Horthy** [22:25]
Okay.

**Alan** [22:25]
Then we can try mine after. Here. Cheers.

**Dex Horthy** [22:26]
All right. Cheers.

**Alan** [22:32]
Mm. The bark chu was done very well.

**Dex Horthy** [22:34]
Really?

**Alan** [22:36]
Yeah. It's like-

**Dex Horthy** [22:36]
If you had given me... If you let me take my time I probably would've overcooked it.

**Alan** [22:39]
Mm-hmm.

**Dex Horthy** [22:40]
I think I used too much, I threw in some paste at the end. I feel like I used too much-

**Alan** [22:43]
Oh, okay

**Dex Horthy** [22:43]
... of the, the sesame paste.

**Alan** [22:44]
Yeah, I think yours is definitely a little spicier, but let's see how mine... Mine's definitely a little more thick.

**Dex Horthy** [22:49]
Let's give this a shot.

**Alan** [22:50]
Great. Cheers.

Mm.

**Dex Horthy** [22:56]
Damn. Okay. I really like the like... Something about this one is like it's more well-rounded. It just like the flavors all come together in a nice way.

**Alan** [23:08]
Yeah, I think this is like-

**Dex Horthy** [23:08]
This one feels a little sharper.

**Alan** [23:09]
Yeah. Let's try the rougher.

**Dex Horthy** [23:11]
Yeah, I'm definitely like a follow the recipe once guy.

**Alan** [23:14]
Yeah.

**Dex Horthy** [23:14]
And then-

**Alan** [23:14]
I also usually just-

**Dex Horthy** [23:15]
... and then, and then, and then riff it myself. 'Cause like now I'm like I made this, so I was like, okay, I can think of like four things I would change next time.

**Alan** [23:20]
Yeah. Cheers.

**Dex Horthy** [23:27]
I feel like this-

**Alan** [23:28]
Mm

**Dex Horthy** [23:28]
... one just has less flavor than both of them.

**Alan** [23:29]
Yeah.

**Dex Horthy** [23:30]
It's just a little more-

**Alan** [23:31]
I, I agree

**Dex Horthy** [23:31]
... like mild.

**Alan** [23:32]
Like I, I think this is like very like safe and-

**Dex Horthy** [23:35]
Yeah

**Alan** [23:35]
... you know, not as-

**Dex Horthy** [23:35]
If I was making food for other people I would make that and then I would put the chili oil on the side-

**Alan** [23:39]
Yeah, yeah

**Dex Horthy** [23:39]
... and be like make it, make it, make it, uh, make it more fun if you want to kind of thing.

**Alan** [23:42]
Yeah. Add a little extra spice in there.

**Dex Horthy** [23:46]
So this one's mine, this one is Alan's, and that's the reference one.

**Alan** [23:49]
Yeah.

Any call-outs? Um, where can people find you?

**Dex Horthy** [23:55]
Uh, I'm on, I'm on Twitter @dexhorthy. Uh, you can get my weird, uh, unfiltered brain trace as I go through the world of building agentic IDEs and playing with coding agents and keeping up with all the new things that all the labs are shipping.

**Alan** [24:10]
Yeah.

**Dex Horthy** [24:10]
Uh, and then we are, we just rebuilt our product from scratch. It's an agentic IDE. It's-

**Alan** [24:15]
Exciting

**Dex Horthy** [24:15]
... uh, coming out soon, and, uh, you can go sign up at humanlayer.dev, and, uh, we'll shoot you a note when it's ready.

**Alan** [24:21]
Yeah. Great. Well, back to our judges. Are you guys ready to say?

**Dex Horthy** [24:25]
Yeah, I want to, I want, I want the verdict here.

**Swyx** [24:27]
So I give Dex the slight win here. It, s- they, they all used the same ingredients-

### Verdict

**Alan** [24:31]
That's cool

**Swyx** [24:31]
... so it's like-

**Dex Horthy** [24:31]
Yeah

**Swyx** [24:32]
... you're, you're gonna be very comparable. But I just like the texture and the flavor. And also I, I really like the, the extra protein. Like you just like went all out with like the flavor. I love-

**Dex Horthy** [24:41]
I took a bunch of noodles out. I was like I want, I want more pork.

**Alan** [24:44]
Yeah.

**Dex Horthy** [24:44]
I want the ratio. I want protein.

**Swyx** [24:45]
Yeah. The, the ratio is just very different, and I like that.

**Dex Horthy** [24:47]
Okay.

**Swyx** [24:48]
Um, I, I have to agree. Dex's is more meatier.

**Dex Horthy** [24:51]
Aw.

**Alan** [24:51]
Yeah.

**Swyx** [24:52]
And-

**Dex Horthy** [24:52]
Damn, that's the-

**Swyx** [24:52]
Sorry, Alan

**Dex Horthy** [24:52]
... is that-

**Alan** [24:53]
Let's go, Alan. Oh, yeah.

**Dex Horthy** [24:53]
Yeah. Win, win any cooking competition with this one weird trick. Just make more meat on the plate.

---

This library is powered by PodHood (https://podhood.com), the podcast website platform.