How to Utilize AI in DevRel  ## Metadata - Author: [[The State of Developer Education]] - Full Title: How to Utilize AI in DevRel - Category: #podcasts - URL: https://share.snipd.com/episode/2f443f8e-7c7a-4a8b-ab55-6a3a29501b3a ## Document Notes ## Highlights - **AI Enables Creative Coding for Kids** - Greg Baugues shared how using AI helped his daughter creatively write and edit a 168-page story. - They built a personalized text adventure that kept her engaged and made coding fun from an early age. Transcript: Speaker 1 Decade or more cultivating a specific voice to reach a specific audience and you structure content in a specific way. And it's a hone skill. It's interesting you talk about author using it and giving you the wrong answer. I think it's called Cunningham's Law, which is the best way to get the right answer on the internet is to give the wrong one. And chat GPT is really good at giving you the wrong answer. And I don't mean like factually, although it hallucinates you, but in terms of writing, it's just really good at giving you something where I'm like, oh no, actually don't like how that Is said, but I think I can change that into how it's said. Also on the writing side, you know, speaking outside of the Debrel, I have a 90 year old daughter and she wants to be an author when she grows up and we were at the car dealership for like Three hours a few months ago. She's like, hey, can I borrow your phone? And she loves talking to chat to you because can I ask you to write the first chapter of a story? I'll make sure and she like giggles when it comes back and she gave me the character and some ideas. And then she just sat there for three hours doing this. And when by the time we got out, we went home and we copied and pasted the chat history to a Google Doc. We had 24 pages, 13 chapters of a story. Over the last four months, she has edited that. I mean, she calls it editing. She doesn't say writing. She's like, can I edit? She's edited that into a 168 page document with 20,000 words. And again, here, you know, now she's at a point where, yeah, probably like 5% of the words in there were originally authored by chat GPT. But to your point about structuring a story, she's tried writing stories before. And I found that she was always really good at writing chapters, but we had a hard time ([Time 0:08:59](https://share.snipd.com/snip/c694d39a-565d-4029-a040-794c2a02727c)) - **AI Meets Devs Where They Code** - Developers will increasingly get AI help directly in their IDEs, reducing distractions from web searches. - DevRel must adapt as AI meets developers closer to where they work, not on company real estate. Transcript: Speaker 2 Like all these different surfaces that DevRel often owns and maintains like forms. Like if I get to the end of a docs page and there happen to be a few forums, threads that are relevant to this, I might want to see that or tech blog posts for YouTube videos that your developer Advocates have done. I'd love to see that. And then at the same time, the other side of that coin is at what point do you allow the user to eject from the bubble that you have put them in with the AI? And so like, that's an open ended question for me today. Speaker 1 I think an interesting thing about change in coding over the last year for me at least is I've tried to adopt more of, I'd say, coding with AI is, you know, the old paradigm was you hit an Error message, you copy, paste into Google and then scroll and then find the stack overflow, right, or whatever it is. I spend so much more time in the IDE now. And if you think about it, like the entire internet's constructed to steal your attention, sort of like the economy. And so I used to get really frustrated because I would get distracted when coding and I had a friend point out, I was like, you're acting like that's your fault. Like in order to find answers to your coding problems, you're going out to the web, which is this whole environment constructed to steal your attention off in all these different directions. And the really interesting bit about using cursor over the last six months or so is that I have my GPT 4 chat right there in my IDE. It ([Time 0:21:56](https://share.snipd.com/snip/accca439-67cc-4e46-867b-91801d5afb56)) - **Text Fatigue Spurs Video Growth** - The barrier to producing text content is lower with AI, but content fatigue is rising among developers. - Video content and human connection will grow in importance in DevRel despite AI advances. Transcript: Speaker 3 In 2030 look like as a result of generative AI? I'm curious to hear each of your thoughts if you have them. Speaker 1 I'll share one as kind of back on the content thing. I think generally, like I do think the barrier to entry to producing text-based content is lower. Although I think we all agree that there's still a high-quality bar. And I think that people are going to become more and more fatigued with text-based content because there's just going to be so much more of it. And I think that folks are going to be looking for personalized content via chat as opposed to wanting to read blogs. That said, I also think that humans just have a strong desire for human connection. And so I personally have spent the last year up leveling my video game because I think on YouTube content, there is a higher barrier to entry. And also, I think people want to make a connection with another human being. And I don't think that's going to go away no matter how good the tools get. I think people are going to look for trusted authorities or what it see faces. And I also just think that post-COVID, there's less of an appetite for travel. When I was at early days at Twilio, I would fly to Minneapolis to speak to a PHP meetup of 30 people. And that would consume two days. Now, why not just sit at home, make a YouTube video on the same topic that could be watched by thousands or tens of thousands of people over the coming months as opposed to a single event? And so personally, I think one just small tactical shift is moving away from IRL events, moving away from text-based content. And ([Time 0:23:58](https://share.snipd.com/snip/8eda45fb-eade-4895-996f-d0b233fc3034)) - **AI as a Patient Thought Partner** - Use AI as a thought partner to brainstorm and explore ideas without fear of judgment. - It offers infinite patience and non-judgmental feedback, useful in early problem-solving stages. Transcript: Speaker 2 I sometimes call it thought partner. Like the phrase I use a lot about this kind of thing is like someone that I can toss the ball back and forth with. And that's a really hard thing to find sometimes in your day-to-day job, right? People are really busy. People want a solution really quickly. I'm a big believer in sitting in problem spaces and putting space between like the initial stimulus and the reaction. I think we get better outcomes when we do that. And I still want that from humans and prize that. But I think the bot really helps me in my initial sort ([Time 0:29:12](https://share.snipd.com/snip/3c8b345d-2a08-41fa-95c2-e4ee69ab5fb0)) - **Fun AI Projects Attract Developers** - Fun and creative AI projects attract developer interest by entertaining alongside educating. - Human stories and engaging content help build brand affinity and draw developers in. Transcript: Speaker 1 Dana White had this line in there, he said, my competition is anything else you might be doing on a Saturday night. Right. And so he's not thinking like my competition's not the other fights, whatever, it's like, how are you going to spend your time? DevRel encompasses a lot. So some DevRel is docs, right? And so if I'm on your docs, I'm probably already kind of committed to your thing. And maybe I don't need cute stories and whatnot. But a lot of my time at Twilio was spent on the evangelism team where our metrics are around signups and awareness, brand awareness and brand affinity. And so if you're producing that type of content, your competition is anything else that a developer might be consuming on the internet. And so your competition is TikTok, your competition is woodworking videos on YouTube, right? And so they're like human beings like stories and human beings like being entertained. And you do want to educate, but you often need to entertain first to grab the attention. And so in that case, having a fun aspect to your projects is a really makes for better stories and makes for better to draw folks in to get them to have the first time user experience to get Them to sign up and then see if the thing works. And then you can push them off to the docs that are we'll just give them the answers they're looking for. But I do things like just telling the fun stories. Also, it's just more fun to build. Like it's just the best rate alert, right? Like, you know, there's enough boring stuff at work. And I think so many of us got into development because it was fun and because of this instagramification, because we like building things. And ([Time 0:33:10](https://share.snipd.com/snip/574b0be7-a64e-46a6-9342-b0684eb45fca)) - **Strategic Content Mix in DevRel** - Tailor DevRel content to company lifecycle phase: SEO-driven posts for organic growth, and social-optimized stories for awareness. - Mix practical tutorials with entertaining stories to capture varied developer attention. Transcript: Speaker 1 Think they answered your question about it depends a lot on the life cycle of the company and also the numbers you're trying to move at that time and the time frame you're trying to get Your ROI. Broadly speaking on, let's just say the Twilio blog, we'd have two different types of content that we'd have the posts that we write for the purpose of getting organic traffic in the Long run. So it's like environment variables in Ruby, right? And we'd know that there wasn't a really great post out there. There's a whole in the content ecosystem for this. You could spend a lot of time writing a really good solution to that. And the day you press publish, you'd get zero visits to it. But over time, the chart would go like this, right? And now drive awareness. And then it would get the cookie and then we could do retargeting and whatnot. We'd also write a lot of posts that were like, how to send a text message in Python, how to send a text message Ruby, how to make a phone call. So those are going to get less traffic, but they're going to convert more. Then you had posts that you want to have blow up on social. So like, when I wrote this, the best story here was like how I taught my dog text me selfies, right? So it's like the dog pressed a button on the Arduino, it took a picture with the webcam, they used Twilio MMS to send a text, right? So that hit Hacker News, the traffic for that was like boom, 50,000 visits, and then nothing, right? Nobody ever Googles how to teach my dog since selfies like Twilio made $0 in revenue off of that use case, right? But I tend to think of it as there's content that developer consumes when they have an IDE open and they're trying to solve a problem. And then there's content that they consume on the toilet. And the dog selfie thing is 100%, like that's on the toilet. But hopefully it plans to see in the back of their head that if they need to send text messages in the future, they're going to do that. And I think at different phases in a company's lifecycle, there's need for awareness, versus need for signups, versus need for organic traffic, because the SEO juice rises all the Posts on the blog. ([Time 0:37:20](https://share.snipd.com/snip/ca052620-5092-4090-9ebe-f00b6fcf84f9)) - **AI Hype to Human Augmentation** - We are in the hype and pain phase of AI adoption but real utility is emerging. - The best outcomes come from augmenting humans with AI, not full automation. Transcript: Speaker 3 With? And you don't have to speak on behalf of the businesses, but existentially, how do you think about that? Speaker 4 It feels like we are at the scary stage where a crapped on a money is being put in and there's maybe a crapped on of losses. But I kind of want to get that over with it so we can get past it. I worry that we're kind of there, but I absolutely see value in this. So I think it's kind of like a necessary pain. We'll have to go through to kind of get to the, this is less a new superhero. And oh, this is a tool I use every single day and it's just normal now. Speaker 1 I sort of feel like the folks who are saying, hey, this like the next phase of crypto, like, haven't actually used the tools. And then the folks who are saying this is going to eliminate all of the jobs also haven't actually used the tool. Like there is absolutely utility here. And also it's not that great. And I think so much of the game right now and it was so exciting as it's all greenfield is we're all trying to figure out, like for what problems is actually the right tool for the job. And we're all sort of collectively discovering that right now. And the place where I look for a lot of inspiration is the world chess, a Casparov in 97 loses to deep blue. And then AI really takes over there and he makes the point he's got this great TED talk and then he's got his book, Deep Thinking that he published in 2017. And he says, look like the human chess players today are the best human chess players who ever lived because they've been training alongside super powered AI is for their entire lives. And so I think a lot of the really great use cases are in the like augmentation, like working alongside AI, not in the automation. And generally I found and we were talking about this with customer support, like generally, the places where you're trying to automate away the work are underperforming. And generally the places where you have a very skilled person or even not a skill, even a nine year old, right? Like who's trying to do something alongside in AI are finding that they're able to do that thing better, faster and cheaper. And those ([Time 0:39:44](https://share.snipd.com/snip/ab67d31b-3b6e-40bb-a342-e302bb218d85)) - **Master AI Prompting and Logic** - Developers must learn to prompt AI models effectively and handle outputs for logic-based decisions. - Choosing the right AI model and deployment method is critical for integrating AI into applications. Transcript: Speaker 2 Decisions on? I think there's been tooling out there for that. But I think there's still a lot more to do. Also, how do I pick the right model in the first place? You can go out to Hugging Face Today and there's like, Jillian models, by the way, Olama, one of my favorites, because you can run it locally. The demo I did on stage at the API Summit in Austin the other day, I was running Olama 2 on an M2 MacBook Air. If you ever want to do a tension-filled tech demo, do that. We have to wait for a little bit, but it works. I think one of the questions a lot of developers and companies will have to ask themselves is like, if I'm putting AI into my product, who's AI? It's less about, am I going straight to open AI or whatever? Am I doing this via AWS or Azure? Or is it worth it to try to make it local to my application? How much is that going to cost? How do I figure out how to think scales? I think a lot of those questions are less about how good is the model? A lot more about, as a mere model app developer, how am I taking advantage of this? I think we'll see a lot more thought in that space. I'd love to see some things ([Time 0:42:43](https://share.snipd.com/snip/9922205e-2600-407d-9923-f93fa35dfffd)) - **Manage Prompts Like Code** - Use tools like PromptLayer to manage, version control, and optimize AI prompts like software code. - Treat prompt engineering as a new profession to ensure cost-effective and quality AI output. Transcript: Speaker 3 Most excited about right now that you think people may not have heard of yet? What's your sort of up-and-coming exciting AI tool? Speaker 1 I'll say Prom Flayer, great New York startup. Actually, John, they probably exist because of you and your work from the hackathon. They went to, it was the BCA, the world's largest high school hackathon. The easiest way to put it is a CMS for your prompts. You'd think about, you probably start off by hard coding your prompts in your source code, and their thing is, hey, much in the same way that you no longer hard code the pros in your website. Let's create a CMS where we can extract the prompts, and then we can have version control, and then we can also have e-vowels so we can test and score the responses to see how our prompts Performing against, let's just say unit tests. Also, if we extract them out of the source code, now you don't need to know how to code to edit and iterate on the prompts. They actually, I think I have here this hack that they gave me. This is prompt engineer audit. These guys are making this bet that there will be a world where there exists prompt engineers as a legitimate job, and you are going to be able to do that much in the way that there are Salesforce Admins who don't know how to code but are highly technical and know the ins and outs, and not just also like the SEO where they not just can write better prompts, but they know how to build The system to say that this prompt is objectively better and cheaper and whatnot, and they're building the tooling for that. I think it's really, really needed, and it's been a really, really fun play. With that, it's an interesting ([Time 0:44:20](https://share.snipd.com/snip/ddb87423-60cf-419a-978e-0a4aba8d3877))