# AI Slop in the Industry (Ep 103)  ## Episode metadata - Episode title: AI Slop in the Industry (Ep 103) - Show: Community Pulse - Owner / Host: Mary Thengvall - Episode publish date: 2026-04-01 - Episode AI description: They dig into low-quality AI-produced developer content and why it erodes trust. They debate who AI content really targets and how answer-engine optimization shifts priorities. They talk layoffs, ethical choices by companies, and security risks from AI-generated code. They close by arguing human curation and communities are vital to counter algorithmic, agent-focused content. - Mentioned books: [I, Robot](https://share.snipd.com/book/b7b9de47-a2da-4202-9e7a-2776734238c1) by [Isaac Asimov](https://share.snipd.com/person/17cf51b6-0ffd-49b4-a18d-3709b3de9650) - Duration: 36:01 - Episode URL: [Open in Snipd](https://share.snipd.com/episode/e0ca2cf0-3f45-4f95-95bb-63429bbff340) - Show URL: [Open in Snipd](https://share.snipd.com/show/558ab5f5-a9ce-46bc-aebb-9b6a42ad68ed) - Export date: 2026-07-08T01:13:39 ## Snips ### [AI Content Is Being Built For Algorithms Not Developers](https://share.snipd.com/snip/6d607767-3658-4f2e-b1d0-4fcae6f91391) 🎧 00:38 - 04:12 (03:34) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/fd52d6bd-8e2a-45f3-a621-be50917a4bfb" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - AI-generated content is mainly produced to feed algorithms and agents rather than help human developers. - Jason argues marketing pushes low-quality, unattributed AI 'slop' into LLMs so brands show up in generative answers, decoupling content from developer trust. #### 💬 Quote > It is made to feed the algorithm. why SEO is being replaced by AEO or GEO, which is answer engine optimization as opposed to GEO is generative answer or generative engine optimization. > — Jason Hand Jason Hand on why AI content targets search/answer engines #### 📚 Transcript **Mary Thengvall:** So to set the stage, there have been a few pieces of content that have come out recently that were clearly AI developed that were less than of a quality that I would consider good and also had information that was incorrect or at the very least misleading. And my concern, what kind of started this off was if we're giving out bad information, is that going to hurt our standing in the world of developers? To which Wesley **Wesley Faulkner:** said. I said that we are not the customers of this type of creation, this content. And also I want to add that that document was also unattributed. So it didn't give any ups or references to even back up what it was saying and did not help with this authenticity by eliminating who might have actually given valuable data that this LLM sucked together. And when we look at what happened, like at block, where they fire 40% of their employees because of AI, whatever dominance. But then I don't know if you saw this post, where someone who did not get fired, did not get laid off was then offered like a huge bonus that was equal to like 90 of her salary to to stay i did not see that there's a they try to give retention bonuses to everyone who who was left behind uh and this person decided not to take it and to quit because they said they work. Their role was working within departments and 70 percent of the people that they worked with were gone and they didn't want to stay for that carnage. **Mary Thengvall:** So that's that's a whole interesting other take on an ethical stance against **Wesley Faulkner:** AI **Mary Thengvall:** slot firings. **Wesley Faulkner:** Yeah. And the reason why I want to bring this up is that people mean nothing. Customers mean nothing. There was a huge firing of 16,000 at AWS. And I knew several in the community space that no longer have jobs, which means that the people that they served are also devalued. the reason why is because if you think about marketing, which I think if we're talking about what content is used for in the developer space, it's for content marketing, not necessarily for developers, it is made to feed the algorithm. why SEO is being replaced by AEO or GEO, which is answer engine optimization as opposed to GEO is answer or generative engine optimization. These new terms are big because that's the main marketing channel is to pump it into those things, have those be a signal for those things. Don't need the references and the authority because all that's obfuscated when it's pushed into either a terminal or into these generative systems where they just like, we have enough signal. So this is not noise, which is wrong. And I think that is the main thing. It's just like, how do we get in front of developers? --- ### [Continuously Update Your Mental Model Of AI Tooling](https://share.snipd.com/snip/c23a7fa6-ad74-40bd-bba4-9e75bce6d862) 🎧 06:22 - 10:08 (03:46) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/cdaf25b1-011c-43ba-8f94-4c936b00aebe" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Stay current with AI tools and models to understand real developer usage and avoid outdated mental models. - Jason describes a Datadog hackathon where experimenting with agents, models, and production use showed practical developer adoption and benefits. #### 💬 Quote > Unless you've been messing with AI here in the last month or so, you're, you're just blatantly, uh, uninformed. > — Jason Hand Jason Hand on the rapid pace of AI change #### 📚 Transcript **Jason Hand:** In fact, this week, uh, I was just sharing with you both. We had a hackathon internally within our advocacy to explore lots of different things, mostly to up level ourselves on the different tools that are available to us, which is pretty much anything we want at Datadog. Like there's just there's lots of different things. There's agents, there's models, there's anything you need. And how can we use these tools to help do our day-to job and level ourselves up? can we use these tools to just understand where development is, what software development looks like in March in 2026? And then relay that on out to other developers out within the community. That's kind of our role too is like, what kind of an advocate would we be if we don't know the latest and greatest of how software is made you know like that's that's sort of what goes through my mind is like i can't just ignore this thing and think it's a trend there are developers who really do use this stuff i know pj you point out there's people who are hesitant and for sure you know like it's a spectrum but i can tell you for sure at least the datadog like we're all in using it very heavily in production. **Mary Thengvall:** And this, this begs the question though, if you're saying you're trying to use it to see how the average developer uses it, but you're using it heavily in production, but **Jason Hand:** the average developer is **Mary Thengvall:** not using it heavily in production. **Jason Hand:** We use it. I'm saying as most people will say, well, you're not using it in production. I'm here to say, no, we use it at every, if we just use it, like it's, it's being used. It is a, it's a, it's a co-pilot for all of us. know, the, all just AI in general is making all of us more, you know, of an improved version of ourselves and more efficient, more accurate, more improved across the board. So there's no argument there that that's happening. I think that there's an echo chamber that gets created when people who are, you know, have one experience and they go out there and say, this is how it is. And this is what I'm experiencing. And then people take that, you know, as like the gospel and then they, but, but I do think that there's enough of us, myself included, who recognizes that that's just not true. Maybe that's their reality, but unless you've been messing with AI here in the last month or so, you're, you're just blatantly, uh, uninformed, like whatever, whatever, like mental model you've created about this stuff, uh, you know, pre February 1st, I would say somewhere around there, February 9th, I remember what, but like, you're just wrong. Uh you got to go back and revisit like what's happening in the world within AI like very regularly just to just to know like what the state of things are and you're right with the the AIEO or AEO or whatever like that stuff is out there I feel like there are people the agents are writing stuff for other agents on the web. I showed you my hackathon is related to those LLMs texts. The, the inner, uh, the internet is becoming increasingly just agents talking to agents, you know, and we're like, kind of like, where do we exist in all this anymore? Are we the ones that are consuming this content online or is it just other agents and other LLMs and other like, you know, algorithms consuming it and making it gospel? And, um, you know, some of that I think comes down to us. I'm very positive that there's going to be a lot of slop and a lot of crap on the internet and there's got to be a way for the good stuff to make it to the top, you know, and whether it's Google or some others, they're going to put weight on things that show they're, you know, they're supported, I would think by the community, some objectively way, objective way. You know that this is bad and this is good based off of, you know, like it can be done. Um, and I think that's going to be the stuff that makes it into like gospel of AI. --- ### [Community Signals Could Rescue Quality From Agent Noise](https://share.snipd.com/snip/cf83398c-2b84-440b-bd96-832fb76325d4) 🎧 09:58 - 12:34 (02:36) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/cae7ef43-e291-456f-9fe5-7bd34897a929" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Good content can still surface if objective community signals are used to weight trust over agent-generated noise. - Jason hopes major platforms will favor community-supported, verifiable content rather than short-term SEO/agent gaming. #### 💬 Quote > I feel like there are people the agents are writing stuff for other agents on the web. The inner, uh, the internet is becoming increasingly just agents talking to agents. > — Jason Hand Jason Hand on agents consuming agent content #### 📚 Transcript **Jason Hand:** You know that this is bad and this is good based off of, you know, like it can be done. Um, and I think that's going to be the stuff that makes it into like gospel of AI. And we don't have to worry about these like posers or grifters who have never done the job and are just trying to put stuff out there thinking it's going to help themselves somehow in a short-term way. And it's really damaging to the industry. I guess I'm, I'm kind of hoping they're going to get kind of just shoved out because it just, it just, it's not going to. **Mary Thengvall:** I think that's a bit optimistic though. **Jason Hand:** I know. Yeah. **Mary Thengvall:** You hope that you hope the big guns like Google are going to do the ethical thing when we know for a fact they do. Well, what's **Jason Hand:** in it for them to put out the wrong information when we're all looking for the correctness? I think correctness is there. Do pay for it? **Mary Thengvall:** It's the same answer. Why did Google remove the first we shall do no harm concept from what they do? Don't Yeah, don't be. Why did they stop saying don't be evil? Because evil pays money. **Wesley Faulkner:** Blue Sky and Mastodon have already demonstrated that without this tipping the scale curation of this agentic or algorithmic feed, that I'm still able to see things from people I care about. And maybe not in the mainstream way. But there are spaces, especially like even like the DevRel Collective, like we share articles that we've seen. So it's a human curated space rather than an algorithmic one. And so those will exist if you are a member, know where to find these communities, know how to join these communities, know how to participate in these communities. And so, but then the mainstream area, the LinkedIn's, the threads the, the, even the Twitter, just say Twitter. X. Um, the, that stuff is still going to be probably prominent there. Like, even if you type into an LLM right now, make a post for X or whatever, it says, Hey, would you like to go viral? You want me to tweak it for its engagement? And it'll just do that for you and so i i already am seeing this thing where it's like everything is trying to be the best though the oh you can't believe this happened is it's the it's the like link bait like engagement stuff for almost all these mainstream posts and it's going to be it's going to people who's educated and know about this stuff that's going to be able to find the places where they can find the right information. And it's that same old, tired and true statement is that the future is here. --- ### [Enable Developers Internally Or Accept Falling Behind](https://share.snipd.com/snip/f7c2a32d-2a1f-4801-9ba4-a8367db5259e) 🎧 12:49 - 14:06 (01:17) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/a4fbcbfb-79f4-4eac-a883-28650576d2ef" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Advocate for enabling developers inside organizations to use AI tools or risk falling behind. - Mary warns most companies lack Datadog-like budgets and that access gaps create 'only-the-rich' participation in AI advantages. #### 💬 Quote > Most companies aren't going to have that. Very few companies are going to have that, in fact. > — Mary Thengvall Mary Thengvall on unequal access to AI resources #### 📚 Transcript **Mary Thengvall:** So only people with means could participate. This is very similar. Like the average person, sure, the average person, let's put the average person aside for a second. We talked about developers. The average developer company does not have a portfolio like Datadog where they can say, we're going to spend hundreds of thousands of dollars on tokens so that our people could use the best of the best. Most companies aren't going to have that. Very few companies are going to have that, in fact. So you already have a only the rich can participate situation. On top of that, the lack of guardrails, and we've seen that this week where your Anthropic was like, no, I will not accept, I'll say no to your defense contracts because we're not dropping our guardrails. And immediately OpenAI and ChatGPT were like, we're totally cool with losing all ethical balance just to make a dollar. And those are the two largest AI companies in the world right now, arguably. Like, so it's, it's kind of like, do we want to, can we survive to a point where this optimistic future that Jason talks about works out or will it be a continual division of if I'm not working for a major, corporation or super large business, I won't be able to ever get the tools that will allow me to advance my career. Well, **Wesley Faulkner:** I don't think it's one or the other. --- ### [Claude Turned Speaker Notes Into Overexplained Slides](https://share.snipd.com/snip/85cc8e1e-a7a2-485b-b9d5-fca14149ee45) 🎧 14:25 - 15:18 (00:53) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/6ef1f36b-4727-4b49-b0de-932042b8de56" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Mary used Claude to 'punch up' a talk and it copied speaker notes onto slides and over-explained jokes. - The result felt unhuman and highlighted how some assistants rewrite content in awkward, incorrect ways. #### 💬 Quote > It then proceeded to put all of my speaker notes onto each slide, then explained to me what the jokes were that I had written. > — Mary Thengvall Mary Thengvall recounting a tool editing mishap #### 📚 Transcript **Mary Thengvall:** And one of the things that Luma does is that very thing that you mentioned, Wesley, which is, hey, I see that you're writing some stuff. Do you want me to perk this up for you? Do you want me to punch it up? So I was like, okay, like I'm going to do, so I did a little bit of testing. I was like, yeah, go ahead and punch it up. Cause all it really says is like, Hey, the, the meeting, the, the dinner we're having is at this place. Uh, these are your, your meal options, you know, whatever it's, BizCat, whatever, that you would usually put on an invitation. And it rewrote it in a way that was so unhuman. And I don't mean inhuman, it was unhuman. Like it was something no one would ever say. Then I was like, that's preposterous. So then I decided I'm in the middle of writing a talk, I'm going to take the talk and ask Claude to punch it up for me. It then proceeded to put all of my speaker notes onto each slide, then explained to me what the jokes were that I had written. **Wesley Faulkner:** Oh, I need that so I could understand. Incorrectly. I'll definitely send that over to you. --- ### [AI Will Create An Agile Versus Legacy Company Divide](https://share.snipd.com/snip/7dd70901-fc55-4ee6-8234-fd3b19608ebb) 🎧 17:30 - 19:26 (01:56) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/5b63d10c-3229-46e3-aa74-a8744f12af9e" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - A bifurcation will form between AI-native, agile companies and slow-moving incumbents with heavy compliance. - Wesley outlines how compliance cycles make large firms miss rapidly iterating foundational models, advantaging nimble players. #### 💬 Quote > By the time that something goes all the way through their quote unquote process, they're not using the latest and greatest because they're coming out with a brand new foundational model every three months. > — Wesley Faulkner Wesley Faulkner on corporate adoption lag #### 📚 Transcript **Jason Hand:** years if they're not enabling their developers today. It's just a fact. **Mary Thengvall:** I wish that were true. I don't think that's true. Okay. I'll **Jason Hand:** stand on that for sure. I **Mary Thengvall:** think that **Wesley Faulkner:** what you're hinting to is like whose companies are enabling developers, their developers, and training their people to do better. And I think there's an ignorance about how people learn in the developer journey from companies, including internal developer relations. And we are going to start seeing a bifurcation in an arms race where those companies that don't teach their own developers how to use these tools will, I won't say be left behind, but will be a few steps back. And the scary part of this right now, we're seeing that the corporate dominance of even an anthropic say, no, we don't need a government contract or whatever, we'll step away is only going to get worse because when you think about big corporations, they have compliance loops to go through. They have some security loops. They have all this stuff in order to adopt these tools. And by the time that something goes all the way through their quote unquote process, they're not using the latest and greatest because they're coming out with a brand new foundational model every three months. And so companies are going to need to find a way to be more agile with their either inside their regulatory framework or these neo companies, these ones that are educated or these, let's call them AI native companies, are going to be the new dominant players. And what does that look like when you have new, like when you wipe the board of the top players for some of these things, and then you replace them with like three years ago, companies we never heard of? --- ### [Treat LLMs As Databases Not Reasoning Engines](https://share.snipd.com/snip/765edeef-eaaa-4038-acab-209e0bbc644a) 🎧 22:31 - 23:42 (01:10) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/f8b5327e-b1cd-4fdf-89ba-ed959848135e" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Treat current LLMs as very large synthetic databases, not reasoning minds; verify outputs rigorously. - Mary advises calling them 'toddlers' that echo training data and says they can't truly reason or replace expert validation. #### 💬 Quote > These things don't reason. They don't think. They're not intelligent. They're easier ways to access a database. > — Mary Thengvall Mary Thengvall on the limits of current AI reasoning #### 📚 Transcript **Mary Thengvall:** These things don't reason. They don't think. They're not intelligent. They're easier ways to access a database. **Wesley Faulkner:** You can't do that. I mean, it will never go away because these top companies need that word to be the thing. **Mary Thengvall:** Well, yeah, to sell the thing. I mean, **Jason Hand:** we've been talking about AI forever. I was delivering talks on AI like in 2018. Like this isn't. And and and what I mean is like LLMs is part of AI is like a part of the category. So like people have not been educated enough along to know the distinction between the two, you know, which is between all the different elements within the greater AI category. So it all gets rolled up. Oh, sure, sure. It's confusing. **Mary Thengvall:** And I give a bit of that, but it's when you say things like reasoning and thinking, it's not. What these things are is you have the ability to make a database query, which is, you know, an LLM is essentially an extremely large scale database. And you can do that without writing a SQL statement. It's synthetic. what AI is. **Jason Hand:** Yeah. It's emulating human reasoning in a synthetic. **Mary Thengvall:** It's not emulating human reasoning. It's literally just saying the thing that has been told. --- ### [AI-Generated Vibe Code Returned Hundreds Of CVE Warnings](https://share.snipd.com/snip/18561b61-6d6b-4f8a-827d-b64e51517aab) 🎧 31:00 - 31:32 (00:32) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/c006e607-09b4-4dce-b86b-2e46823d81c4" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - A public GitHub repo's 'vibe-coded' examples returned hundreds of CVE warnings that Dependabot couldn't fully fix. - Mary observed AI-generated front-end code with massive security issues, showing real risks from one-shot generation. #### 💬 Quote > The Vibe coded stuff was the only stuff that came back with like over five, like nothing else had security risks. This had 500 CVE issues that had to be dealt with by Dependibot. > — Mary Thengvall Mary Thengvall on security problems in AI-written code #### 📚 Transcript **Mary Thengvall:** We put all of our example code in the public GitHub repo. The Vibe coded stuff was the only that came back with like over five, like nothing else had security risks. Everything else was written in Python. This had 500 CVE issues that had to be dealt with by Dependibot. And Dependibot was even like, I can't even touch some of this stuff. I don't even know how this got coded. Like, and that's AI talking AI. And I would argue that things like Dependibot and some of the things coming out from GitHub are some of the most powerful tools out there. And yet, it's like, that's just shit code. --- ### [Human Curation Will Remain Essential In An Agent-Driven Web](https://share.snipd.com/snip/22e743cd-478c-4b0c-8d5f-b0e1a6161e0a) 🎧 32:51 - 35:13 (02:22) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/b11e3ae7-3bee-4240-aa91-07671799d8b8" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Community curation and human dialogue remain essential to navigate AI noise and validate quality. - Wesley and Mary stress human-curated spaces like DevRel Collective or Mastodon as refuges from mainstream algorithmic slop. #### 💬 Quote > We need to actually be part of a community where we can have these types of dialogue. > — Mary Thengvall Mary Thengvall on the role of community in the AI era #### 📚 Transcript **Wesley Faulkner:** Yeah. **Jason Hand:** And I know we're short on time. should probably try to wrap it up here, but just to sort of add on to that, just begin based on very recent memory with the hackathon stuff, you can one shot an app, you know, and I showed you my, my, my project. And there are some like little things in there that were very much one-shotted. They're buggy for sure. There might be a CVE. I doubt it. But I didn't pay close, close attention because this is just a proof of concept. I'm not here to build this hardened security around something that literally just a little game. But what was hard was once I built this beautiful thing and it worked perfectly on my machine is getting it up into the cloud is deploying it into GCP using Docker. And not only that, but like once it's in, once it's in Docker, regardless of the tools, that, that thing is heavy. It's got a lot of 3d stuff. It's heavy on like resources. And so like my phone, when I tried to test it on my phone, turned into a little hand warmer because it was just like so hot with whatever's going on. So I had to use Datadog to figure out what in the world's going on with my app here to like reduce the CPU. The AI can't do all of that just yet. So there's like pockets of things that it is very helpful for me to, like I was able to like troubleshoot that, but it can't one shot it. You know, the very first time I had a problem with my deployment, it told me I had some sort of, you know, security issues that I need to go address first. And it turned out I was just trying to deploy from the wrong directory. So like, you know, like things like that still happen. But in general, if it comes down to just creating code and stuff like that, I'm seeing much and very few things happen or very few things built anymore that have these, you know, wrong packages that are security vulnerabilities or they're like using all these things that are just slop. Like just not happening with these frontier models. So anyway, I know we're getting right out of time. Leslie's got **Mary Thengvall:** to go. A quick, quick, quick note Jason mentioning this hackathon thing. He's not promoting AI. What he's actually doing is trying to trick you to use the app so you listen to bluegrass music. That's really what this is all about with him. **Jason Hand:** Maybe. **Mary Thengvall:** Unethical, Jason. **Jason Hand:** But anyway, yeah, we should end. Wesley, I just want say you made an excellent point earlier that a's not the, a lot of the AI stuff out there is we're not the customers, you know? And, and I think the truth is like, we're not the only customers. --- Created with [Snipd](https://www.snipd.com) | Highlight & Take Notes from Podcasts # Snips updated after 2026-04-05T10:16:17 ### [AI Content Is Being Built For Algorithms Not Developers](https://share.snipd.com/snip/6d607767-3658-4f2e-b1d0-4fcae6f91391) 🎧 00:38 - 04:12 (03:34) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/fd52d6bd-8e2a-45f3-a621-be50917a4bfb" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - AI-generated content is mainly produced to feed algorithms and agents rather than help human developers. - Jason argues marketing pushes low-quality, unattributed AI 'slop' into LLMs so brands show up in generative answers, decoupling content from developer trust. #### 💬 Quote > It is made to feed the algorithm. why SEO is being replaced by AEO or GEO, which is answer engine optimization as opposed to GEO is generative answer or generative engine optimization. > — Jason Hand Jason Hand on why AI content targets search/answer engines #### 📚 Transcript **Mary Thengvall:** So to set the stage, there have been a few pieces of content that have come out recently that were clearly AI developed that were less than of a quality that I would consider good and also had information that was incorrect or at the very least misleading. And my concern, what kind of started this off was if we're giving out bad information, is that going to hurt our standing in the world of developers? To which Wesley **Wesley Faulkner:** said. I said that we are not the customers of this type of creation, this content. And also I want to add that that document was also unattributed. So it didn't give any ups or references to even back up what it was saying and did not help with this authenticity by eliminating who might have actually given valuable data that this LLM sucked together. And when we look at what happened, like at block, where they fire 40% of their employees because of AI, whatever dominance. But then I don't know if you saw this post, where someone who did not get fired, did not get laid off was then offered like a huge bonus that was equal to like 90 of her salary to to stay i did not see that there's a they try to give retention bonuses to everyone who who was left behind uh and this person decided not to take it and to quit because they said they work. Their role was working within departments and 70 percent of the people that they worked with were gone and they didn't want to stay for that carnage. **Mary Thengvall:** So that's that's a whole interesting other take on an ethical stance against **Wesley Faulkner:** AI **Mary Thengvall:** slot firings. **Wesley Faulkner:** Yeah. And the reason why I want to bring this up is that people mean nothing. Customers mean nothing. There was a huge firing of 16,000 at AWS. And I knew several in the community space that no longer have jobs, which means that the people that they served are also devalued. the reason why is because if you think about marketing, which I think if we're talking about what content is used for in the developer space, it's for content marketing, not necessarily for developers, it is made to feed the algorithm. why SEO is being replaced by AEO or GEO, which is answer engine optimization as opposed to GEO is answer or generative engine optimization. These new terms are big because that's the main marketing channel is to pump it into those things, have those be a signal for those things. Don't need the references and the authority because all that's obfuscated when it's pushed into either a terminal or into these generative systems where they just like, we have enough signal. So this is not noise, which is wrong. And I think that is the main thing. It's just like, how do we get in front of developers? --- ### [Continuously Update Your Mental Model Of AI Tooling](https://share.snipd.com/snip/c23a7fa6-ad74-40bd-bba4-9e75bce6d862) 🎧 06:22 - 10:08 (03:46) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/cdaf25b1-011c-43ba-8f94-4c936b00aebe" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Stay current with AI tools and models to understand real developer usage and avoid outdated mental models. - Jason describes a Datadog hackathon where experimenting with agents, models, and production use showed practical developer adoption and benefits. #### 💬 Quote > Unless you've been messing with AI here in the last month or so, you're, you're just blatantly, uh, uninformed. > — Jason Hand Jason Hand on the rapid pace of AI change #### 📚 Transcript **Jason Hand:** In fact, this week, uh, I was just sharing with you both. We had a hackathon internally within our advocacy to explore lots of different things, mostly to up level ourselves on the different tools that are available to us, which is pretty much anything we want at Datadog. Like there's just there's lots of different things. There's agents, there's models, there's anything you need. And how can we use these tools to help do our day-to job and level ourselves up? can we use these tools to just understand where development is, what software development looks like in March in 2026? And then relay that on out to other developers out within the community. That's kind of our role too is like, what kind of an advocate would we be if we don't know the latest and greatest of how software is made you know like that's that's sort of what goes through my mind is like i can't just ignore this thing and think it's a trend there are developers who really do use this stuff i know pj you point out there's people who are hesitant and for sure you know like it's a spectrum but i can tell you for sure at least the datadog like we're all in using it very heavily in production. **Mary Thengvall:** And this, this begs the question though, if you're saying you're trying to use it to see how the average developer uses it, but you're using it heavily in production, but **Jason Hand:** the average developer is **Mary Thengvall:** not using it heavily in production. **Jason Hand:** We use it. I'm saying as most people will say, well, you're not using it in production. I'm here to say, no, we use it at every, if we just use it, like it's, it's being used. It is a, it's a, it's a co-pilot for all of us. know, the, all just AI in general is making all of us more, you know, of an improved version of ourselves and more efficient, more accurate, more improved across the board. So there's no argument there that that's happening. I think that there's an echo chamber that gets created when people who are, you know, have one experience and they go out there and say, this is how it is. And this is what I'm experiencing. And then people take that, you know, as like the gospel and then they, but, but I do think that there's enough of us, myself included, who recognizes that that's just not true. Maybe that's their reality, but unless you've been messing with AI here in the last month or so, you're, you're just blatantly, uh, uninformed, like whatever, whatever, like mental model you've created about this stuff, uh, you know, pre February 1st, I would say somewhere around there, February 9th, I remember what, but like, you're just wrong. Uh you got to go back and revisit like what's happening in the world within AI like very regularly just to just to know like what the state of things are and you're right with the the AIEO or AEO or whatever like that stuff is out there I feel like there are people the agents are writing stuff for other agents on the web. I showed you my hackathon is related to those LLMs texts. The, the inner, uh, the internet is becoming increasingly just agents talking to agents, you know, and we're like, kind of like, where do we exist in all this anymore? Are we the ones that are consuming this content online or is it just other agents and other LLMs and other like, you know, algorithms consuming it and making it gospel? And, um, you know, some of that I think comes down to us. I'm very positive that there's going to be a lot of slop and a lot of crap on the internet and there's got to be a way for the good stuff to make it to the top, you know, and whether it's Google or some others, they're going to put weight on things that show they're, you know, they're supported, I would think by the community, some objectively way, objective way. You know that this is bad and this is good based off of, you know, like it can be done. Um, and I think that's going to be the stuff that makes it into like gospel of AI. --- ### [Community Signals Could Rescue Quality From Agent Noise](https://share.snipd.com/snip/cf83398c-2b84-440b-bd96-832fb76325d4) 🎧 09:58 - 12:34 (02:36) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/cae7ef43-e291-456f-9fe5-7bd34897a929" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Good content can still surface if objective community signals are used to weight trust over agent-generated noise. - Jason hopes major platforms will favor community-supported, verifiable content rather than short-term SEO/agent gaming. #### 💬 Quote > I feel like there are people the agents are writing stuff for other agents on the web. The inner, uh, the internet is becoming increasingly just agents talking to agents. > — Jason Hand Jason Hand on agents consuming agent content #### 📚 Transcript **Jason Hand:** You know that this is bad and this is good based off of, you know, like it can be done. Um, and I think that's going to be the stuff that makes it into like gospel of AI. And we don't have to worry about these like posers or grifters who have never done the job and are just trying to put stuff out there thinking it's going to help themselves somehow in a short-term way. And it's really damaging to the industry. I guess I'm, I'm kind of hoping they're going to get kind of just shoved out because it just, it just, it's not going to. **Mary Thengvall:** I think that's a bit optimistic though. **Jason Hand:** I know. Yeah. **Mary Thengvall:** You hope that you hope the big guns like Google are going to do the ethical thing when we know for a fact they do. Well, what's **Jason Hand:** in it for them to put out the wrong information when we're all looking for the correctness? I think correctness is there. Do pay for it? **Mary Thengvall:** It's the same answer. Why did Google remove the first we shall do no harm concept from what they do? Don't Yeah, don't be. Why did they stop saying don't be evil? Because evil pays money. **Wesley Faulkner:** Blue Sky and Mastodon have already demonstrated that without this tipping the scale curation of this agentic or algorithmic feed, that I'm still able to see things from people I care about. And maybe not in the mainstream way. But there are spaces, especially like even like the DevRel Collective, like we share articles that we've seen. So it's a human curated space rather than an algorithmic one. And so those will exist if you are a member, know where to find these communities, know how to join these communities, know how to participate in these communities. And so, but then the mainstream area, the LinkedIn's, the threads the, the, even the Twitter, just say Twitter. X. Um, the, that stuff is still going to be probably prominent there. Like, even if you type into an LLM right now, make a post for X or whatever, it says, Hey, would you like to go viral? You want me to tweak it for its engagement? And it'll just do that for you and so i i already am seeing this thing where it's like everything is trying to be the best though the oh you can't believe this happened is it's the it's the like link bait like engagement stuff for almost all these mainstream posts and it's going to be it's going to people who's educated and know about this stuff that's going to be able to find the places where they can find the right information. And it's that same old, tired and true statement is that the future is here. --- ### [Enable Developers Internally Or Accept Falling Behind](https://share.snipd.com/snip/f7c2a32d-2a1f-4801-9ba4-a8367db5259e) 🎧 12:49 - 14:06 (01:17) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/a4fbcbfb-79f4-4eac-a883-28650576d2ef" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Advocate for enabling developers inside organizations to use AI tools or risk falling behind. - Mary warns most companies lack Datadog-like budgets and that access gaps create 'only-the-rich' participation in AI advantages. #### 💬 Quote > Most companies aren't going to have that. Very few companies are going to have that, in fact. > — Mary Thengvall Mary Thengvall on unequal access to AI resources #### 📚 Transcript **Mary Thengvall:** So only people with means could participate. This is very similar. Like the average person, sure, the average person, let's put the average person aside for a second. We talked about developers. The average developer company does not have a portfolio like Datadog where they can say, we're going to spend hundreds of thousands of dollars on tokens so that our people could use the best of the best. Most companies aren't going to have that. Very few companies are going to have that, in fact. So you already have a only the rich can participate situation. On top of that, the lack of guardrails, and we've seen that this week where your Anthropic was like, no, I will not accept, I'll say no to your defense contracts because we're not dropping our guardrails. And immediately OpenAI and ChatGPT were like, we're totally cool with losing all ethical balance just to make a dollar. And those are the two largest AI companies in the world right now, arguably. Like, so it's, it's kind of like, do we want to, can we survive to a point where this optimistic future that Jason talks about works out or will it be a continual division of if I'm not working for a major, corporation or super large business, I won't be able to ever get the tools that will allow me to advance my career. Well, **Wesley Faulkner:** I don't think it's one or the other. --- ### [Claude Turned Speaker Notes Into Overexplained Slides](https://share.snipd.com/snip/85cc8e1e-a7a2-485b-b9d5-fca14149ee45) 🎧 14:25 - 15:18 (00:53) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/6ef1f36b-4727-4b49-b0de-932042b8de56" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Mary used Claude to 'punch up' a talk and it copied speaker notes onto slides and over-explained jokes. - The result felt unhuman and highlighted how some assistants rewrite content in awkward, incorrect ways. #### 💬 Quote > It then proceeded to put all of my speaker notes onto each slide, then explained to me what the jokes were that I had written. > — Mary Thengvall Mary Thengvall recounting a tool editing mishap #### 📚 Transcript **Mary Thengvall:** And one of the things that Luma does is that very thing that you mentioned, Wesley, which is, hey, I see that you're writing some stuff. Do you want me to perk this up for you? Do you want me to punch it up? So I was like, okay, like I'm going to do, so I did a little bit of testing. I was like, yeah, go ahead and punch it up. Cause all it really says is like, Hey, the, the meeting, the, the dinner we're having is at this place. Uh, these are your, your meal options, you know, whatever it's, BizCat, whatever, that you would usually put on an invitation. And it rewrote it in a way that was so unhuman. And I don't mean inhuman, it was unhuman. Like it was something no one would ever say. Then I was like, that's preposterous. So then I decided I'm in the middle of writing a talk, I'm going to take the talk and ask Claude to punch it up for me. It then proceeded to put all of my speaker notes onto each slide, then explained to me what the jokes were that I had written. **Wesley Faulkner:** Oh, I need that so I could understand. Incorrectly. I'll definitely send that over to you. --- ### [AI Will Create An Agile Versus Legacy Company Divide](https://share.snipd.com/snip/7dd70901-fc55-4ee6-8234-fd3b19608ebb) 🎧 17:30 - 19:26 (01:56) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/5b63d10c-3229-46e3-aa74-a8744f12af9e" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - A bifurcation will form between AI-native, agile companies and slow-moving incumbents with heavy compliance. - Wesley outlines how compliance cycles make large firms miss rapidly iterating foundational models, advantaging nimble players. #### 💬 Quote > By the time that something goes all the way through their quote unquote process, they're not using the latest and greatest because they're coming out with a brand new foundational model every three months. > — Wesley Faulkner Wesley Faulkner on corporate adoption lag #### 📚 Transcript **Jason Hand:** years if they're not enabling their developers today. It's just a fact. **Mary Thengvall:** I wish that were true. I don't think that's true. Okay. I'll **Jason Hand:** stand on that for sure. I **Mary Thengvall:** think that **Wesley Faulkner:** what you're hinting to is like whose companies are enabling developers, their developers, and training their people to do better. And I think there's an ignorance about how people learn in the developer journey from companies, including internal developer relations. And we are going to start seeing a bifurcation in an arms race where those companies that don't teach their own developers how to use these tools will, I won't say be left behind, but will be a few steps back. And the scary part of this right now, we're seeing that the corporate dominance of even an anthropic say, no, we don't need a government contract or whatever, we'll step away is only going to get worse because when you think about big corporations, they have compliance loops to go through. They have some security loops. They have all this stuff in order to adopt these tools. And by the time that something goes all the way through their quote unquote process, they're not using the latest and greatest because they're coming out with a brand new foundational model every three months. And so companies are going to need to find a way to be more agile with their either inside their regulatory framework or these neo companies, these ones that are educated or these, let's call them AI native companies, are going to be the new dominant players. And what does that look like when you have new, like when you wipe the board of the top players for some of these things, and then you replace them with like three years ago, companies we never heard of? --- ### [Treat LLMs As Databases Not Reasoning Engines](https://share.snipd.com/snip/765edeef-eaaa-4038-acab-209e0bbc644a) 🎧 22:31 - 23:42 (01:10) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/f8b5327e-b1cd-4fdf-89ba-ed959848135e" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Treat current LLMs as very large synthetic databases, not reasoning minds; verify outputs rigorously. - Mary advises calling them 'toddlers' that echo training data and says they can't truly reason or replace expert validation. #### 💬 Quote > These things don't reason. They don't think. They're not intelligent. They're easier ways to access a database. > — Mary Thengvall Mary Thengvall on the limits of current AI reasoning #### 📚 Transcript **Mary Thengvall:** These things don't reason. They don't think. They're not intelligent. They're easier ways to access a database. **Wesley Faulkner:** You can't do that. I mean, it will never go away because these top companies need that word to be the thing. **Mary Thengvall:** Well, yeah, to sell the thing. I mean, **Jason Hand:** we've been talking about AI forever. I was delivering talks on AI like in 2018. Like this isn't. And and and what I mean is like LLMs is part of AI is like a part of the category. So like people have not been educated enough along to know the distinction between the two, you know, which is between all the different elements within the greater AI category. So it all gets rolled up. Oh, sure, sure. It's confusing. **Mary Thengvall:** And I give a bit of that, but it's when you say things like reasoning and thinking, it's not. What these things are is you have the ability to make a database query, which is, you know, an LLM is essentially an extremely large scale database. And you can do that without writing a SQL statement. It's synthetic. what AI is. **Jason Hand:** Yeah. It's emulating human reasoning in a synthetic. **Mary Thengvall:** It's not emulating human reasoning. It's literally just saying the thing that has been told. --- ### [AI-Generated Vibe Code Returned Hundreds Of CVE Warnings](https://share.snipd.com/snip/18561b61-6d6b-4f8a-827d-b64e51517aab) 🎧 31:00 - 31:32 (00:32) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/c006e607-09b4-4dce-b86b-2e46823d81c4" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - A public GitHub repo's 'vibe-coded' examples returned hundreds of CVE warnings that Dependabot couldn't fully fix. - Mary observed AI-generated front-end code with massive security issues, showing real risks from one-shot generation. #### 💬 Quote > The Vibe coded stuff was the only stuff that came back with like over five, like nothing else had security risks. This had 500 CVE issues that had to be dealt with by Dependibot. > — Mary Thengvall Mary Thengvall on security problems in AI-written code #### 📚 Transcript **Mary Thengvall:** We put all of our example code in the public GitHub repo. The Vibe coded stuff was the only that came back with like over five, like nothing else had security risks. Everything else was written in Python. This had 500 CVE issues that had to be dealt with by Dependibot. And Dependibot was even like, I can't even touch some of this stuff. I don't even know how this got coded. Like, and that's AI talking AI. And I would argue that things like Dependibot and some of the things coming out from GitHub are some of the most powerful tools out there. And yet, it's like, that's just shit code. --- ### [Human Curation Will Remain Essential In An Agent-Driven Web](https://share.snipd.com/snip/22e743cd-478c-4b0c-8d5f-b0e1a6161e0a) 🎧 32:51 - 35:13 (02:22) <iframe src="https://share.snipd.com/embed/obsidian-player/snip/b11e3ae7-3bee-4240-aa91-07671799d8b8" width="100%" height="100" style="border: none; border-radius: 12px;" sandbox="allow-scripts allow-same-origin allow-forms allow-popups allow-clipboard-write" ></iframe> - Community curation and human dialogue remain essential to navigate AI noise and validate quality. - Wesley and Mary stress human-curated spaces like DevRel Collective or Mastodon as refuges from mainstream algorithmic slop. #### 💬 Quote > We need to actually be part of a community where we can have these types of dialogue. > — Mary Thengvall Mary Thengvall on the role of community in the AI era #### 📚 Transcript **Wesley Faulkner:** Yeah. **Jason Hand:** And I know we're short on time. should probably try to wrap it up here, but just to sort of add on to that, just begin based on very recent memory with the hackathon stuff, you can one shot an app, you know, and I showed you my, my, my project. And there are some like little things in there that were very much one-shotted. They're buggy for sure. There might be a CVE. I doubt it. But I didn't pay close, close attention because this is just a proof of concept. I'm not here to build this hardened security around something that literally just a little game. But what was hard was once I built this beautiful thing and it worked perfectly on my machine is getting it up into the cloud is deploying it into GCP using Docker. And not only that, but like once it's in, once it's in Docker, regardless of the tools, that, that thing is heavy. It's got a lot of 3d stuff. It's heavy on like resources. And so like my phone, when I tried to test it on my phone, turned into a little hand warmer because it was just like so hot with whatever's going on. So I had to use Datadog to figure out what in the world's going on with my app here to like reduce the CPU. The AI can't do all of that just yet. So there's like pockets of things that it is very helpful for me to, like I was able to like troubleshoot that, but it can't one shot it. You know, the very first time I had a problem with my deployment, it told me I had some sort of, you know, security issues that I need to go address first. And it turned out I was just trying to deploy from the wrong directory. So like, you know, like things like that still happen. But in general, if it comes down to just creating code and stuff like that, I'm seeing much and very few things happen or very few things built anymore that have these, you know, wrong packages that are security vulnerabilities or they're like using all these things that are just slop. Like just not happening with these frontier models. So anyway, I know we're getting right out of time. Leslie's got **Mary Thengvall:** to go. A quick, quick, quick note Jason mentioning this hackathon thing. He's not promoting AI. What he's actually doing is trying to trick you to use the app so you listen to bluegrass music. That's really what this is all about with him. **Jason Hand:** Maybe. **Mary Thengvall:** Unethical, Jason. **Jason Hand:** But anyway, yeah, we should end. Wesley, I just want say you made an excellent point earlier that a's not the, a lot of the AI stuff out there is we're not the customers, you know? And, and I think the truth is like, we're not the only customers. ---