Buong transkripsyon ng S8E6 Documentation Is Becoming the Interface with Niko Berry

KOPYA NG SALIN 🔗

WEBVTT 00:00:02.240 --> 00:00:06.878 Hello friends and welcome to season eight of Inside Tech Comm with Zohra Mutabanna. 00:00:06.878 --> 00:00:16.160 This season features a collection of conversations that explore the ideas, challenges, and opportunities shaping technical communication today. 00:00:16.160 --> 00:00:21.838 From AI and content strategy to leadership, product thinking, and the future of our profession. 00:00:21.838 --> 00:00:26.160 Each episode offers a fresh perspective from people doing the work. 00:00:26.160 --> 00:00:27.039 Let's get started. 00:00:27.039 --> 00:00:34.719 Season 8 is proudly sponsored by LavaCon. 00:00:34.719 --> 00:00:41.679 Use discount code ITC26 to save $200 on registration for this year's conference. 00:00:41.679 --> 00:00:44.960 I'll share more about LavaCon later in this episode. 00:00:44.960 --> 00:00:50.960 Hello, listeners. Welcome to another episode of Inside Tech Comm with Zohra Mutabanna. 00:00:50.960 --> 00:00:53.840 I have with me here Niko Berry. 00:00:53.840 --> 00:00:59.359 And I was very fortunate to actually get introduced to him through Manny Silva. 00:00:59.359 --> 00:01:03.118 We were at the Write the Docs conference in Portland back in May. 00:01:03.118 --> 00:01:10.400 And Niko and I started talking, and then I kind of learned that he's moved from technical communication into product management. 00:01:10.400 --> 00:01:12.480 And that just got me curious. 00:01:12.480 --> 00:01:15.519 And I invited him to join my show. 00:01:15.519 --> 00:01:18.879 And he was very kind enough to accept that. 00:01:18.879 --> 00:01:20.560 So here we are today. 00:01:20.560 --> 00:01:23.040 And welcome to my show, Niko. 00:01:23.040 --> 00:01:23.840 How are you doing? 00:01:24.319 --> 00:01:24.718 I'm doing well. 00:01:25.200 --> 00:01:29.280 Niko, tell us a little bit about yourself, your journey, and what brought you where you are today. 00:01:29.840 --> 00:01:30.319 Hi, I'm Niko. 00:01:30.319 --> 00:01:33.438 I am not actually a technical writer anymore. 00:01:33.438 --> 00:01:40.959 Just to give the quick background on this, I got into technical writing kind of on accident after failing to be a journalist. 00:01:40.959 --> 00:01:52.799 And from there, I found that contrary to you know the famous blog I'd rather be writing, it was actually every other part of the technical job that I enjoyed most. 00:01:52.799 --> 00:01:58.799 And so what it turned out was that there was a lot of overlap between that skill set and product management. 00:01:58.799 --> 00:02:06.640 And so eventually somebody at the company I was at had the bright idea to give me a shot as a product manager, and it just kind of went from there. 00:02:06.640 --> 00:02:09.598 But I remain really interested in technical writing. 00:02:09.598 --> 00:02:12.800 I've remained in this developer experience niche. 00:02:12.800 --> 00:02:19.038 And really, I'm gonna say 60% of developer experience is just having good documentation. 00:02:20.158 --> 00:02:22.800 That's such a refreshing thing to hear from product management. 00:02:22.800 --> 00:02:25.360 So uh thanks for that, Niko. 00:02:25.360 --> 00:02:27.199 What were the skills? 00:02:27.199 --> 00:02:30.479 You said that there were skills that you kind of leaned into. 00:02:30.479 --> 00:02:39.680 For somebody who doesn't know what product management is, what would you say were the transferable skills and a little bit about product management itself in your context? 00:02:40.800 --> 00:02:42.080 So, what is a product manager? 00:02:42.080 --> 00:02:53.919 The way that I would describe the core responsibility of a product manager is the person who tells developers and really the business in general what we need to build next and why we need to build it. 00:02:53.919 --> 00:03:01.840 And so within that, there are a number of things that are complemented quite well by the technical writer's skill set. 00:03:01.840 --> 00:03:24.400 The first one is just clear technical storytelling, being able to articulate to three, four, five, however many different audiences within a company, what you are about to build at different levels, whether that's talking about the low-level stack to developers or just talking about the highest level business implications of it to an executive leadership team. 00:03:24.400 --> 00:03:34.800 And beyond that, the other core part of the job is talking to people and getting information from them that may not be easy to initially understand. 00:03:34.800 --> 00:03:38.800 And that's another very common skill set for technical writers. 00:03:39.840 --> 00:03:40.318 That's right. 00:03:40.318 --> 00:03:45.199 So as you kind of moved into product management, how long have you been doing this? 00:03:45.199 --> 00:03:52.800 Because I kind of want to understand the pre-AI and the post-AI shifts that may have happened with your role in product management. 00:03:53.360 --> 00:03:53.758 Yeah, definitely. 00:03:53.758 --> 00:03:56.878 I've been in product management for about five years. 00:03:56.878 --> 00:04:04.718 I got in right as the pandemic started, and then have been in here to some extent for a half a decade since. 00:04:04.718 --> 00:04:13.038 And the pre-AI and post-AI like phases of it have been, I'm trying to decide whether they have been all that different, honestly. 00:04:13.038 --> 00:04:31.838 I think that like the core role of the product manager has not really changed all that much, but AI has provided a number of tools that complement it, accelerate it, and provide ways to do it with a new skill set and a new tool set. 00:04:31.838 --> 00:04:40.160 But really, at the end of the day, I've seen a lot of people discuss, you know, product managers taking better to AI than many careers do. 00:04:40.160 --> 00:04:53.680 And I think that a portion of it is just that like a lot of product management is already talking to somebody else, having them make something on your behalf, and then having to stand by the results

Load full transcript

of it however it comes out. 00:04:53.680 --> 00:05:06.160 And I think that that's some a reason why product managers are a lot more comfortable with like what AI does than a lot of other career paths, where AI produces slop in a lot of cases. 00:05:06.160 --> 00:05:18.399 It takes a lot of work to get it to do what I would describe as C work for a developer or a technical writer, but I've already got practice standing by work that I can't really like do myself. 00:05:19.199 --> 00:05:19.439 All right. 00:05:19.439 --> 00:05:29.439 You mentioned accelerated and with the new tool sets, and that's something that I'm also experiencing as a technical writer at the moment. 00:05:29.439 --> 00:05:35.439 That said, with the acceleration, have you noticed a drop in quality of content? 00:05:35.439 --> 00:05:37.040 Has it gotten better? 00:05:37.040 --> 00:05:39.920 What has the outcome been with AI? 00:05:39.920 --> 00:05:42.800 I'm assuming I think I'm kind of jumping ahead. 00:05:42.800 --> 00:05:43.838 So is there AI? 00:05:43.838 --> 00:05:49.199 Are the technical writers on your team using AI tools to generate content? 00:05:49.199 --> 00:05:52.079 And if so, what has the outcome been? 00:05:52.720 --> 00:06:03.759 Yes, uh, my team is using AI tools not for generation of content necessarily, but for a lot of parts of the research and initial drafting phase. 00:06:03.759 --> 00:06:13.199 And the reason we don't just go straight into directly generating content so far has been because the quality is just lower. 00:06:13.199 --> 00:06:15.759 Like, hallucinations are real. 00:06:15.759 --> 00:06:25.439 There isn't a like, unless you are testing the uh claims that your AI assistant is making, you have no way of knowing if it's correct or not. 00:06:25.439 --> 00:06:27.600 It's also just really verbose. 00:06:27.600 --> 00:06:29.838 Like, it's a pain to read. 00:06:29.838 --> 00:06:48.560 Also, another part of it, too, which is just that like in a lot of cases, the problem that like certain sets of docs have is just that they have gaps in it, that there are more surfaces of the product to cover than you have technical writers to do it or time to do it or the ability to prioritize. 00:06:48.560 --> 00:06:59.519 And I think that there is a determination that like product managers and technical writers need to make about certain portions of the docs of is something better than nothing. 00:07:00.319 --> 00:07:18.399 Yeah. So in your current role, with the acceleration that's happening in most tech companies, what I'm hearing and what I'm experiencing is the amount of work that comes your way in the pipeline has just kind of accelerated as well, exponentially gone up. 00:07:18.399 --> 00:07:25.040 And so, from a product management perspective, are you expecting anything differently from technical communicators? 00:07:25.040 --> 00:07:28.560 And is that something that your team also experiencing, perhaps? 00:07:29.439 --> 00:07:44.560 You know, that's a question

that I should have been asking myself sooner, because I don't think my expectation around technical writers changed, but I think the context around it has changed enough that secretly it means my expectation has actually changed. 00:07:44.560 --> 00:08:04.639 Like what I mean is if my expectation is, hey guys, you make sure that all of the new features are documented, and if anything is like breaking our SLAs or there's anything critical in the current docs, go back, update it, try to like make the docs better where you can, practice Boy Scout documentation where you leave things cleaner and better than you found them. 00:08:04.639 --> 00:08:12.959 But if like my engineering teams are producing, you know, 1.5 times as many features, then really the expectation has gone up too. 00:08:12.959 --> 00:08:15.040 Sort of a blind spot on my part, I guess. 00:08:16.160 --> 00:08:17.278 becomes important as well. 00:08:17.278 --> 00:08:20.560 And that's how I think in our conversation we came up with the headless CMS. 00:08:20.560 --> 00:08:27.920 So kind of taking a step back and giving that context to our listeners, but then also talking about what a headless CMS is. 00:08:27.920 --> 00:08:29.920 Would you like to check on that question? 00:08:29.920 --> 00:08:32.399 You know, I'd like to pull apart that thread. 00:08:33.440 --> 00:08:33.678 All right. 00:08:33.678 --> 00:08:35.678 So content management systems. 00:08:35.678 --> 00:08:42.158 I'm just gonna go out on a limb and assume that an audience of technical writers already knows what a content management system is. 00:08:42.720 --> 00:08:43.360 They should know that. 00:08:43.360 --> 00:08:43.678 Yes. 00:08:43.678 --> 00:08:48.879 And if they don't, then we can always use amazing tools out there to learn more about it. 00:08:50.080 --> 00:09:00.320 So in the past, content management systems would entangle the content that you wrote, but then also tie in a bunch of technology about its presentation. 00:09:00.320 --> 00:09:19.038 The main form that this took was you had WYSIWYG editors, where as you make changes to it, it's creating this completely unreasonable, totally inaccessible HTML that is tightly bound to like the website that you are using right now. 00:09:19.038 --> 00:09:38.798 And so at a certain point in this journey, somebody had the bright idea of what if we just completely remove content from this front or from the front end and have it be stored in a form that makes no opinions about how it is presented or what technology should be presenting it. 00:09:38.798 --> 00:09:42.639 A content management system without the head on top of it, effectively. 00:09:42.639 --> 00:09:47.038 And this idea took off a decent bit. 00:09:47.038 --> 00:09:54.879 Currently, I'm actually product or a product manager at Contentful, which I believe was the first commercial headless CMS. 00:09:54.879 --> 00:10:15.360 And the reason why is because when you have this kind of agnostic approach to delivering your

content, it is able to be quickly adapted to all sorts of front ends, whether that's a new web framework or your mobile app, or maybe like digital signage in a store. 00:10:15.360 --> 00:10:24.000 And now the most recent thing that has come out as like a channel for distributing content and the consumption of content is AI. 00:10:24.879 --> 00:10:26.798 Can you give a more real-world example? 00:10:26.798 --> 00:10:29.440 I think for my benefit as well. 00:10:29.918 --> 00:10:33.038 Tell me a little bit more about what you want the real-world example to be, exactly. 00:10:33.038 --> 00:10:33.440 real world exactly. 00:10:34.000 --> 00:10:40.240 I think you kind of give gave a good definition and you talked about how you pulled out content from uh from that framework. 00:10:40.240 --> 00:10:45.600 And then you talked about how you would publish that content, essentially. 00:10:45.600 --> 00:10:53.600 So when we are removing content, like what exactly is happening there? Like I think I'm trying to get a little more technical there. 00:10:54.158 --> 00:11:35.678 So what is happening when you're removing content from like the front end is you are starting to move from HTML that is something like a div with a style class applied to it, or a bunch of like indicators of how something should be styled, to instead an object, probably JSON, on your server that is like call to action, and then here is the text of it, where you start to associate the content with its semantic purpose, and then how that is articulated to the customer on their client is a matter for the UI or for whatever is consuming content from the headless CMS. 00:11:35.678 --> 00:11:37.678 I hope that is making sense. 00:11:38.240 --> 00:11:48.639 So, you know, when we've removed content here, in in simple words, the formatting is now no longer something that is part of the content itself. 00:11:48.639 --> 00:11:49.918 That is separated. 00:11:50.639 --> 00:11:50.960 Yeah, right? 00:11:50.960 --> 00:11:52.480 In a perfect world, yes. 00:11:53.278 --> 00:11:54.320 In a perfect world, right. 00:11:54.320 --> 00:11:57.600 And we are talking about the headless CMS here now, right? 00:11:57.600 --> 00:11:59.519 Where this framework applies. 00:11:59.519 --> 00:12:08.158 With AI, with agentic workflows that we are starting to kind of interact with, I think this lends well. 00:12:08.158 --> 00:12:09.840 What is your take on that? 00:12:10.399 --> 00:12:12.720 Yes, I think that it lends well as well. 00:12:12.720 --> 00:12:40.399 I think that being able to have content with a clear purpose attached to it rather than a purpose that needs to be inferred from the LLM looking at the raw markdown of it, it tends to be able to allow the LLM to make better assumptions about the content and articulate it in better ways to whoever is having the LLM work as a middleman between them and like the content that it's consuming. 00:12:41.120 --> 00:12:41.918 Yeah, that makes sense. 00:12:41.918 --> 00:12:49.038

So we've talked a little bit about the headless CMS and why you know people may be paying attention to it. 00:12:49.038 --> 00:12:53.759 And then we also talked about how it lends better to the new workflow. 00:12:54.720 --> 00:12:58.158 But I think that there's like one other part of it, which is beyond content. 00:12:58.158 --> 00:13:07.360 When you start to think about software as headless, then AI can become an entirely new layer of user interface on top of it. 00:13:07.360 --> 00:13:20.158 At present, one of the beautiful things about Contentful is that we provide all of the APIs that our UI uses to customers to like write scripts for if they wanted to. 00:13:20.158 --> 00:13:35.759 And what this has meant is that if this is the approach that you want to take for using the software, you can either install our command-line interface and give that to Claude Code, or download our MCP server, a Model Context Protocol server. 00:13:35.759 --> 00:13:39.360 I'm assuming a fairly technical audience who already knows what that is. 00:13:39.360 --> 00:13:49.678 Give that to either Claude or ChatGPT or any LLM that supports it, and you can just start asking it to do things within the system, and it's able to do it. 00:13:50.399 --> 00:14:00.080 When organizations start managing content, it seems like we are starting to look at content as a reusable component instead of a document. 00:14:00.080 --> 00:14:03.678 Would that be accurate in this scenario? 00:14:04.558 --> 00:14:19.678 Yes, and I think that part of it is a philosophic question that has emerged that every organization is going to need to deal with if they want to use their existing content and AI effectively, is to what extent is content context? 00:14:20.798 --> 00:14:21.840 Content okay. 00:14:22.639 --> 00:14:32.399 By which I mean to what extent can you just take your content and throw it into an LLM and expect it to produce useful results when you hand it to a user? 00:14:32.399 --> 00:14:36.480 To what extent do you need to add additional things that are not already in it? 00:14:36.480 --> 00:14:47.918 Whether that's additional meta-elements, skills, or extra instructions about how the LLM needs to do things differently from how a user would if they were doing it through the UI. 00:14:48.960 --> 00:14:53.278 But you consider those as the reusable part of a content chunk? 00:14:53.278 --> 00:14:54.480 Okay. 00:14:54.480 --> 00:15:02.320 And that kind of also, again, when you look at it in a more broader sense, that's part of structured content, right? 00:15:02.320 --> 00:15:05.759 Where you have content that is enriched with context. 00:15:06.720 --> 00:15:25.840 Yes. I would say like, so just going into my personal philosophy on it, I think that content and context are almost one-to-one, but there is more to context than just content, because in many cases, context can be things that are shifting and need to be assessed in a sort of like at the moment manner. 00:15:25.840 --> 00:15:33.038 It's things like a call to an API that gets you information about the status of a system right now. 00:15:33.038 --> 00:15:38.720 Or it could be like just a running log from something else. 00:15:38.720 --> 00:15:50.639 And by comparison, content is it is reusable, but it's also something that is sort of expected to stay mostly the same unless a human comes in and updates it. 00:15:51.200 --> 00:15:53.918 You had also mentioned content as input. 00:15:53.918 --> 00:15:56.558 What did you mean by that? 00:15:57.038 --> 00:15:58.080 I think I'm starting to remember. 00:15:58.080 --> 00:16:07.759 It was the question of if your documentation is effectively being consumed by middlemen who re-articulate it to users. 00:16:07.759 --> 00:16:16.399 And there is actually value in that because a user can bring additional context to an LLM for your docs. 00:16:16.399 --> 00:16:19.360 They can say things like, look, I'm not technical. 00:16:19.360 --> 00:16:33.519 I'm trying to figure out like we got two developers on my marketing team, and I want to see about updating the way that like our content management system handles publishing blog posts so that it can do a few additional things automatically. 00:16:33.519 --> 00:16:36.639 I don't speak developer. 00:16:36.639 --> 00:16:39.678 Can you explain this to me like I'm five? 00:16:39.678 --> 00:16:42.558 Or explain it to me in terms that are easier to understand. 00:16:42.558 --> 00:16:50.720 Or on the other hand, you can show up and say, I'm a developer, I'm working in the C# stack, don't bring me any JavaScript examples. 00:16:50.720 --> 00:16:56.158 I want this all interpreted into like the tech stack that I'm working in. 00:16:57.038 --> 00:17:04.880 with one of the things that I want to ask you, and you may not have the answer, all of us are starting to feel this fear about AI. 00:17:04.880 --> 00:17:08.640 Has that been the case in your field as well? 00:17:08.640 --> 00:17:10.798 Or you feel like it has empowered you? 00:17:10.798 --> 00:17:11.680 AI has empowered. 00:17:12.480 --> 00:17:14.240 I think that it's a very complicated question. 00:17:14.240 --> 00:17:24.078 To a certain extent, I definitely feel fear because this is a technology where—when did ChatGPT get released? 00:17:24.078 --> 00:17:25.200 Was it like 2022? 00:17:26.558 --> 00:17:27.358 Yeah, end of 2022. 00:17:28.960 --> 00:17:34.798 And it came out and was just like, oh, this can write sometimes correct code if you prompt it to. 00:17:34.798 --> 00:17:40.558 And it can like talk with you, and sometimes it gets things right, but really it hallucinates all the time. 00:17:40.558 --> 00:18:02.240 And then you cut forward four years, and you have these like incredibly complicated or not complicated, competent models, Claude Fable, where I can just tell it to recreate some video games in the HTML5 canvas element from scratch, and it's able to get really close on a single attempt. 00:18:02.240 --> 00:18:06.798 It's crazy how

far it came in such a short amount of time. 00:18:06.798 --> 00:18:18.078 And I think that there's just the sort of feeling of it's not gonna replace me now, and if the gains are modest and it sort of stalls out here, I think I'm definitely safe. 00:18:18.078 --> 00:18:20.960 But does anybody know what's gonna happen with it? 00:18:20.960 --> 00:18:23.920 And I think that that's like one source of anxiety. 00:18:23.920 --> 00:18:39.358 The other bit of fear that I have about AI is less about just losing my job and more about getting so or AI making me so fast at it that I just have my attention fragmented across 20 different projects all the time. 00:18:39.920 --> 00:18:52.960 And I think that's a very valid concern because we were talking about it at work, and anybody that I talk to who's from the tech industry and using AI is fragmentation and the challenge to critical thinking. 00:18:52.960 --> 00:18:57.200 Because if you're going to be that fragmented, how does that impact your critical thinking? 00:18:57.200 --> 00:19:03.680 We've been we've been making the case that AI will free you up so that you have you can do more high-level thinking. 00:19:03.680 --> 00:19:08.078 But on some days, I feel like that's not true at all. 00:19:08.720 --> 00:19:22.000 Definitely. And a question that I've always had about AI since like we started applying it to workflows is is there more you can do with it than make this devil's bargain of go faster but with much lower quality? 00:19:22.000 --> 00:19:32.160 And like with a question mark around, are you even going faster if the quality starts to get so low that you need to come in and correct it over and over and over again? 00:19:32.160 --> 00:19:37.440 I've got some answers there that point towards a potentially optimistic direction. 00:19:37.440 --> 00:19:47.038 And one of them is that AI does provide some novel mechanisms for being more deliberate about your thinking, but it does make you slower. 00:19:47.038 --> 00:19:53.200 I'm gonna point you I'm gonna point you and your audience towards this uh skill for Claude called Grill Me. 00:19:53.200 --> 00:19:55.358 It's worth looking up right now. 00:19:55.358 --> 00:19:56.798 It is really cool. 00:19:56.798 --> 00:19:57.358 Yeah. 00:19:57.358 --> 00:19:58.640 Grill me. 00:19:58.640 --> 00:20:01.519 Cook out, grill in, grill me. 00:20:01.519 --> 00:20:03.200 Let me grab it and I'll send you a link. 00:20:03.200 --> 00:20:11.358 Grill Me is interesting in that I'm just going to explicitly state if you use this, this will make you slower at a task. 00:20:11.358 --> 00:20:52.318 But what it does is it prompts your assistant to look at all of the context it can get about a project that you are currently on, and then just starts asking you hard questions about it, forcing you to really think through it, examine it from different angles, explain the parts where you used an abstraction, or put forward a like term that doesn't seem to be fully defined, and just make you be explicit about everything, get your thoughts down onto paper, and really force you to think through what you are proposing from end to end. 00:20:52.318 --> 00:21:12.400 And in product management, this is just absolutely fantastic because you have to deal with so many different potential stakeholders on your product, and so many like different people who want to know and make sure that their opinions are considered for the thing that you are about to build. 00:21:12.400 --> 00:21:30.798 And I've found that this is the best way for me to like start to consider those opinions up front, even when it's the end of the day and I'm burnt out and under pressure and all of these other things that starts to really narrow your field of view while you're working on something. 00:21:31.519 --> 00:21:35.358 You mentioned something about being deliberate that can come with AI. 00:21:35.358 --> 00:21:39.920 And that got me thinking that maybe that's true. 00:21:39.920 --> 00:21:44.400 Uh, you do have to be very deliberate when you are using AI. 00:21:44.400 --> 00:21:55.118 And in my context, right now, with the costs escalating, companies are starting to clamp down on your AI tools usage, right? 00:21:55.118 --> 00:21:58.078 You have you so you have to be deliberate right there because of the cost. 00:21:58.078 --> 00:21:59.519 So now I think about whether. 00:21:59.519 --> 00:22:00.880 Am I going to use AI? 00:22:00.880 --> 00:22:01.680 What are the scenarios? 00:22:01.680 --> 00:22:04.078 Does this consume a lot of tokens? 00:22:05.038 --> 00:22:07.118 It depends on what you're doing with it. 00:22:07.118 --> 00:22:12.640 Like typically I know my token cost per month for like my actual day job. 00:22:12.640 --> 00:22:20.480 And there has never been a month as a product manager where I've gone above $250 of tokens spent. 00:22:20.480 --> 00:22:22.640 This is like paying through the API. 00:22:22.640 --> 00:22:25.118 And part of that is just that I don't use Opus. 00:22:25.118 --> 00:22:34.880 I basically have reserved Opus for a few very high value tasks in touching the code base. 00:22:34.880 --> 00:22:46.798 And apart from that, it's Sonnet all the way with a few subagents that use Haiku for summary or other things that can be token intensive and not terribly difficult. 00:22:46.798 --> 00:22:53.838 I think that using this with Opus could create problems and might get you on like your accountants, you know, shitlist. 00:22:55.278 --> 00:22:56.640 Yeah. And I do the same thing. 00:22:56.640 --> 00:23:00.078 So the Grill Me, how did you discover this? 00:23:00.640 --> 00:23:10.400 Honestly, every morning when I wake up, I pour myself a cup of coffee and I open up a Y Combinator's Hacker News and just read through the top 20 articles for the day. 00:23:10.400 --> 00:23:12.480 And this was in there. 00:23:13.118 --> 00:23:22.000 Okay. And can you share when you used like an aha moment when you used Grill Me and it really helped you? 00:23:22.640 -->

00:23:29.200 Yes. So one of the things that I am in charge of as product manager at Contentful is our Model Context Protocol Server. 00:23:29.200 --> 00:23:40.240 And I started using Grill or I used Grill Me when I was putting together the state of the product document for the current quarter that we are in. 00:23:40.240 --> 00:23:43.838 And it made me talk about what we are currently seeing. 00:23:43.838 --> 00:23:51.598 It forced me to go into our data and answer some specific questions for it that ended up going into this state of the product document. 00:23:51.598 --> 00:24:04.480 And then ultimately at the end of it, the aha moment that it led me to was a question about how granular we were or weren't being with the audience of users for the Model Context Protocol server. 00:24:04.480 --> 00:24:10.640 Because the main goal that I have had in building it out is this is a strange way to put it. 00:24:10.640 --> 00:24:16.798 When you work in like B2B enterprise SaaS, you're making some of the most boring software in the world. 00:24:16.798 --> 00:24:33.598 And at best, the like good human moments that make it worthwhile are when you can make something that makes somebody's job so much easier, or like gets rid of some task that annoys them a lot, and they go, oh, so much better. 00:24:33.598 --> 00:24:59.038 Providing the Model Context Protocol server and Claude to like the common users who aren't developers that use Contentful, I could visibly see their faces light up and like joy hit them, that they had hit a point where they could start to describe what they wanted to have done with their content strategy and their marketing strategy, what they needed to do to maintain it mechanically through our software. 00:24:59.038 --> 00:25:15.838 And it can just happen without them needing to remember how the UI works, how to like click through 20 different nested menus and grapple with the not great parts of having a UI that can serve highly complicated use cases. 00:25:15.838 --> 00:25:26.400 And so what this early realization led me to was going, how can we get this into the hands of more people who are not developers? 00:25:26.400 --> 00:25:54.798 And ultimately, what my grill session led me to was, and I hope that this isn't just like super obvious, that the actual thing that needs to happen is making all of these different governance cases to control what an agent is able to do and be able to hand it out to people with like the confidence that even if they don't know what they're doing, they can't cause serious damage to like the setup for your company. 00:25:54.798 --> 00:26:07.759 And really focusing on providing that to developers with clear messaging about why they should care about it and what they should be building for their internal team. 00:26:08.180 --> 00:26:11.880 Yeah, that's such a valid point. 00:26:11.880 --> 00:26:17.191 The episode is sponsored by LavaCon, the content strategy conference taking place this October in Charlotte, North Carolina. 00:26:17.191 --> 00:26:23.565 With more than 70 sessions and workshops featuring speakers from companies like Salesforce, TikTok, T-Mobile, and Amazon. 00:26:23.565 --> 00:26:31.000 LavaCon brings together content strategists, documentation leaders, and content professionals to share practical ideas you can apply right away. 00:26:31.000 --> 00:26:37.678 And yes, it's known as the Fun Conference, too, with networking events, therapy dogs, comfort llamas, storytelling, and karaoke. 00:26:37.678 --> 00:26:41.700 Use discount code ITC26 to save $200 on registration. 00:26:41.700 --> 00:26:59.680 one of my questions was how are customers thinking about AI generated content, or if they are being, you know, through your tools, being uh using the MCP protocol and able to generate their own content using your content as input, what has their experience been like? 00:26:59.680 --> 00:27:04.480 So this example that you shared kind of illustrates that aha moment for them as well, probably. 00:27:05.358 --> 00:27:14.558 And I want to be clear, the thing about this is also something that applies to technical writing as well, but a lot of the customers that we have are content writers and marketers. 00:27:14.558 --> 00:27:21.358 And when they ask me about using AI to generate content, I actually encourage them not to do that. 00:27:21.358 --> 00:27:44.720 Because AI generated content, unless you are very careful with it and very deliberate about having lots and lots of highly refined content that you're inputting to it, and like you're basically just generating little variants for different markets, different times of year, just like the last 10% of doing content strategy. 00:27:44.720 --> 00:27:57.680 If you don't do that, the content will come out sounding like AI. And there's a decent amount of evidence showing up in the industry that content that sounds like AI just makes your brand less trustworthy. 00:27:59.278 --> 00:28:06.160 And that's where the human in the loop, human oversight, governance, transparency, all these things start to matter. 00:28:06.160 --> 00:28:12.798 Have you had any pushback from customers at all that they don't want any that is? 00:28:13.278 --> 00:28:13.680 Yes, definitely. 00:28:13.680 --> 00:28:22.480 There are plenty of customers, like I would say a decent third-ish of our customer base just doesn't want to touch the stuff. 00:28:22.480 --> 00:28:26.318 And you know, we accommodate that and provide a thorough off-switch for it. 00:28:26.318 --> 00:28:33.838 I think that both approaches are valid depending on what you are trying to accomplish and which market segment you are in. 00:28:33.838 --> 00:28:45.038 Like there are places where using AI makes a lot of sense because what's going on is you're writing content that you need to get to rank on long tail SEO keywords. 00:28:45.038 --> 00:28:52.960 And a lot of what you need to do is just take some core themes of your content and have dozens of variations.

00:28:52.960 --> 00:29:08.400 It's things like if you are a food company and you make hot dogs, there are only so many ways you can provide recipes and instructions for making hot dogs, but you can always find ways to connect it into like which season is it right now? 00:29:08.400 --> 00:29:10.480 What holiday is coming up next? 00:29:10.480 --> 00:29:11.759 Things like that. 00:29:12.400 --> 00:29:23.200 Okay. So the resistance that some customers thought of the customers, so I understand that market segments definitely will determine if they want to use AI or not. 00:29:23.200 --> 00:29:39.118 But are there customers who you think are in a segment, market segment that would benefit, but are pushing back perhaps because either they are not AI fluent or just AI resistant and being more old school? 00:29:40.078 --> 00:29:42.400 have definitely seen customers who are like that. 00:29:42.400 --> 00:29:57.838 Part of it is just I think that there is some deliberate efforts on the part of frontier model providers and the providers of like LLM tools to obfuscate what exactly is happening under the hood. 00:29:57.838 --> 00:30:08.558 And this can start to like make it intimidating, complicated, and feel a lot less trustworthy than it actually is. 00:30:08.558 --> 00:30:20.720 And like I think that that results in some people just not really knowing things like what is happening with my data as it is being passed around through this stuff. 00:30:20.720 --> 00:30:21.200 Right. 00:30:21.200 --> 00:30:48.720 Where in some cases, like, depending on which provider you're using, you may either genuinely not know, or it could be things like, oh, this what's happening is the data is being routed through AWS, it is going to hardware that is owned by Amazon and governed by them and adheres to their privacy policy, and it gets run through an open weight model that is hosted on their hardware and returned back to you. 00:30:49.278 --> 00:30:49.598 I see. 00:30:49.598 --> 00:30:52.960 So those kind of scenarios can turn off the customer. 00:30:52.960 --> 00:30:56.640 And and I don't think that's fair, especially when it comes to protecting their data. 00:30:57.278 --> 00:31:12.078 Yes. The other thing is just depending on like what software you're using, there are things like there are situations out there where the stakes for getting something wrong are very high, and AI makes it easier to get things wrong. 00:31:12.798 --> 00:31:16.240 Yeah. So I was going to flip the question that I had for you. 00:31:16.240 --> 00:31:28.000 Are there customers who are very gung-ho about using AI, but where you feel as a product manager that probably their adoption should be slower and you might want them to go slower? 00:31:29.118 --> 00:31:38.000 I'm going to say that has never been something that I have personally encountered, thank God, because some of the customers I work with are in the healthcare niche. 00:31:38.000 --> 00:31:53.200 Some of them work in government, some of

them work in finances, and there are a lot of places where the rollout of AI is very slow, very deliberate, very conservative, and they are making like every or taking every pain possible to get it right. 00:31:53.200 --> 00:32:01.358 And I'm so happy that that's the case because there are actually substantial downsides to them getting it wrong. 00:32:02.078 --> 00:32:07.519 I rarely end up thinking about these different industries, but I'm thinking about education where things definitely would be probably slower too. 00:32:08.318 --> 00:32:19.519 The thing that I think is a lot more common that I see is companies that are going headlong into AI, but don't really have a specific goal apart from use AI. 00:32:19.519 --> 00:32:28.240 And like I'm still not sure what to think about that because this is an expensive technology to roll out. 00:32:28.240 --> 00:32:34.480 I'm sure we've all seen some headlines about companies racking up multimillion dollar clawed bills. 00:32:34.480 --> 00:32:51.038 To some extent, it's just unless you know what you're getting from this or have a very good use case for it, it feels wasteful and disruptive to roll that out to a bunch of different teams and also then pay for what is effectively getting you nothing. 00:32:51.038 --> 00:33:06.558 At the same time, I've also seen some product managers make the argument that you actually kind of need to have this phase of just getting people to start using it so that they can discover the thing that it is valuable for in their workflow. 00:33:07.680 --> 00:33:14.000 Yeah. Probably I would say three years ago, when AI was new, generative AI was new, it made sense. 00:33:14.000 --> 00:33:23.519 But I think if companies are a little late in the game, they probably need to have some controls, some guardrails, and not just free for all. 00:33:23.519 --> 00:33:25.680 I think that train has left the station. 00:33:25.680 --> 00:33:28.960 But I would be concerned equally about that. 00:33:29.759 --> 00:33:36.240 Yeah. I kind of agree, but also, man, in a highly regulated industry, three years is no time at all. 00:33:36.880 --> 00:33:39.680 Right. In a highly regulated industry, yes. 00:33:39.680 --> 00:33:44.318 But in a non-regulated industry In that case, you're right. 00:33:44.318 --> 00:33:44.880 Yeah. 00:33:44.880 --> 00:33:48.000 Because in a regulated industry, anyways, things are going to go slower. 00:33:48.000 --> 00:33:58.640 And there are guardrails, but it's the non-regulated industry where I feel that education needs to happen, that training needs to happen, and guardrails absolutely need to exist. 00:33:58.640 --> 00:34:05.118 Because again, if you have customer data and if you're not considering that aspect, that would be a challenge, I think. 00:34:05.118 --> 00:34:08.000 We have covered good ground here. 00:34:08.000 --> 00:34:11.119 This is just a question that I'm curious about. 00:34:11.119 --> 00:34:13.199 And let me know what your take on that is. 00:34:13.199

--> 00:34:16.719 content practice or even product management? 00:34:16.719 --> 00:34:19.840 I mean, you've been in product management, you said about five years. 00:34:19.840 --> 00:34:24.960 What was a best practice five years ago that's no longer a best practice, in your opinion? 00:34:24.960 --> 00:34:26.239 As a product manager. 00:34:27.360 --> 00:34:30.159 Writing your own user stories for sure is one of them. 00:34:31.039 --> 00:34:31.518 Okay. Yeah. 00:34:31.518 --> 00:34:33.360 How do you feel about that? 00:34:33.920 --> 00:34:35.440 I'm so happy that that's gone. 00:34:35.440 --> 00:34:45.280 Like, that's one of those ones where it's like every once in a while I'll see a developer like complain about Jira on the internet and say something like only product managers like Jira. 00:34:45.280 --> 00:34:47.280 And I'm like, no, bro. 00:34:47.280 --> 00:34:52.559 Saying product managers like Jira is like saying that people with leprosy like leprosy. 00:34:52.559 --> 00:34:57.920 It's not something we use because we like, it's a disease, but I see. 00:34:58.239 --> 00:35:01.518 So you'll feel like liberated that Yes. 00:35:02.400 --> 00:35:26.480 Having something that can sit between me and that system and offer a layer of abstraction that helps like remove this mechanical task and can just take the higher level goal and enumerate out all it into all of these different chunks that mostly have like boilerplate between like what is in the initial requirement document, the initial spec document, and what needs to go into stories. 00:35:26.480 --> 00:35:28.320 That's been a dream come true. 00:35:28.320 --> 00:35:37.840 And it is a place where like a couple of hours of my day on average have been freed up, and I've been able to put it right back into stuff that feels better to do because it's more impactful. 00:35:38.639 --> 00:35:39.760 Yeah, that's a great example. 00:35:41.119 --> 00:35:55.679 The technical writer version of this that we've started to play around with is using Claude Code as a way to deal with just the nitty-gritty of the Git-based workflow for docs as code. 00:35:55.679 --> 00:36:01.199 And that's an interesting spot because like in technical writing, there's a number of technical writers out there. 00:36:01.199 --> 00:36:25.440 I was one of them at the time, or when I was a technical writer, who love docs as code, who love the fact that you get to get down into like the details, go into the command line, learn Git, play around with different markup languages, define your own components, run the static site generator locally, make a GitHub action for it, and like do all of this really technical stuff. 00:36:25.440 --> 00:36:30.559 And then there's a lot of technical writers where that's just like the miserable part of the job. 00:36:30.559 --> 00:36:40.079 It's something they're not interested in that stresses them out, that they view as like, uh, okay, if I mess this up, I'm gonna have to go contact like a developer Googling

a solution for 30 minutes. 00:36:40.079 --> 00:36:46.000 And for those people, using Claude Code and just saying, hey, I got these updates that I need to make, let's do it. 00:36:46.400 --> 00:36:46.960 Oh, yeah, definitely. 00:36:46.960 --> 00:36:57.360 Initially I was a little hesitant to use that, but I'm one of those who likes to be doing docs as code and be in the developer workflow. 00:36:57.360 --> 00:36:58.960 So I feel very empowered. 00:36:58.960 --> 00:37:03.760 But yeah, you're right, that it it can empower somebody who doesn't want to do this as well. 00:37:03.760 --> 00:37:05.679 It kind of deburdens them. 00:37:06.239 --> 00:37:10.000 I think that this is one of the parrot, like I don't know if it's actually a paradox. 00:37:10.000 --> 00:37:11.199 I'm gonna call it a paradox. 00:37:11.199 --> 00:37:38.559 One of the conundrums of AI right now is that for people who are very, very advanced and established and good at a trade in like tech, the benefits that AI provides can be like a lot lesser because search, looking things up, well, if you've already got your programming language and the systems you're working in and your code base committed to memory because you've been doing it for years, it starts to be less useful. 00:37:38.559 --> 00:37:52.880 Then there are people who don't have any of those things who can benefit a lot from having like a little machine assistant who's able to spin up like a prototype or handle the Git workflow for doing docs as code stuff. 00:37:52.880 --> 00:38:02.800 And it frees them from having to commit a lot of time to learn these skills that are kind of ancillary to the thing that they actually do for their job. 00:38:02.800 --> 00:38:09.679 But also because they don't learn those skills, they're they have less in place to defend them when things go wrong. 00:38:09.679 --> 00:38:12.239 I'm still figuring out what to do with all that. 00:38:13.039 --> 00:38:15.119 has been, for me, Niko, a good conversation. 00:38:15.119 --> 00:38:17.920 You've really picked up on a lot of good stuff. 00:38:17.920 --> 00:38:21.840 And I came scripted, but I think we went off the script entirely. 00:38:22.960 --> 00:38:25.679 This happens when I start talking about AI with people. 00:38:25.920 --> 00:38:26.800 It's which is great. 00:38:26.800 --> 00:38:27.760 That's my goal. 00:38:27.760 --> 00:38:33.679 The script was for me, not really for you, because I knew this would end up being a great conversation. 00:38:33.679 --> 00:38:37.039 I just had to prepare myself to be able to understand. 00:38:37.039 --> 00:38:45.440 And thankfully, as you were giving these examples, since I've been in these tools, I was hopefully able to ask some intelligent questions of you. 00:38:46.480 --> 00:38:46.960 Yes, definitely. 00:38:46.960 --> 00:38:50.000 And hopefully I gave intelligent answers. 00:38:50.800 --> 00:38:53.920 Oh, that was—I mean, that was not even up for question. 00:38:53.920 --> 00:38:55.679 I that was anticipated. 00:38:55.679 --> 00:39:06.320 But thank you, Niko, for being on the show and sharing your experience and your insights with us because it's definitely helped me. 00:39:06.320 --> 00:39:11.039 And you've challenged me to think about AI differently as well on a personal level. 00:39:11.039 --> 00:39:12.559 So thank you again. 00:39:12.559 --> 00:39:16.639 Have a lovely evening, and uh hope to run into you at another conference. 00:39:16.639 --> 00:39:21.199 Absolutely. Thank you for listening to Inside Tech Comm. 00:39:21.199 --> 00:39:25.760 If there's a guest you would love to hear on the show, please let me know. 00:39:25.760 --> 00:39:31.039 And don't forget to follow and catch every episode on your favorite podcast app. 00:39:31.039 --> 00:39:32.320 See you next time.

I-edit ang transcript