Sign in or Join the community to continue
AI Agents Need a Babysitter
Posted Oct 27, 2026 | Views 0
Share
Speakers
Erica Hughberg
Community Advocate @ Tetrate
Erica Hughberg is a technical leader and community advocate passionate about helping engineering teams build scalable, secure, and human-centric application platforms. With a background in software engineering and a deep understanding of cloud-native technologies, she specializes in driving the adoption of open-source projects like Envoy Gateway, Istio, and Kubernetes Gateway API, which enable organizations to simplify traffic management, security, and API distribution.
As a maintainer of Envoy AI Gateway, she plays a key role in shaping the future of API infrastructure. She focuses on features to ensure organizations can securely and efficiently integrate AI-powered services while simplifying traffic management, security, and API distribution. In the Envoy community, she drives collaboration, mentorship, and contributions that advance the project and its adoption.
Lastly, as a believer in the power of storytelling, Erica enjoys translating complex technical concepts into engaging, accessible narratives in the form of social media posts, conference talks, podcasts, and educational content.
+ Read More
Demetrios Brinkmann
Chief Happiness Engineer @ MLOps Community
At the moment Demetrios is immersing himself in Machine Learning by interviewing experts from around the world in the weekly MLOps.community meetups. Demetrios is constantly learning and engaging in new activities to get uncomfortable and learn from his mistakes. He tries to bring creativity into every aspect of his life, whether that be analyzing the best paths forward, overcoming obstacles, or building lego houses with his daughter.
+ Read More
SUMMARY
AI can write your code, choose its own tools, and book your next trip. But who checks that it actually did the job?
In this episode of Debug Mode, we explore what it takes to give AI agents more responsibility—and how to keep track of what they’re doing. We discuss intent-based tool selection, MCP and WebMCP, and why a convincing demo doesn’t always translate into reliable automation.
We also get into the personal AI assistant everyone seems to be building: one that manages travel, handles household tasks, and hopefully remembers to order coffee before you run out.
Then it’s back to code: QA agents, automated verification, tests that capture business intent, and the growing challenge of reviewing AI-generated contributions.
+ Read More
CONTENT & TRANSCRIPT
Erica Hughberg: [00:00:00] I am very lucky nowadays because I got Grok in my car, so I can chat to Grok in my car and have a wonderful time. We don't need radio anymore. We can talk to AI about things. As humans, are, are, are humans going to stop being the drivers, right? Yeah. Of computers effectively. With the whole, like, super intelligence situation, h- how are we gonna deal with that for the AIF?
Erica Hughberg: I mean, like, is it just gonna be the ASIF now?
Demetrios: Got a design. So I'm into calligraphy, and so I created, like, a calligraphy design probably five years ago. Okay. And I've been [00:01:00] flip-flopping about getting it, and now-
Erica Hughberg: Yeah ...
Demetrios: I ended up got it, getting it, and it's, it w- it basically says, um, "Let's talk about Jev."
Erica Hughberg: Oh, great. That's like you designed this lot years ago.
Erica Hughberg: That's, like, amazing. You know? How, how, how-
Demetrios: Reminiscing, my friend ...
Erica Hughberg: that's, like, crazy. Let's talk about Jev. Yeah, like-
Demetrios: Oh ...
Erica Hughberg: my goodness.
Demetrios: But what are, what are you feeling on Jev? It feels like we've kinda, we're on the down slope of the hype train. A lot of stuff has come out. Where are you at? I
Erica Hughberg: think that's exciting-
Demetrios: Yeah
Demetrios: to be
Erica Hughberg: fair.
Demetrios: Break it down. Why?
Erica Hughberg: Okay. So what I was doing in my car the other morning when I was driving home from the gym, as you do, um, I am very lucky nowadays because I got Grok in my car, so I can chat to Grok in my car and have a wonderful time. We don't need [00:02:00] radio anymore. We can talk to AI about things.
Demetrios: That's what I was doing when I was going to Japan. I just was talking to AI and trying to learn Japanese phrases that would be useful to me.
Erica Hughberg: Oh, dude, like I've been Of-- That's another topic of learning languages, by the way. Like, let's talk about that in a moment, because I've been trying to use AI to learn languages as well, which is related to my gym adventures because my-
Demetrios: All right
Erica Hughberg: gym instructor or like two of the gym instructors speak Spanish. So for some reason, like, I said, they speak English during the class, but they can also speak Spanish. Yeah. So I was like, "Oh, maybe I wanna know Spanish." But, but, uh, anyway, on the dev part, what I was thinking about is, you know how we've been speaking about security and like MCP tool calling security, as well as having the next layer of security is auditability, right?
Erica Hughberg: So being able [00:03:00] to make have an audit trail of what happened. Okay.
Demetrios: Yeah.
Erica Hughberg: So then I had this idea. Okay, I have not tried this idea. Let's be very clear. This was a morning conversation where like, okay, let-- And then I talked to Groq about it, but Groq said that there are some papers out there in the world that's come out, so I'm not completely crazy, but-
Demetrios: I love how it's just like, there are some.
Demetrios: Yeah, you wanna know what they are? We'll figure that out later. But they're
Erica Hughberg: out there. He told me stuff. He gave me names of authors. Do I remember those? I just was happy that I wasn't crazy. That was like step one. But-
Demetrios: Well, what's the theory? What is the-
Erica Hughberg: Okay. So imagine you have an agent, a tool-calling agent, but the agent doesn't know what tools are available to it, okay?
Erica Hughberg: So if you have that scenario, the agent cannot make a recommendation to call a tool that it is not allowed to call. It also cannot-
Demetrios: Doesn't know exists. Yeah.
Erica Hughberg: Yeah. So instead of the agent defining, saying which tool to [00:04:00] call, the agent defines what it wants to do, an intention. Okay. Right? What do I want to achieve?
Erica Hughberg: And then you use Jev with the intention, and then you have the selection you can make is different tools, right? Uh-huh. So you have Jev recommending the tool based on the intention and the need, okay? And what's kind of fun about this is that then we can actually start evaluating and fine-tuning. Like, okay, do we need better tool descriptions?
Erica Hughberg: Do we need better, like a better way for the agent to define its intent and the action that it wants to take for what purpose?
Demetrios: Mm-hmm.
Erica Hughberg: So I thought that was kind of interesting. Then it also becomes-- It's also auditable beyond which tools were called when. It is then auditable to why- Yeah ... was the tool called.
Erica Hughberg: Not
Demetrios: just- This is assuming that your, the [00:05:00] agent that pings over to your MCP server is going to give up that information, right?
Erica Hughberg: So this is data that, that doesn't send... Having Jev making the-- Effectively, you're giving your agent a single tool, which is the tool-finding
Demetrios: The Jev tool? ...
Erica Hughberg: tool. Yeah, it's a tool for finding tools that you use Jev underneath, and the contract is provide me with your desired action and effect- Okay
Erica Hughberg: and why you are making this. why you wanna use this tool, then it can find if it has access to any tool like that, or otherwise it can, you can come back and say there is no tool for doing what you wanna do. Like, there is nothing available to you.
Demetrios: Yeah. Right. Impossible task you just gave me. The-
Erica Hughberg: Yeah. Like, there, there is nothing for you.
Erica Hughberg: But there was... I mean, there are several things out there that's like, how can we use Gem for better tool selection for the task? So that's when I found out I wasn't crazy. That ob- obviously that's a [00:06:00] use case.
Demetrios: Yeah. I've
Erica Hughberg: seen that. But I, but I was thinking about it from a how can we effectively not even make the agent aware of all the tools, and also how can we make it so that it becomes auditable at the level of intent.
Demetrios: Well, did you see recently Pi, the harness, the creators of Pi, I think the dude's name is Marco, he put out a blog post about how Pi now is making MCP a first class citizen.
Erica Hughberg: Mm-hmm.
Demetrios: And the reason- What does that mean? It means that it's in the core harness. Uh, whereas before it was, like, an extension or an add-on.
Demetrios: And the reason that- Mm ... he was saying this, or one of the things that you get for free in, in the blog post, if I understood it correctly, was- Yeah ... that now you can use JEV also with [00:07:00] MCP and do what you're talking about, like classifying tool calls or- Yeah ... just using JEV in the MCP selection or the tool selection, all of that-
Erica Hughberg: Yeah
Demetrios: fun stuff. So I thought that was fun to think about and see how JEV almost, like, propelled MCP in a way. Yeah. And the use cases are plentiful out there on the internet. What I saw and I thought was one of the most interesting takes out of all of this-
Erica Hughberg: Mm ...
Demetrios: noise was Steve from builder.io. I don't know if you ever follow him, but I love his content.
Demetrios: Mm. It is top-notch stuff. Uh, he broke down all of these different posts and what they were saying they were using JEV for, and kind of like- Mm ... demystified them and said- Yeah ... "Eh, look, this is probably bullshit," on some of them. And then on others- Yeah ... he said, "Well, we've been hearing a lot about JEV's computer use skills."
Demetrios: So [00:08:00] he ran benchmarks to see if it is going to save you a ton of money, and if it is actually just as good. And what he found was, like, eh, it's not that good com- in the, uh, accuracy. Yeah. I think some of his benchmarks were like it only got the stuff right 33% of the time. Mm. And-
Erica Hughberg: Like what you use in, like, a computer use situation.
Erica Hughberg: Yeah. Like effectively a tool selection. Yeah. My criteria.
Demetrios: And the, the other piece was that that 33% was after a bunch of retries. So, like, the one- the first time it was 18%, and then you would have to retry and get to 33%. So it ended up not being that cheap either. Like, the whole value prop of JEV is, hey, it's faster and you get the same amount of accuracy, and it's super cheap, right?
Demetrios: But after he ran these benchmarks, he said, "Eh, not quite." What is actually the way to do it though [00:09:00] is something- Yeah ... that he was calling, like, a blended JEV or a hybrid JEV- Uh-huh ... type mode, because he Was saying, "Look, Luna or the smallest model, whatever- Yeah ... Haiku plus Jev." So you use Jev first. If it doesn't work, then you fall back to Luna or Haiku.
Demetrios: Have to
Erica Hughberg: make it more like, you know-
Demetrios: So that you just have that robustness ...
Erica Hughberg: reasoning-based decision rather than a-
Demetrios: Well, yeah, and a lot of times the input that you are constricted to or confined to with Jev- Yeah ... is actually one of the downfalls. It's like a double-edged sword because it's great that you can do this with this classifier model, but then you can't just feed it a screenshot, right?
Erica Hughberg: No. Okay, so that is where computer use becomes really complicated.
Demetrios: Well, no, it's, it's, it's more that's where if you have the fallback of a [00:10:00] smaller model- Yeah ... you get the accuracy higher-
Erica Hughberg: Yeah ...
Demetrios: when Jev can't figure it out.
Erica Hughberg: Mm. And like, what I think is interesting, like on the topic of like Jev and the inputs, I feel like do, there must be like a new art.
Erica Hughberg: You know how we talked about prompt engineering before, right?
Demetrios: Yeah.
Erica Hughberg: And, and context engineering and whatever else we wanna call those things. I, I think that Jev is gonna have that entire like learning curve for people as well. Like how do I actually get the most out of- All of a sudden- ... how do we get good stuff out of it?
Demetrios: Yeah. People are becoming machine learning engineers. I-
Erica Hughberg: Oh ...
Demetrios: like I saw a funny take where it was something along the lines of, "Wow, Jev made 1,000 AI bros on Twitter have to learn classical machine learning."
Demetrios: It's
Erica Hughberg: like, [00:11:00] I think that's another fun part about AI or at least that AI world, is that there is-- It's coming, like there's so many parts to it that touches on things that are not new- Yeah ... constantly. And I find that wonderful and fascinating how we are running into problems, you know, whether they are perceived or real.
Erica Hughberg: And when I say they're perceived, it's like, "Oh, it's a problem and I need to figure out to solve it," but actually it's already been solved. Or whether it is like in this scenario, say like Jev, it's like, "Oh, now we're gonna introduce all of these like classific- classifier-" Yeah ... "concept," and now people have to learn a different way of thinking that they maybe haven't used in their system designs or applications before.
Demetrios: Well, one thing that I feel like Jev could be really useful for, and I haven't seen people talking about it as much in this computer [00:12:00] use scenario is, or just browser use, whatever it- Mm ... may be, is if you combine it with WebMCP.
Erica Hughberg: Okay.
Demetrios: Because did we talk about WebMCP last week? I can't remember.
Erica Hughberg: Yes, we talked, yeah, we talked a bit about it.
Erica Hughberg: All right, so- Okay, so you're thinking that if you have WebMCP, which effectively allows a web-hosted site to expose tools to a agent that is interacting with it.
Demetrios: Exactly.
Erica Hughberg: Are you thinking then you can, how are you thinking you combine Jev with this?
Demetrios: Well, just the way that you were talking about-
Erica Hughberg: Yeah ...
Demetrios: Jev is classifying which tool to use.
Erica Hughberg: Yeah. Okay. Yeah- It's the same 'cause- ... that would make it a lot easier, yeah, wouldn't it?
Demetrios: Right now, exactly. It has to figure out stuff without any tools to grapple with in the computer usage. Just kind of like determining, do I press button now? Do I press that link? Do I go up and press something in the [00:13:00] header?
Demetrios: And it's weighing its options. Mm-hmm. But if you said, "Hey look, here's the four happy paths that people usually take on our landing page, and we-" Yeah "... expose those as tools," so maybe it's an add to cart tool. It is a like- Yeah ... go see the catalog tool, whatever. I'm using an e-commerce example.
Erica Hughberg: Yeah.
Demetrios: Then Jev can classify that based on the intent like we were talking about before.
Erica Hughberg: Yeah. Okay, and then I think to me, it sort of make, then it also makes sense in the world of like- If you have the, you know, you have Geff for decision-making, then you have a lighter model that is the act- that powers the more agentic behavior overall, right? So then you can even, okay, I don't know if this is a crazy idea or a sane idea, but actually have, if you have your like Luna or Haiku or even if you just go Sonnet or whatever, like you have that model [00:14:00] that can even enhance the options that are passed to Geff for the, uh, context, as in the intention of the user or the task.
Erica Hughberg: So I don't know, like could you then construct a Request to Jav with help from your inference
Demetrios: From another model? Yeah. When it's interacting with a WebMCP server, you mean? Or just
Erica Hughberg: like- Or like, yeah, like any tool selection effectively. But I mean, like if you are on the, I mean, the most typical, uh, use case, right, it's the pet store demo.
Erica Hughberg: It's a pet- Yeah ... store demo, right? And, uh, we're in the pet store demo, and like if we have the context of whether we are shopping for cat food or dog food, right? Or something like that-
Demetrios: Yeah ...
Erica Hughberg: maybe if we know the job, the ultimate task we're up for is buy cat food. Mm-hmm. [00:15:00] Maybe by having the inference model provide that context of the overarching agent task to the input for Jav, or maybe you don't even need inference.
Erica Hughberg: Maybe you just need a dumb agent harness that doesn't need inference for it. When I say dumb agent harness, I mean something that just puts these things together. It might not need inference. Maybe that's a better example.
Demetrios: You know what? You're, you're making an interesting point, because where should that responsibility lie?
Demetrios: Should we, as the agent builders or whatever the agent harness we're using, should that be on us, or should it be on the WebMCP builder to say, "Look, there is this pet store, and a lot of things that people wanna do when they come on this pet store website is buy dog food." Yeah. So that is a tool that I give you, and if you call that tool buy dog food, it just is straight [00:16:00] to, uh- Like-
Demetrios: you know, like pass go, and you're good. You're straight
Erica Hughberg: to- Or maybe you, like, then get to a different pla- okay, I have a question about WebMCP. Now I have a fun question- Yeah ... 'cause I don't actually know if this is how it can work or does work. So imagine you have buy dog food tool. Yeah. So you landed on the pet store shop.
Erica Hughberg: Mm-hmm. This, it will buy dog food. Okay, now you're at d- uh, so you've gone through the first door of buy dog food. So now we're in the virtual dog food aisle. If we're in the virtual dog food aisle, can, in WebMCP, it now surface new tools on that website?
Demetrios: Yes, exactly.
Erica Hughberg: Right?
Demetrios: Exactly. So now we can- So each page is a new, you get new tools.
Demetrios: So- Okay ... if it is, for example, you're just saying I'm this pet shop store, and what I'm going to expose as these four main tools when someone hits my landing page is- Yeah ... buy dog food, buy cat food, buy rabbit food, or buy, like, accessories. [00:17:00] And then the fifth tool is, like, search tools, because I have a million other tools there, but that's kind of the code mode type of-
Erica Hughberg: Yeah
Demetrios: method where you don't wanna overload it with tools. Yeah. You wanna give the four top ones and then search. Yeah. And so you then say, "Okay, buy dog food." Boom, you land in the dog food page, and now you're met with four new tools, which are, like, big dog food, s- small dog food, vegan dog food.
Erica Hughberg: GMO.
Demetrios: Yeah, whatever.
Demetrios: You get, you get all these different descriptions, or it's probably more like, um, select type of dog food, select- Yeah ... size and all that, and then add to basket.
Erica Hughberg: Yeah.
Demetrios: And then boom, you go to basket, and then the next one is decide how you're paying, just like- Yeah ... you kind of would with a human thing. But each- Yeah
Demetrios: new stage after you call a tool, you are met [00:18:00] with new tools. It serves
Erica Hughberg: new tools. Yeah.
Demetrios: Yeah.
Erica Hughberg: Okay, I just had, like, a picture in my head. There was, like, the skill trees in computer games where you get- Yeah ... one skill and you unlock another skill in the skill tree.
Demetrios: It really is that. Yeah, it's, it's very much a, a decision tree, uh, if you look at it that way.
Demetrios: But then there's also, there's some other cool things that I saw folks doing with WebMCP servers that are- Mm ... in a way built Uniquely for the capabilities that WebMCP has, and I realized this 'cause I went back and I listened to the interview that I did with the keynote speakers, um, Frederick and Dominic from Agent Con Amsterdam, and they were talking about how one of them, if I'm not mistaken, said that there is a WebMCP server, like, issue that [00:19:00] just got merged that is specifically for debugging.
Demetrios: So you can create a server on your website-
Erica Hughberg: Yeah ...
Demetrios: that is only for debugging, and so that agents can then go and use your website-
Erica Hughberg: Mm ...
Demetrios: and debug, like basically test it out and make sure everything's working properly. And if it's
Erica Hughberg: not- So like a QA, for a QA agent- Yeah ... can like- Yeah ... QA the site and click all the things they shouldn't click and see how it breaks.
Demetrios: And it's not exposed to... It's not like you would expose that to the general public, but you would wanna use that to sanity check if your WebMCP server or if your, just your website in general is- Yeah ... working as you think it should be working.
Erica Hughberg: Mm. I think, okay, the more we speak about WebMCP, I realize, you know how we talked about WebMCP last week, and we talked about [00:20:00] at one point is humans are, are, are humans going to stop being the drivers, right- Yeah
Erica Hughberg: of computers effectively, and are we gonna get, are we gonna go completely headless? And I
Demetrios: think- Yeah. And spoiler alert for anybody that didn't listen, no. We decided no, right? That's pretty-
Erica Hughberg: Yeah ...
Demetrios: clear.
Erica Hughberg: We think humans like looking at stuff, uh, and being part of it, but maybe not everything. There are some things that are really boring.
Erica Hughberg: But-
Demetrios: Yeah ...
Erica Hughberg: what I did find really fascinating, you know how we, you'd said earlier around Dev is making people think about machine learning concepts that- Yeah ... they may not have been exposed to before. Now with WebMCP, something that a- people who've been writing and creating APIs for the last, like couple of decades, something that has always been hard and frustrating is to move the idea of APIs to be workflow enablers rather than, [00:21:00] oh, the sort of CRUD create, you know?
Demetrios: Hmm. Yeah, that
Erica Hughberg: couple of some experience. Update, delete, retrieve.
Demetrios: Yes.
Erica Hughberg: Yes. It's like, not like it's just you have to figure out your workflow because a lot of APIs were sort of exposing the raw tools, if you will, and you then had to assemble your own workflow around the available tools. It's sort of Lego without instructions.
Erica Hughberg: But something that emerged in API design probably over the last bit more than five years, and it was there before, but it started to get popularized, was to do workflow-based APIs. Hmm. So think, you know, how you enable. Well, what is the program that is calling our system actually trying to achieve, and how can we design the APIs to achieve the workflow instead of just interacting with data or [00:22:00] functions?
Erica Hughberg: And now when you talked about WebMCP-
Demetrios: It's literally
Erica Hughberg: that ... I think it's the s- it's that like shift of thinking going from I'm going to turn my APIs into MCP Tools. I think actually fundamentally whatever it is you're building, whether you are building APIs or MCP, uh, tools, I, I think just focusing more on the workflow is- Yeah,
Demetrios: that end state
Demetrios: better It's the, it's the jobs to be done framework, right? Where- Yeah ... people say, "Oh, you don't want a drill, you want a picture on your wall."
Erica Hughberg: Yeah.
Demetrios: Or you
Erica Hughberg: don't- Or like you want clean floors. You don't want a vacuum cleaner. Yeah. Exactly. You want clean floors. So-
Demetrios: It just so happens that the vacuum cleaner is the tool to get you clean floors, and there's other- Yeah
Demetrios: tools that will help depending on what kind of floors you have, and so you have to be able to choose which tool is right for the job in that moment. But the-
Erica Hughberg: Yeah ...
Demetrios: intention- Or you
Erica Hughberg: can [00:23:00] hire a cleaner ... end state. Yeah ... and never buy a vacuum cleaner.
Demetrios: Yeah. Yeah, and then that's like hosted SaaS software, right?
Demetrios: Or the... I do like that metaphor and the thinking around it of you just want the outcome. For example, I want the dog food to show up- Yeah ... at my house.
Erica Hughberg: Yeah.
Demetrios: I tell my agent, "I don't care, like, what it needs to do." Ideally, the agent is able to then go and seamlessly, m- using- Yeah ... the marketing term, get that done very-
Erica Hughberg: Yeah
Demetrios: very quickly and with maybe one tool from a web MCP server. Yeah. But the bigger... Uh, but then you kinda go into this question of Does it matter if it has the WebMCP server, if it can do it on its own already without it?
Erica Hughberg: Mm. Yeah, that's the other-
Demetrios: That was the big- ...
Erica Hughberg: interesting-
Demetrios: Yeah ...
Erica Hughberg: question. But I think like fundamentally, if [00:24:00] we can design better for workflows, whether that is better user experiences, better APIs, better WebMCP tools or MCP tools, fundamentally I think how you achieve it-
Demetrios: Yeah
Erica Hughberg: is, this sounds silly, is not necessarily that important. It's, it's gonna be important in s- because things only take off if people adopt them. Yeah. But I like the idea of getting people into workflow thinking, and like how do I achieve something? And I just, you know, we talked about the pet food, and also being able to make decisions, right?
Erica Hughberg: When I shop for my cat's pet food, because I am a last-minute kind of person, do I not tend to try and get my cat food like the day before they're about to run out? Yeah. So sometimes I have to pick a more expensive cat food so I can get it delivered tomorrow, because the other one that they normally have, I would have to wait two days.
Erica Hughberg: Oh, two days, how [00:25:00] long? But you know, that is sometimes a problem. And I think it's nice, like could- I feel that ... could like you have the, you know, the goal of the agent is like you, I need to get cat food tomorrow, right? Mm-hmm. So even if there is cheaper food available in two days, but I need it tomorrow, right?
Erica Hughberg: So having
Demetrios: that- Having constraints. Yeah, more constraints- Yeah ... or more context around it.
Erica Hughberg: Yeah.
Demetrios: I haven't figured that out. That is something, not with cat food, but f- for me it's with coffee. I need- Yeah ... coffee beans, and I tend to order my coffee beans like just a little bit too late. And what happens is I have to go and ask my neighbor for coffee beans- Mm
Demetrios: uh, to hold me over while I have like one or two days of no coffee. And- Yeah ... I always think like remember back in the day when Internet of Things y- was coming out and people were like, "Wow, it's gonna be so incredible because we're never gonna have to order milk again. [00:26:00] It'll just magically appear at our doorstep when we run out of milk."
Demetrios: And that didn't happen, and I think we went about it all wrong, but I'm not sure how we could go about it so that I don't have to think about ordering coffee It just comes I
Erica Hughberg: mean, a subscription is sort of the answer
Demetrios: to that problem. But I've tried subscriptions, and the problem with subscriptions is I either get too many coffee beans, too much, because sometimes I go and I go on a trip, and I'm gone for two weeks or three weeks.
Demetrios: And you don't want
Erica Hughberg: old coffee beans. You want fresh ones, right?
Demetrios: Well, it's not even that. It's just that all of a sudden I have three bags of coffee beans at my house 'cause normally if I'm home, I'm drinking a lot of coffee, and so I'm ordering a subscription-
Erica Hughberg: We, we established that last week, the amount of coffee you were drinking.
Demetrios: Yeah. Okay. You gotta get that L-theanine to counterbalance it. Uh, so the, but the... So I don't know [00:27:00] where we're going with that, but it is something that I've been thinking about is how can I take that mental task-
Erica Hughberg: Yeah ...
Demetrios: and it's not only coffee, it's just mental tasks of everything around the house. I, I am just in my
Erica Hughberg: head thinking about the amount of coffee you consume and how this is a problem that follows you around in your life, whether that's in Japan and you've got too many coffee cups- Oh my God.
Erica Hughberg: It's so bad ... that you're carrying around 'cause you don't have a rubbish bin, or whether it is I need to figure out how I have coffee beans in my house.
Demetrios: The right amount of coffee beans. I don't want too many. I don't wanna overflow with coffee beans 'cause then I feel like I need to drink more coffee, and that's also not a good thing.
Demetrios: I just want the right amount of coffee beans. And I was listening to a podcast, uh, from those, those YC dudes, uh, I think it's like Michael and Dalton, and they were talking about how to create... They wanna see startups being created that are trying to bring the luxury experiences that rich- Yeah ... rich people have- Oh, yeah
Demetrios: [00:28:00] into the average everyday person's world.
Erica Hughberg: Okay.
Demetrios: And in this example, uh, it was kind of a douchey thing to say, but
Erica Hughberg: it's like- Was it coffee delivery?
Demetrios: No.
Erica Hughberg: The selection of the right coffee for you.
Demetrios: No. No, j- just taking mine off coffee for a moment. Let's, let's go with this one. What it was is they said, "You know, I am a person who does not have to worry about things being empty in my house anymore, because I've reached a certain level of wealth that I can have somebody else do that for me.
Demetrios: And- Yeah ... I wonder if there is a way to bring that experience to the average person- Mm ... like me, you know? So those guys- Yeah ... they can afford to have, like, a house manager. I do not have a house manager, as we've- Yeah ... established, because of my poor coffee timing. Wouldn't it be great if-
Erica Hughberg: You have your daughter, though.
Erica Hughberg: I think you need to train your daughter [00:29:00] for this
Demetrios: responsibility. That's a much cheaper and less energy-wasting way than starting a business around how can I bring perfect coffee beans to your door at the right moments.
Erica Hughberg: Uh- Yeah, I don't know what... She, she might have, like, compensation if she wants for a responsibility like this.
Erica Hughberg: Always, uh- I'll get her on coffee ... keep a tab on the coffee level, coffee bean level. Yeah. No, but it is-- To be fair, though, like, this conversation, so we hosted, like, a few weeks ago, like, a coffee, Lean Coffee After Dark in Atlanta as part of, like... It was our first little AIF meetup. Nice. And we had, like, conversations.
Erica Hughberg: But- Right ... what you were saying, though, one of the guys, he said that he really just wants An AI agent that can be a, uh, what do they call them? Those people who plan vacations for you. A travel agent, yeah. Yeah, your travel agent.
Demetrios: Like
Erica Hughberg: a proper personalized travel agent that doesn't [00:30:00] just, you know, come up with suggestions, but books, manages.
Erica Hughberg: Okay, uh, flights are delayed or, like, something happens also. You know, a good travel agent, like, will help you book new ones- They're on top of it ... like replan your trip. Yeah. Yeah. And he's like, he's like, "That's what I want," right? And I feel like this topic of having AI help you with managing your household or even, like, bring those luxury experience, like a personal, personal travel agent is one of those.
Erica Hughberg: Home manager is, uh, probably one of those. Personal tutor. You know, we said we were trying to learn languages with the help of AI, right? The having a personal language tutor is, uh, a level of luxury.
Demetrios: Yeah. I agree. I, I agree 100%. And you see that, like, it, it's funny because it feels like, uh... Saw a Twitter thread where, or a post on X that [00:31:00] someone said, "Everyone is building the same thing."
Demetrios: And- Yeah ... it does feel like everyone right now is building what I've heard termed the personal assistants or life assistants. Yeah.
Erica Hughberg: Muse. Was it Meta Muse? And
Demetrios: now- Muse came out ... OpenAI Dots. You've got w- it was, like, Inception, I think, or Instinct. That's it. Where- Okay ... that was the first one that was valued at 3 billion in their seed round, and everybody was looking for the promo code to get involved.
Demetrios: Yeah. And then recently, Dots. Yeah.
Erica Hughberg: Yeah.
Demetrios: OpenAI just released it. I'm actually very stoked about Dots, not because of the life assistant thing, but because- Yeah ... they finally brought in the Notion-like document editing to the interface. But that's- Yeah ... a bit of a, that's a different story. I've been asking everyone I know that works at OpenAI and Claude or [00:32:00] Anthropic-
Erica Hughberg: Yeah
Demetrios: to create that type of an experience, because I notice I'm dealing much more with the chat as a document and a living document- Yeah ... than I am, uh, wanting to take what I get from the chat and then pipe it over to a- Yeah ... Google Sheet or pipe it over to my Notion workspace.
Erica Hughberg: Yeah.
Demetrios: I just wanna have that inherent in my-
Erica Hughberg: And, and be able to organize the information Orga-
Demetrios: yeah, and tag in the agent when I want.
Demetrios: Mm-hmm. Like highlight something and tag it in or leave a comment and do all that stuff, or drag a certain text block or create a, um, some kind of a visualization with data or- Yeah ... @ someone and tag them in or @ a database and tag it in. And so I wanted that whole like Notion vibe, but inside of my-
Erica Hughberg: Yeah
Demetrios: chat. [00:33:00] And turns out, m- open AI, OpenAI listened to me because now they have that in
Erica Hughberg: HyperTalks. They listened to you, but there hopefully more people who felt the same way.
Demetrios: I, it must have been me. I'm the-
Erica Hughberg: Just for you, the single user.
Demetrios: Have you ever heard that expression, "Squeaky wheel gets the oil"?
Erica Hughberg: Yes.
Demetrios: Yeah.
Demetrios: That w- I was- Yeah ... quite squeaky on this one about it, and hopefully it was because of me that they- Yeah ... decided to do that. And so-
Erica Hughberg: I like to view you as a, uh, champion of the underserved.
Demetrios: I'm, I'm like a, a visionary, I think is what they would call me.
Erica Hughberg: Yeah.
Demetrios: That's... Should I just say that? I sound like such a...
Demetrios: I felt like a douchebag saying
Erica Hughberg: that. No, it's okay. But I can't imagine what it sounded like. You just said you're a visionary and not a doer. That's fine.
Demetrios: Exactly. Exactly. I don't
Erica Hughberg: want to actually- Just say that other people did the things.
Demetrios: Yeah.
Erica Hughberg: OpenAI made the thing. You just-
Demetrios: But it was my idea, all right?
Demetrios: Of course. Don't forget. Uh,
Erica Hughberg: but- If you have loads of ideas, all [00:34:00] of them are yours at one point or another. It's fine.
Demetrios: I know. It is, it, that, that's a whole 'nother topic of inspiration and where ideas come from, but, uh, that's for different podcasts.
Erica Hughberg: How, how, uh, how is... I have a funny question. Not funny, I don't know.
Erica Hughberg: It's a question. With the whole, like, super intelligence situation, h- how are we gonna deal with that for the AIF? I mean, like, is it just gonna be the ASIF now? How do
Demetrios: I know
Erica Hughberg: what to do with it?
Demetrios: There was never, yeah, there was never a AI, uh, AGI, AF, you know? Like, it, it was always just AI, so whatever.
Erica Hughberg: I think, yeah, it's, it's interesting, right?
Erica Hughberg: 'Cause some people are obviously arguing it's... AI has always been the umbrella term- Yeah ... as super intelligence is just a type of AI. Like
Demetrios: something. Exactly.
Erica Hughberg: Um, [00:35:00]
Demetrios: but- You know what I wanted to actually talk about, though-
Erica Hughberg: Yeah ...
Demetrios: is this idea of building verifiers in your code and in your-- And it goes back to, 'cause I remembered it as I was mentioning the QA assistant for, or just the QA WebMCP type of server that you can create.
Demetrios: And what I recognize is I wanna learn from people who are building their verifiers, how they're building their verifiers, because it feels like that's where, that single point on how you're verifying the code that is being churned out by your agents.
Erica Hughberg: Yeah.
Demetrios: What does that look like? What is that whole process, and what are the new rituals that we're creating?
Demetrios: Mm. Because we used to have it where you would submit PRs, and now maybe you can get rid of a lot of [00:36:00] PRs by creating really good verifiers. And verifiers, I guess, uh, uh- Yeah ... another term for it could just be, like, uh, tests- Yeah ... or a really robust CI/CD pipeline. But- But
Erica Hughberg: I do think this is interesting 'cause people, again, people have been dusting off an old approach that had, um...
Erica Hughberg: So this is f- I mean, this is what I've learned from some of my friends and colleagues is... So do you know about cucumber testing?
Demetrios: No, what is
Erica Hughberg: this? Okay. It's very much a Java-based testing framework. I wonder if it works for other programming languages, but let's go back to the basics of what it's trying to solve.
Erica Hughberg: The fact it's called Cucumber is probably rather irrelevant, but it is funny because it's so old, uh, as a framework. But instead of testing like method functionality or like feature [00:37:00] functionality, you are testing like business outcomes, like more like-
Demetrios: Oh ...
Erica Hughberg: you are testing against-- You, you start off defining, again, coming back to the idea of workflows, and I think some people back when it was popularized, might get this wrong, but I think people referred to as like business-driven development.
Erica Hughberg: It was a time when-- It was also the time when you often had in large companies, like a very divided, like here's the tech engineers, the software engineers, people doing that, and the business is over there. And then you had business analysts and people who sort of translated business needs into requirements for software.
Erica Hughberg: And this idea was effectively take the requirements a- and sort of encode them into-
Demetrios: The tests? ...
Erica Hughberg: tests, if you will. Ugh. And, uh, I've seen [00:38:00] people been, uh, using this because that allows you to maintain the business impact and outcome-based information into your code effectively. So you don't have to ask yourself- Well, why did the agent write this code?
Erica Hughberg: Because there's a test that is using the code to verify that the intended outcome actually happens.
Demetrios: Mm-hmm.
Erica Hughberg: Right? So what could be a really simple idea, maybe think of onboard, set up a new user in the system. Setting up a new user in a system would be more than just creating the user in the database, right?
Erica Hughberg: It would probably involve creating the user, set up permissions, send like email to like notify them that they have an account to activate, blah, blah, blah, whatever, right? It would probably m- make sure they're part [00:39:00] of a certain team and whatnot. And the idea that my friends showed me was that that would mean one of the big questions, they said that it comes before quality, is why was this code created in the first place?
Erica Hughberg: Ah. So being able to go back, okay, this code was created to solve this particular problem, and this is especially relevant maybe in very large code bases and da- like code bases where you are working to try and understand why the coding agent even wrote that code, for what outcome and purpose.
Demetrios: Uh, yeah, it makes me think about the auditability of it, and- Yeah
Demetrios: tracing back the decision that was made to actually include this code into the code, because there was some-
Erica Hughberg: Yeah ...
Demetrios: there was some spark that caused this to actually materialize.
Erica Hughberg: [00:40:00] Yeah. And I, and I think like what's interesting about writing these type of tests is that you have to be, you know how I said back in big enterprise, like 15 years ago, you had the business, you had the software engineers, and then you had business analysts in the middle.
Erica Hughberg: You have to operate more like a business analyst yourself. You have to have a deeper understanding of the actual business, not just how to execute the code as a software engineer or builder or whatever you want to call it.
Demetrios: Yeah.
Erica Hughberg: And I think that's what's interesting to me in how we have seen these, like, in between technical people become very powerful in the era of AI because they have the ability to express their intent.
Erica Hughberg: 'Cause they've had to s- they've had to express it over so many years already to humans, and since AI operates on natural language, they [00:41:00] have now been able to interact with this much better than- Yeah ... they have before, say. I think that to me is quite fascinating.
Demetrios: Well, so the verifiers that I was talking about were way, it was a totally different field or ballpark than what you're talking about, but I do like that you took it here because In a way, it reminds me of shift left and trying to shift left so far left that you're at the inspiration of what then comes out of it.
Demetrios: Mm. And so that's great. The verifiers I was thinking about though are, and how to architect verifiers and what I've heard folks saying that they do i- are things like just making sure that before any PR gets [00:42:00] submitted- Yeah ... you spin off a sub-agent and that sub-agent goes and it tests the whole new changes and- Yeah
Demetrios: your whole application to make sure that nothing broke and everything is showing up the way that it should, and it will write a document with screenshots that verifies- Mm ... okay, I clicked this button and it should go here, and it does go here. Here's a screenshot of me being here. And oh, I click this button, it looks like this is off.
Demetrios: Now I need to go and change. I broke this page or I broke this link or I broke whatever. And so you're able to have a, a- again, it's like this QA agent that is- Yeah ... going and verifying the code that's trying to be changed- Yeah ... before it gets to the whole PR and CICD
Erica Hughberg: pipeline.
Demetrios: Like
Erica Hughberg: limit the amount of human [00:43:00] review- Interaction.
Erica Hughberg: Yeah ... so that we can reduce it. I think like even just the simple things that are not a- agent driven quality controls-
Demetrios: Mm-hmm ...
Erica Hughberg: is, um, so often not checked by people. Say, for example, a common thing that you would have in a software project is that you have linting, testing, et cetera, right? That's like the, some basic stuff.
Erica Hughberg: But The amount of, like in open source, where in an open source project this exists, the amount of PRs I see get opened that don't even pass the existing-
Demetrios: Uh-huh ...
Erica Hughberg: sort of testing harness, right?
Demetrios: Yeah.
Erica Hughberg: Uh, and I find that interesting, right? Like, that's another human problem. Like, the, the human is [00:44:00] clearly being told by the PR that stuff's not working, or their coding agent, or whichever we wanna see it as, right?
Erica Hughberg: But then no action is taking, taken to address it. Mm. So that's another, like, how do you do that, deal with that as a maintainer. Like, I try to like ping people and be like, um, "Do you wanna fix this?" Or sometimes it's hard because sometimes part of what they've done is actually not bad, then I'm like, "Do I need to go rescue this and like bring it in because there is something useful in there?"
Demetrios: Yeah. The-- Man, that's something we should talk about next week, is just open source projects right now and how you are fielding all of the AI-generated code that's coming through, and how you are looking at maintainers or just fly by night contributors that wanna get something merged, and they'll throw some slop at you, and [00:45:00] you get inundated with slop, and you have to set hard rules and what those rules are.
Demetrios: So let's do that next week. I gotta jump now, though.
Erica Hughberg: Yeah. But I like the idea of trying to figure out how can we-- We have got some help from a Google Gemini tool to try and help us with like reviews, but- Nice ... it's, it's still like, I wonder can you even enable people even more? Like you were saying, c- can we try and even help coding agents help themselves with a testing agent before-
Demetrios: Build a verifier that will explain to the agent why this code is going to automatically be rejected if you submit it as a PR.
Demetrios: Yeah. That type of thing.
Erica Hughberg: Exactly.
Demetrios: There's-- Yeah. Awesome. Well-
Erica Hughberg: Cool ...
Demetrios: I will see you next week.
Erica Hughberg: Yes. Or hopefully I see you too. No, I will see you. Yeah.
Bye-bye.
+ Read More
Watch More
Generative AI Agents in Production: Best Practices and Lessons Learned // Patrick Marlow // Agents in Production
Posted Nov 15, 2024 | Views 6.5K
# Generative AI Agents
# Vertex Applied AI
# Agents in Production