Are rolling token blackouts and late-night AI coding shifts about to become the new normal for developers? This week on the Friday Deploy, Andrew and Ben explore the shifting economics of AI compute before debating whether the Model Context Protocol (MCP) was fundamentally overhyped. The hosts also dive into "context anchoring" to prevent model compaction during long coding sessions, why optimizing the wrong bottlenecks makes AI an amplifier for bad processes, and the nostalgic resurgence of the decentralized "small web." Finally, they break down the new rules of workplace AI etiquette to help you avoid serving your coworkers "sloppy pasta" disguised as real work.
Show Notes
- MCP is Dead; Long Live MCP!
- Context Anchoring
- If you thought the speed of writing code was your problem - you have bigger problems
- The “small web” is bigger than you might think
- Stop Sloppypasta
Transcript
(Disclaimer: may contain unintentionally confusing, inaccurate and/or amusing transcription errors)
[00:00:00] Ben Lloyd Pearson: Andrew, are you taking advantage of Claude's recommendation to start working nights and weekends now?
[00:00:06] Andrew Zigler: What do you mean, Ben?
[00:00:08] Ben Lloyd Pearson: Uh, you haven't gotten the notification yet and when you sign in it's like, Hey, we'll, we'll give you more tokens if you work late at night because it's cheaper for us.
[00:00:17] Andrew Zigler: Oh yes. If you distribute your workloads, the times when not everyone is rushing to, to slam GitHub with their, with their pull requests. Definitely getting people off of the morning shake up on Mondays for sure. I don't know about you,
[00:00:28] Ben Lloyd Pearson: Yeah.
[00:00:29] Andrew Zigler: morning it's just like, oh, is this gonna be, is it gonna fall over today?
[00:00:33] Andrew Zigler: It's
[00:00:33] Ben Lloyd Pearson: Yeah, I, well, I.
[00:00:34] Andrew Zigler: week. Now I'm, I'm like betting on my repositories.
[00:00:37] Ben Lloyd Pearson: mornings, it's like, I feel like almost every AI tool is like barely surviving, just like scraping by, suffering through the like first few hours of the, the work week before, like things sort of flatline, you know.
[00:00:49] Andrew Zigler: It, it makes me think of like, uh, a very kind of apocalyptic future where there's like rolling token blackouts where you're in like a, a token blackout because the consumptions [00:01:00] are, are too high in your area. And imagine what that would turn off. It has electricity. When it shuts off, you can't
[00:01:05] Ben Lloyd Pearson: Yeah,
[00:01:05] Andrew Zigler: things. Maybe when the token stop flowing in your area, a lot of things don't work anymore either.
[00:01:10] Ben Lloyd Pearson: you know, like people get like batteries for their house when they have like solar PA panels, you know, to collect energy and then like use that battery energy when pri electricity prices are high. Like, I wonder if there's a way for me to just like collect tokens, you know, so that I can leverage my, my bank of tokens when, when token costs are high.
[00:01:28] Andrew Zigler: Okay, well, somebody had been a solar farm and let him get cooking. 'cause that
[00:01:33] Ben Lloyd Pearson: Yeah, it's true.
[00:01:34] Andrew Zigler: it would be nice to just get more tokens in general. So I love Claude's idea of distributing, uh, folks around. I would just wonder, uh, who will do it? Obviously, if you can orchestrate your agents and they're running
[00:01:44] Ben Lloyd Pearson: Yeah.
[00:01:45] Andrew Zigler: all the time anyways.
[00:01:46] Andrew Zigler: What's it matter? It's like they're, they're probably always
[00:01:49] Ben Lloyd Pearson: Okay.
[00:01:49] Andrew Zigler: you could just have them run later. So.
[00:01:51] Ben Lloyd Pearson: Yeah. Which is saying is all the more pressure to just have our agents constantly working for us when we're not working, I guess. Yeah. Well, [00:02:00] and on that note, welcome to the Friday Deploy. I'm your host, Ben Lloyd Pearson.
[00:02:04] Andrew Zigler: And I am your host, Andrew Zigler.
[00:02:06] Ben Lloyd Pearson: And in this week we've got MCP is it Dead? How to anchor context to prevent model compaction, the resurgence of the small web and an AI etiquette guide.
[00:02:16] Ben Lloyd Pearson: So let's just start at the top with MCP. Andrew is MCP dead.
[00:02:21] Andrew Zigler: Okay, well this spicy article is about how MCP is gone and no MCP is back and MCP is somewhere in the middle. Yes, I think AI's Darling Child MCP has certainly ran its course of popularity. I love how this article calls out the fault of AI hype influencers for making MCP more popular than it needed to be, I don't know if this was directed at me, I'm just kidding. But if you, if you have been listening to Dev Interrupted, especially around this time, about a year ago, you'll know we were talking about MCP on this, show a lot and,
[00:02:54] Ben Lloyd Pearson: Actually, the timeline in this article actually almost perfectly lines up when we were hyping MCP, so I feel like [00:03:00] he probably is calling us out directly on this one.
[00:03:02] Andrew Zigler: yes. Clearly a Dev Interrupted reader, clearly disgruntled by our effusive coverage of MCP. And you know what? I don't blame them because reflecting on it, it was a different problem, a different solution for a different time, because around this time, a year ago, let's not forget, models were a lot dumber. Their ability to use tools was not as great as it were as it was today. And MCP was a fundamental day and night change in the ability to get consistent tool usage from LLMs. It really paved the way for more token optimized and simpler solutions. You know, we all kind of ended up going back to CLIs in some form or another instead of carrying around massive MCP servers.
[00:03:43] Andrew Zigler: And the benefits of this are, are numerous and lots of folks talk about them, but what this article calls out is why was MCP in the spotlight in the first place? And I think it's a great reminder that, you know, we're in a growing pain. We're going to be, uh, embracing and using technology that maybe's not gonna stick [00:04:00] around forever. So you should always be working in a way that expects your tools to be temporary and moving with you, not standing still.
[00:04:07] Ben Lloyd Pearson: Yeah. And I want to point out that I, I kind of thought it was obvious that CLIs are like more of a crutch for AI agents than they are like, uh, an interface that, that they would, should be standardized around. And it's really just because it's like, it's just the most common interface, right? Like it's been, it's been around for decades now.
[00:04:23] Ben Lloyd Pearson: and so if, if you're gonna have an agent do work, it's gonna naturally pick up a CLI because what other, what other option out there is, is there? You know, I think if you were to actually ask like your agents, like how would you design and implement your ideal interface, like it would look nothing like a CLI even though it's become such a popular tool for AI agents.
[00:04:42] Ben Lloyd Pearson: But yeah, I really like the, just the pragmatic perspective on MCP. In this article. So, you know, like I said, we probably contributed to this, but MCP was overhyped months ago. Um, it kind of seemed like everyone was MCPing all the things like not a week went by that there [00:05:00] wasn't half a dozen companies announcing their new MCP product thing that they've attached to their, their platform.
[00:05:07] Ben Lloyd Pearson: Um, but the reality is that there's really like more situations where MCP shouldn't be used, and there are actually situations where it should be used. It's not like this one size fits all agent connector thing. So if, if you're like still trying to understand like what's the difference between a, a use case where MCP is great and where it's not like this is, this is some really great reading that I think is super timely.
[00:05:29] Andrew Zigler: You know, one last thing I wanna end on there is a, a big thing MCP solves that CLIs still won't, is distribution. MCP is a really great way of getting your tool and your and, and. If you're widely used API in the hands of a lot of people really quickly and consistently, and while CLIs can certainly be, uh, applied in the same way, just MCPs are more portable, uh, just at the end of the day.
[00:05:53] Andrew Zigler: And so there is definitely still a use case for them.
[00:05:56] Ben Lloyd Pearson: Yeah. All right. Let's move on to context anchoring. What do we have here, [00:06:00] Andrew?
[00:06:00] Andrew Zigler: Yeah, so
[00:06:01] Ben Lloyd Pearson: I.
[00:06:01] Andrew Zigler: is an article from ThoughtWorks. It's covering latest research about how AI coding sessions, you know, when they run very long, and after a while you get to a point where you have to ask yourself like, would I be stressed out if this chat closed right now? Would it be disrupting to my work if this chat were to end?
[00:06:20] Andrew Zigler: And most, most developers, uh, once they get in into a few turns with any, a AI coding tool would answer yes. And it's because so much of the usefulness of any session gets wrapped up within this temporary window that over time experiences an event we all know as compaction. And this compaction event is pretty savage.
[00:06:40] Andrew Zigler: It, degrades what the AI can recall. Uh, this research shows that the things that go first are the reasoning behind decisions, not the decisions themselves, which creates an even shakier ground. The further you go with other experts, like we've had Dexter Horthy on the show. Called out rightly that around 50% it really drops off a cliff. [00:07:00] And, the solution for this becomes a technique known as context anchoring. I'm a huge proponent of this. We'll talk about this like a little bit more too. but the idea is that you capture. Like a living feature document that externalizes context and decisions. It's a little different from an architectural decision record.
[00:07:18] Andrew Zigler: Those are more formal, um, and exist within a different, uh, like a different place. This is more like a substrate. Where you and the LLM are working together to capture this very temporary bit of knowledge. Uh, there's lots of different tools for doing this, but I gotta say this, uh, article comes at a really fascinating time when Opus 4.6 is just right now reaching its 1 million context window.
[00:07:41] Andrew Zigler: I've been using it on a daily basis, and my sessions now just have so much room in them. It gets really tempting. To go for way longer than I would've even at like a week or two ago. Ben, what did you think about this article?
[00:07:55] Ben Lloyd Pearson: Yeah, well this is part of a series that we've been falling for quite some time [00:08:00] now, uh, from the ThoughtWorks team over on Martin Fowler's blog. Uh, and the, you know, it's a part of a series called The Patterns for Reducing Friction in AI Assisted Development. And I, I really. Love this term or this phrase, context anchoring.
[00:08:14] Ben Lloyd Pearson: And you know, partly because like, I feel like we're in this era right now of AI where we're, we're coming up with a lot of new terms to describe like, phenomenon that, that we're experiencing within this space. And this is one of those terms that, that I think is coming out of that. Uh, and I think it, but I, I really think it's a great term to describe what, what you and I, Andrew, have really been focusing on in recent weeks with our agents.
[00:08:36] Ben Lloyd Pearson: It's like the more that you can build that context behind everything. You know, we've, we've been talking about context engineering for a while, but, uh, this is sort of like a, a way to mature that practice. And, and I want to just like step back for a minute, minute because, um, the article did call out a practice that I, I think it's a very common anti pattern.
[00:08:56] Ben Lloyd Pearson: Um, and that's when you do things like over-relying on like a long [00:09:00] chat thread. Um, with your GPTs, like it kind of, that kind of illustrates like the challenge that context anchoring can help solve, you know, and, and on one hand, like I've always tended to work in a much more ephemeral way. Like, you know, browser tabs to me, like, I open them when I need them, when I close them, when I'm done.
[00:09:17] Ben Lloyd Pearson: They only exist as long as I need them. Notes that I take are on like temporary notepads, and then they're deleted after they serve their purpose. Working documents are like forgotten once the, the thing that I'm working on gets published out to the, to the world. and in a sense that like, that's made, that's given me this really strong default approach to LLMs where I'm, I'm always treating them as like these ephemeral workers.
[00:09:40] Ben Lloyd Pearson: Um, and, and as a result, I've kind of avoided that anti-pattern I was describing where I would have these longstanding chat threads. And, and a big part of that was because I understood that if, if you want to hand off a repeatable problem to ai, um, you need to keep it as constrained as possible, uh, to avoid model failure and like those [00:10:00] long running chat history type things.
[00:10:02] Ben Lloyd Pearson: But this practice breaks down when you need historical context for the model to perform well. AI needs to know what was decided, why that decision was made, who were the people that were involved in making that decision, what additional context is out there behind, you know, like what sort of external constraints exist on this situation?
[00:10:24] Ben Lloyd Pearson: And so now, like, you know, I, I still kind of like, I truly. GPT is ephemerally, but the context is, uh, you know, I'm, I'm recording and archiving context as much as possible now because everything is important for, for helping AI to understand where I am. Like not only the, the problem set that I'm trying to solve, but, but where I am and how I got to where I am at this moment.
[00:10:46] Ben Lloyd Pearson: So, yeah, another, just a great article from ThoughtWorks. Everyone should be following the work over there, and I'm, I'm sure we'll continue to have awesome information from them to share.
[00:10:55] Andrew Zigler: Really well said, really great summary as well. I, it's funny you say that you've always worked in a [00:11:00] more ephemeral way. I actually find myself more gravitating towards the opposite. I've always kind of worked in a very kind of
[00:11:06] Ben Lloyd Pearson: Yeah.
[00:11:07] Andrew Zigler: way. I think the future belongs to collectors, people who have been curating their own content and knowledge and archives of things. Um, maybe I just love to collect things in folders, but a lot of them are marked down and it's just been crazy how future proof and future compatible that way of thinking was and still continues to be. So if you're not somebody who is keeping a, a daily dot journal or is capturing, uh, things from every, every day that, maybe would otherwise fleet out of your mind, like I would encourage you to do so. It's an amazing mindfulness practice that benefits you and all of these other things that you're working on in your life.
[00:11:45] Ben Lloyd Pearson: Yeah, yeah. You know, capturing transcripts into obsidian notes and having Claude Cowork connected to that is, is a wild experience. But, uh, it is, it is immensely powerful.
[00:11:55] Andrew Zigler: Or if you just have a big list of coding challenges and a markdown folder that you just love to go [00:12:00] back
[00:12:00] Ben Lloyd Pearson: Yeah.
[00:12:00] Andrew Zigler: in a while and revisit, that's just fine too.
[00:12:03] Ben Lloyd Pearson: Yeah. Yeah. All right. Well let's talk about some of the challenges that, that, that come up when you, when you don't have great AI practices, particularly around things like context anchoring, anchoring, uh, and this article comes from Andrew Murphy. It's titled, if you thought the speed of writing code was There Problem, you have Bigger problems.
[00:12:20] Ben Lloyd Pearson: So, you know, we've got engineering leaders all over the place out there that are rushing to deploy AI coding assistance. Claiming all these increases to output. And, but you know, the reality is that a lot of them are really just like optimizing the wrong bottleneck. a topic we've covered consistently here on Dev Interrupted is that, you know, writing code is, has never really been the constraint.
[00:12:42] Ben Lloyd Pearson: It, it has sometimes been a bottleneck, but in, in software delivery, it often is not the thing that slows it down. Um, and if you, you know, you operate on the theory of constraints, you know, optimizing something that isn't the bottleneck is just going to make your existing bottlenecks bigger and have more problems build [00:13:00] up.
[00:13:00] Ben Lloyd Pearson: so the real bottlenecks in most organizations is things like, unclear, uh, requirements, you know, lengthy review and deployment processes. We talk about code reviews frequently and how that is almost always or very frequently, uh, the most, the biggest bottleneck in the typical software engineering organization.
[00:13:19] Ben Lloyd Pearson: Um. Beyond that, you know, fear-based cultures, organizational coordination, these are all things that can create bottlenecks, uh, within your teams and that get worse as you add AI on top of them. So if you're as a team, not mapping that entire SDLC and understanding where those constraints are, and, and you know which portions of the cycle, the software cycles are the ones that are slowing you down, um, then AI is actually, you know.
[00:13:46] Ben Lloyd Pearson: It, it, it's an amplifier. It's, it's gonna make all of those bottlenecks, all of those problems worse rather than better. So, we love this article just because it's, you know, just another expert out there agreeing with something that, that we hear over and over again from, [00:14:00] from our community.
[00:14:01] Ben Lloyd Pearson: Uh, so yeah. What did you think about this, Andrew?
[00:14:03] Andrew Zigler: Great summary. There's not much more that I would, I would call out that you didn't already cover. But in terms of the size of really the scope of this problem and how much we've been attacking on the show, this is like a, a pervasive issue that I think every engineering team, um, is encountering right now.
[00:14:19] Andrew Zigler: And if it's lurking and you haven't identified it yet, it's really critical that you do the idea of having the AI rollout without tying it to the impact and understanding downstream effects of, of, of using these tools within your SDLC. Uh, this really calls out how all of the other bottlenecks besides, writing the code are what make that possible.
[00:14:40] Andrew Zigler: So you have to focus on the whole system, not just one part of it. Really great, uh, really great article. I also that it called out the reality that it becomes a horror show, um, at three x the code output without hardening your other systems. Um, I think that's very much the case. Everyone, I think at this point has been in a, a [00:15:00] situation where, like a deluge of something slammed into a wall, maybe a CI/CD check or a flaky test that was not expected.
[00:15:07] Andrew Zigler: And, and you've probably had a meeting about it, right? So having those kinds of, uh, clear requirements is, is how we can avoid those problems.
[00:15:15] Ben Lloyd Pearson: Yeah, and I'm gonna make a shameless plug just because I can, you know, an Andrew and I, you both work for LinearB, we're, it's a company that helps engineering leaders solve these problems all the time. So, if you're listening to this and or reading this article after this and feel like, man, this is really the experience that I'm going through right now, uh, you know, this is a problem that we help engineering leaders solve all the time.
[00:15:34] Ben Lloyd Pearson: So make sure you go check out LinearB and we'll get back into the news. All right, let's move on to the small web, which may actually be bigger than you think. So the small web, what I'm referring to here are these like non-commercial personal websites that are free of like ads and tracking. And you know, really what it is is going back to like the mid nineties when
[00:15:58] Ben Lloyd Pearson: everyone is like building their own [00:16:00] websites and raw HTML and CSS and, and just doing it to share information with the world. You know, and I remember back then, like, uh, an author of this article brings it up how people were just so idealistic. It's like universities and nonprofits will, will be out there like populating the, the web with knowledge.
[00:16:18] Ben Lloyd Pearson: And to an extent that's true. But then you also look at the, the, the last 20 years or so of the web and it's, it's been more a history of consolidation commercialization, um, near monopolistic and sometimes actual monopolistic behaviors in the market. Uh, so this art, this author is, uh, highlighting a trend where that original v vision of the web is actually starting to trend back in a positive direction and start to come back.
[00:16:46] Ben Lloyd Pearson: So there's more of these websites, showing up. And, and from my understanding, maybe you can correct me, Andrew but I think this author is basing it on, uh, some sort of, uh, like library for building websites or that, that, uh, is now proliferating across the web. And, and [00:17:00] you can use that to sort of track like the emergence of these websites.
[00:17:04] Andrew Zigler: That's
[00:17:04] Ben Lloyd Pearson: but yeah, it's,
[00:17:05] Andrew Zigler: check. It's checking new profiles on cgi, I believe is what it was.
[00:17:08] Ben Lloyd Pearson: yeah. Yeah. Okay.
[00:17:09] Andrew Zigler: are like per a personal website repository.
[00:17:12] Ben Lloyd Pearson: Yeah. And, and I, and I feel like, uh, our, our producer Adam added this because he, he just knew I wouldn't be able to help myself, but go into a nostalgic trip to, to back to my early days of technology. 'cause you know, it has me thinking back to like the very first website that I built, which is back, you know, on our local mom and pop, ISP, they just had their like little building across the field from us.
[00:17:34] Ben Lloyd Pearson: Uh, you know, as a part of our monthly internet subscription, we had hosting space that they gave us. Like literally hosted in our neighborhood, you know, which is pretty wild to think. Uh, so I, you know, I built my own HTML CSS, uh, website, you know, of course optimized for both Internet Explorer and Netscape.
[00:17:50] Ben Lloyd Pearson: And I remember, you know, doing things like downloading game guides off of these websites from people just like me who just wanted to share something that they thought was cool, you know? Uh, so yeah, I, I mean, [00:18:00] centralization tends to ebb and flow. Uh, maybe we are adding back towards decentralization, but what do you think Andrew?
[00:18:06] Andrew Zigler: I, I love that very nostalgic trip back down memory lane. I actually too, I, I had a local little ISP that also too, came with a little bit of hosting,
[00:18:17] Ben Lloyd Pearson: Nice.
[00:18:17] Andrew Zigler: originally. I learned to use a computer so that I could make websites. And why did I wanna make websites? Because I saw these cool pictures and I wanted to do what, save them and collect them in folders.
[00:18:28] Andrew Zigler: And so I wanted to make websites so that I could put my, my collection of images somewhere where people could see them.
[00:18:34] Ben Lloyd Pearson: You've been collecting context for your AI since you first started touching a computer. That's wild.
[00:18:40] Andrew Zigler: Yeah, honestly, same year I was writing, I was on a computer. It's computer's been part of my life my whole life. But it's funny you say that, Adam put this article in here, our producer for you, uh, because he actually definitely put this in here for me because, uh, I feel like I talk about this part of the internet all the time, uh, is there's definitely a resurgence of the small web [00:19:00] here.
[00:19:00] Andrew Zigler: Um, it's a reality called the post naive internet. You know, we pre previously exist in an internet where we were all consolidating into massive
[00:19:08] Ben Lloyd Pearson: Mm.
[00:19:08] Andrew Zigler: in the Web 2.0 era, which then became apps on our phone, became these ubiquitous, huge services. But with the arrival of things like AI and you know, more invasive practices by these more monopolistic platforms, people are retreating with their personal data, with their time, with their attention, and most of all with their creativity. Back to the world of creating personal websites that speak to who they are, and AI is making that easier than ever now for anybody to be able to make their own website. And I think any developer, especially those who in their like earliest days of development, especially anyone near the front end world. Probably your first project that you really wanted to do was make a really cool portfolio website for yourself. And so the idea of having a spot on the web that you own, uh, is really universal to just being a participant on it. I think, and I say more to the [00:20:00] merrier. I have a personal website. I love growing my personal website and connecting with others.
[00:20:04] Andrew Zigler: Web rings are back. You know, I think all of these, uh, these things speak to a, a world of the internet where it's hyper customized for you. It speaks to changing market economics and attention economics as well, because huge platforms that lock you in for advertising, you pay with your attention. But owning your own platform and content means you can do anything with it and connect with like-minded people in ways you never could before. And I maybe as these larger off. The shelf experiences shrink or get replaced by something else. I think we'll find ourselves increasingly in a world where software is very customized and
[00:20:40] Ben Lloyd Pearson: Mm-hmm.
[00:20:40] Andrew Zigler: Like, you
[00:20:41] Ben Lloyd Pearson: Absolutely.
[00:20:42] Andrew Zigler: digital and consumption, pricing based, because all that compute is tokenized underneath because all your experiences are hyper customized to you and exactly where you want them. And so maybe instead of paying for a. A, a c or a flat rate or a tier on some SaaS company somewhere, you might just find [00:21:00] yourself turning on and off a bunch of really cool inference systems that you pull into your own website or you pull into your own shared space with your friends. I think the the web is set up for any kind of future that we want it to be.
[00:21:12] Ben Lloyd Pearson: yeah, and to, to touch on your point, uh, about, you know, a AI encouraging you to, to build more, uh, I, I do kind of feel like maybe we are on the cusp of a wave of decentralization within the web. Like, we've kind of discussed a little bit here about how, it feels like it's probably never gonna be this free again.
[00:21:31] Ben Lloyd Pearson: Like the way you can just use it for a very low cost and do a, a huge range of tasks with it. Um, I'm not sure that that, that that will continue and, and things may over time sort of get locked down. But when you think about the fact that AI makes it so, like basically anyone can write code now or at least code that's good enough for a simple website.
[00:21:51] Ben Lloyd Pearson: Um, coupled with the fact that like hosting space for simple websites is incredibly cheap, uh, sometimes free, uh, and it is quite plentiful [00:22:00] and it's very simple to publish your own content to the web now. So. You know, I could definitely see this creating some level of, of like a renaissance for a, for lack of a better phrase, in terms of just digital creation.
[00:22:12] Ben Lloyd Pearson: And, you know, personally what I'm waiting for. Like let's bring back the free internet arcades. Like, come on. Like, those were such a cool era within the, the internet era. And like with ai, we should be able to do so much better now. Like, am I right Andrew?
[00:22:26] Andrew Zigler: You know, there's a time and a place for it. I, I could see a really great arcade coming back and just really hitting, I gotta say every kid had that bookmarked folder of cool arcade
[00:22:35] Ben Lloyd Pearson: Yeah.
[00:22:36] Andrew Zigler: I just think kids, these days, they just have that bookmarked folder of Roblox experiences
[00:22:40] Ben Lloyd Pearson: Yeah, that's true. That's true. All right, well let's close out with a topic that's near and dear to my heart, and that's sloppy pasta and how we're all gonna stop it. So this article, it, it highlights a, a, an an an emerging workplace etiquette problem that the author refers to a sloppy pasta where people forward [00:23:00] raw AI generated text without reading it or verifying it.
[00:23:04] Ben Lloyd Pearson: Uh, creating an as asymmetry where recipients have to do things like fact check and like distill information and, and spend the effort. Um, understanding the information while the person who sent it spent basically no effort creating it. Uh, and ultimately this is the kind of thing that erodes trust in the workplace.
[00:23:23] Ben Lloyd Pearson: Uh, Andrew, what, what do you think are, have you been a victim or a perpetrator of sloppy pasta?
[00:23:28] Andrew Zigler: I certainly try not to be a perpetrator, you know, as much of, even like a huge
[00:23:32] Ben Lloyd Pearson: Yeah.
[00:23:33] Andrew Zigler: voice of text and whispr flow that I am, I don't even really use it to talk to people in Slack. I just use it for, for code. You know, I'm actually, I've really, really diligent about the time that I put into messages I send to people and my emails.
[00:23:46] Andrew Zigler: You know, if you're listening to this,
[00:23:47] Ben Lloyd Pearson: Yeah.
[00:23:47] Andrew Zigler: received an email from me, you'll know that it has this certain flavor on unhinged and those are all written by me. And so, uh, yeah, I, I definitely think that like. I take a lot of care about the stuff I write. So because of that, [00:24:00] I do in some way kind of expect that from others, especially since we're a human to human interaction.
[00:24:05] Andrew Zigler: It doesn't ever really feel good to feel like you're just getting a copy paste outta someone's like, ChatGPT session. Uh, I've definitely been on the receiving end of this. And if you do it the other folks and you don't think that they notice, I promise you that they do.
[00:24:19] Andrew Zigler: about you, Ben? Are, do you find yourself perpetrating or being perpetrated?
[00:24:23] Ben Lloyd Pearson: Yeah, so, so first of all, I'll explain there's, there's a couple of categories of sloppy pasta perpetrators that are outlined here. So the first is the eager beaver. So that's somebody who, who wants to contribute to the topic at hand. Uh, so they go and ask their chat bot and just share whatever that chat bot shares with them.
[00:24:39] Ben Lloyd Pearson: You know, the intention's good, not necessarily helpful. Uh, then the second category is the Oracle. Uh, and to be clear, I feel like I have seen, at least seen all three of these, and I'll get into the one that I think that, that I may be as on occasion. Uh, so the Oracle, this is someone who, asks a question and then somebody else just takes that question [00:25:00] and goes and pastes it into to, to their a AI and brings it back into the chat.
[00:25:05] Ben Lloyd Pearson: Uh, sometimes this is valid though. I will point out there are, there have been times where I, I kind of wish I had the, um, the AI version or the AI equivalent of, let me Google that for you. It's like you could have just taken that to, to chat GPT before you brought it to me. But, but that's besides the point.
[00:25:19] Ben Lloyd Pearson: That's a different situation. Alright. And then the last category, which is the one that I think I do occasionally fall into, and that is the ghost writer. Uh, and this is where the sender shares AI output as their own work. And, but you know what, and I've gotten this, this feedback directly from you, Andrew.
[00:25:34] Ben Lloyd Pearson: You once told me this is. This is, I don't remember the exact words, but it was a like a two or three page document that was purely AI generated and you're like, this is the best kind of AI slop. And I was like, of course it is.
[00:25:48] Andrew Zigler: Well, the, the AI slop that you serve, Ben, is great. It's just, you know, maybe smells a little bit of ai, but, you
[00:25:54] Ben Lloyd Pearson: Yeah. Yeah.
[00:25:55] Andrew Zigler: I will say, you know, I think everybody who worked does knowledge work stuff, does the ghost writing [00:26:00] thing to an extent. I read the ghost writing thing and it's like, oh, that's like, uh, oh my, like my agents do that.
[00:26:05] Andrew Zigler: But. And re, or not like saying to other people, but produce artifacts just like this that are like my outputs. Right? So it's just more about like properly addressing where the content's coming from. Framing expectation, being like if this was AI generator or part of a workflow, I think it's. Helpful to disclose that. A lot of times
[00:26:23] Ben Lloyd Pearson: yeah,
[00:26:23] Andrew Zigler: are implied to be that way now, though, so it just depends.
[00:26:27] Ben Lloyd Pearson: Yeah. It.
[00:26:28] Andrew Zigler: Definitely fall in the trap of perhaps being the eager beaver, but I will say that it's just really, uh, it speaks to the ability to have like instant finger, uh, tips on your, the information that you need. And so if you're a really eager beaver and you're not well attuned to what you're
[00:26:46] Ben Lloyd Pearson: Yeah.
[00:26:47] Andrew Zigler: then that is a, a. Very much a detriment to your team. But if you're an eager beaver and able to spot and be calibrated to the right thing, the right information, the signal, you know, I, I think that's less of an eager [00:27:00] beaver and more of like a, a like a dog, right? Like you're gonna go out like, find what, what needs to be found.
[00:27:05] Andrew Zigler: It's like a could be intentional. So all of these have a positive side as well.
[00:27:09] Ben Lloyd Pearson: Yeah. And, and there's some great tips in this article too that, that, uh, if you think you might be in one of those three categories, or you, you want to help somebody who, who is, and, you know, articulate how, how to get better, uh, there's some great tips. Uh, there's a couple that resonated with me that I, I have definitely incorporated into my daily work.
[00:27:28] Ben Lloyd Pearson: Um, and the fir like first, first of all. Your first pass of anything that comes out of ai, you should effectively, like judiciously delete things out of it. Like, uh, you know, models like Claude in particular are, are incredibly verbose. It's, I found that like, potentially like 20% of what it outputs just off the bat, just get rid of it.
[00:27:47] Ben Lloyd Pearson: It's not relevant. Like it, it, it just went a little too far, uh, because it can, Uh, and, you know, and that helps like distill like you're trying to distill down to like, the important stuff and just get, get rid of all this stuff that doesn't really add value. [00:28:00] Um, and then the second one that is, I think it, it, it's, it's perhaps more critical than the first is you have to validate every single factual claim.
[00:28:08] Ben Lloyd Pearson: Um, and then fix spots where AI missed context. So sometimes AI will just get the facts wrong, um, which you need to go validate and figure out. Um, sometimes it will add context to a fact that it, it comes up with that the fact may be true, but the context is out of place. there's a lot of assumptions that LLMs make that, you know, you just need to use your, your human expertise to validate.
[00:28:30] Ben Lloyd Pearson: So, and there are other tips in this article. I, I feel like that the problems that, that these three categories of, of perpetrator create, they're all solvable. You know, it's just a matter of rigor and how you work.
[00:28:42] Andrew Zigler: I agree.
[00:28:43] Ben Lloyd Pearson: All right, Andrew, what are your agents up to this weekend?
[00:28:46] Andrew Zigler: Oh, well, I guess they're gonna be, uh, trying to work between the rolling token outages that are gonna be hitting the nation.
[00:28:51] Ben Lloyd Pearson: all right.
[00:28:52] Andrew Zigler: but, uh, not, probably not all too much to be honest, because I find myself more and more the more [00:29:00] capable that they get. It's more again, about the communication in the alignment for what we need at any given moment. You know, I don't exist within a massive churning enterprise system, at least not right now. And so, I don't have this constant h like output demand from a huge array of tools and services that I manage or maintain. Instead, all of the knowledge work I do requires a lot of
[00:29:22] Andrew Zigler: thinking. And so what I find is my agents free me up a lot of time to really get aligned on what exactly
[00:29:29] Ben Lloyd Pearson: Yeah.
[00:29:29] Andrew Zigler: done, and then when the coding happens, it's actually so surgically fast because my harness is so well tuned from everything that I learned here from our experts on the show, that I actually don't spend too much time there at all.
[00:29:40] Andrew Zigler: But what about you, Ben?
[00:29:41] Ben Lloyd Pearson: so I, I just got a new laptop, which has been, it's been fun setting it up in the, the, like agent first era, you know? most work every day, like I'm starting with working with. Agents first. Like, it's like, what do I have to build with my agents to solve the things that I've put on my to-do list today?
[00:29:58] Ben Lloyd Pearson: Uh, and it's a, it is a, [00:30:00] it is a pretty wild way to work,
[00:30:02] Andrew Zigler: You know, I'll, I'll answer one better. It's not that my agents will be making anything. I will be spending a lot of time creating the right environment
[00:30:10] Ben Lloyd Pearson: Yeah.
[00:30:11] Andrew Zigler: to know. What to do. All of the agents and, and, and what they need to do are effectively pretty efficient for me. But what I spend a lot of time doing now is curating their environment, making sure the right information's available, the right tools are at their fingertips,
[00:30:24] Ben Lloyd Pearson: Yeah.
[00:30:25] Andrew Zigler: communicate with the right other agents and becomes like a whole other practice.
[00:30:29] Ben Lloyd Pearson: It's almost like tending a garden, you know? Isn't that what it feels like? Yeah. Yeah. Cool. All right. Well thanks everyone for joining us for this segment of the the Friday Deploy. Uh, give us a rating on wherever you're listening to us right now with thumbs up. You know, anything you can do to, to just helps us in drive engagement and get, get our reach out there.
[00:30:47] Ben Lloyd Pearson: So thanks for joining us this week and thanks Andrew, and we'll see you next week.
[00:30:52] Andrew Zigler: see you next time.
