How the best writers use AI

00:00:00,600 --> 00:00:33,740 [Eric] [instrumental music] Welcome back to the Token Intelligence Show. AI is changing the way we work, so we are going to help you understand the state of the art, cut through all the hype, of which there is a lot, and apply wisdom to become a great leader or worker in this new era that we live in. And today, John is going to ask me a bunch of questions about writing- 00:00:33,740 --> 00:00:33,750 [John] Yes 00:00:33,750 --> 00:00:40,580 [Eric] ... and AI. And the context here is that I am technically a professional writer. 00:00:40,580 --> 00:00:41,200 [John] Yes. 00:00:41,200 --> 00:00:44,710 [Eric] Uh, I'm technically a professional writer, so- 00:00:44,710 --> 00:00:44,710 [John] [laughs] 00:00:44,710 --> 00:00:46,810 [Eric] ... which is, which is cool. [laughs] 00:00:46,810 --> 00:00:53,400 [John] Yeah. I'm so excited about this topic because, as you know, I pretty much exclusively use AI to code. 00:00:53,400 --> 00:00:53,440 [Eric] Yep. 00:00:53,440 --> 00:00:56,760 [John] Occasionally, like maybe we're doing a website update, I would use- 00:00:56,760 --> 00:00:56,810 [Eric] Mm-hmm 00:00:56,810 --> 00:01:05,980 [John] ... AI for some copy, but I'm also not... don't feel very good at that use of AI, and I'm also often frustrated and just, you know, end up writing. 00:01:05,980 --> 00:01:06,280 [Eric] Mm-hmm. 00:01:06,280 --> 00:01:11,480 [John] With, with some exceptions recently. Um, but I wanna frame this in 2 ways. 00:01:12,560 --> 00:01:22,610 [John] One, one of the interesting themes that I've seen going around recently is on applying old rules for writing, like that you would've learned in high school- 00:01:22,610 --> 00:01:22,630 [Eric] Mm-hmm 00:01:22,630 --> 00:01:24,440 [John] ... or a college English class. 00:01:24,440 --> 00:01:25,060 [Eric] Yep. 00:01:25,060 --> 00:01:28,980 [John] And then having these things called AI skills, which we've talked about on the show before. 00:01:28,980 --> 00:01:29,640 [Eric] Mm-hmm. 00:01:29,640 --> 00:01:35,300 [John] Um, that's just a concise, um, way of, of saying s- you know, something you want the model to do, and you can- 00:01:35,300 --> 00:01:35,310 [Eric] Mm-hmm 00:01:35,310 --> 00:01:41,460 [John] ... refer to it by name. Um, there's been one... I wanna... I wanna kind of quiz you on 2 of 'em. 00:01:41,460 --> 00:01:41,620 [Eric] Yep. 00:01:41,620 --> 00:01:44,760 [John] There's been one going around, um, X, like this last week- 00:01:44,760 --> 00:01:44,770 [Eric] Mm-hmm 00:01:44,770 --> 00:02:00,050 [John] ... and then another one that I've, um, run across before. But before we dive in, I'd love to just hear your general take on, on AI and writing. Like, what is your current experience with, with, like, how you're interacting with AI and writing- 00:02:00,050 --> 00:02:00,300 [Eric] Mm-hmm 00:02:00,300 --> 00:02:04,200 [John] ... like, right now, on a, on a everyday professional type level? 00:02:04,200 --> 00:02:04,400 [Eric] Yeah. 00:02:06,360 --> 00:02:25,930 [Eric] We, we use AI... So context, I lead what's called the content engineering team at Vercel. And it's called content engineering because we are a technical product built for people doing technical things, so building with AI, building applications, building websites, deploying those to the public internet. Um, 00:02:27,540 --> 00:02:42,840 [Eric] so our target audience is developers, and increasingly developers using, you know, AI coding agents and, and building AI applications and AI agents. And so everyone on our team is deeply technical. Everyone on our team, we're a team of writers, so we deal- 00:02:42,840 --> 00:02:42,989 [John] Mm-hmm 00:02:42,989 --> 00:02:51,850 [Eric] ... on words every day. Everyone on our team, before joining Vercel, had built some sort of agent related to content, which- 00:02:51,850 --> 00:02:51,920 [John] Huh 00:02:51,920 --> 00:02:52,670 [Eric] ... is really interesting. 00:02:52,670 --> 00:02:53,470 [John] Very cool. Yeah. 00:02:53,470 --> 00:03:10,060 [Eric] Yeah. So, uh, one of the... just a couple examples of the types of things that people have done. So, um, we have, uh, one guy on the team who came from GitHub, and he built a content agent inside of GitHub- 00:03:10,120 --> 00:03:10,380 [John] Nice 00:03:10,380 --> 00:03:16,880 [Eric] ... and did a, a bunch of interesting thinking and work around how to help people, um, 00:03:18,080 --> 00:03:20,600 [Eric] develop, uh, clear arguments that would be- 00:03:20,600 --> 00:03:20,630 [John] Okay 00:03:20,630 --> 00:03:22,320 [Eric] ... an input for a content agent- 00:03:22,320 --> 00:03:22,620 [John] Yeah 00:03:22,620 --> 00:03:23,530 [Eric] ... um, to generate something, 00:03:24,560 --> 00:03:40,460 [Eric] which is probably something that will come up as our c- conversation progresses. Another, um... So one woman on the team did a lot of work around developing content for, uh, discovery. So if you think about SEO- 00:03:40,460 --> 00:03:40,690 [John] Right 00:03:40,690 --> 00:03:51,769 [Eric] ... um, optimizing for keywords and that whole realm, building a content agent that sort of generate, you know, generated that content at an, at an extremely high scale, you know? 00:03:51,769 --> 00:03:51,780 [John] Right. 00:03:51,780 --> 00:03:56,260 [Eric] So we're talking about hundreds or thousands of pages per month, you know, being generated- 00:03:56,260 --> 00:03:56,520 [John] Right 00:03:56,520 --> 00:03:57,340 [Eric] ... of content. So 00:03:58,480 --> 00:04:05,160 [Eric] that's why we say content engineering. It's maybe one of the hardest roles I've ever had to even think about- 00:04:05,160 --> 00:04:05,250 [John] Right 00:04:05,250 --> 00:04:06,660 [Eric] ... hiring for because 00:04:08,000 --> 00:04:11,740 [Eric] the technical acumen is mandatory- 00:04:11,740 --> 00:04:12,000 [John] Right 00:04:12,000 --> 00:04:21,380 [Eric] ... but the standard for the content that we produce at Vercel is extremely high, higher than any other company I've ever seen- 00:04:21,380 --> 00:04:21,579 [John] Right 00:04:21,579 --> 00:04:22,560 [Eric] ... or heard about. 00:04:22,560 --> 00:04:22,680 [John] Right. 00:04:22,680 --> 00:04:28,600 [Eric] And I mean, our CEO is an extremely articulate communicator- 00:04:28,600 --> 00:04:29,180 [John] Right 00:04:29,180 --> 00:04:40,220 [Eric] ... and is... just has an, an amazing standard for the quality. So we, we can only hire people who are absolutely excellent writers- 00:04:40,220 --> 00:04:40,469 [John] Mm-hmm 00:04:40,469 --> 00:04:41,070 [Eric] ... as a core- 00:04:41,070 --> 00:04:41,180 [John] Right 00:04:41,180 --> 00:04:42,980 [Eric] ... as a core skill. Um, 00:04:44,000 --> 00:04:47,160 [Eric] so that's sort of the context of, like, what we do and who's on the team. 00:04:48,240 --> 00:04:57,300 [Eric] And we use AI for all sorts of different things. So I'll just give you a couple of examples- 00:04:57,300 --> 00:04:57,310 [John] Yeah 00:04:57,310 --> 00:04:58,260 [Eric] ... that I think will be helpful. 00:04:59,500 --> 00:04:59,680 [Eric] So 00:05:01,660 --> 00:05:04,360 [Eric] we use AI very heavily for review. 00:05:04,360 --> 00:05:04,860 [John] Okay. 00:05:04,860 --> 00:05:15,800 [Eric] So we have a content agent that has a repository of really fundamental information about, 00:05:16,840 --> 00:05:20,400 [Eric] uh, about core things at Vercel. 00:05:20,400 --> 00:05:20,880 [John] Okay. 00:05:20,880 --> 00:05:25,440 [Eric] So you can think about things like our voice and tone. 00:05:25,440 --> 00:05:26,280 [John] Sure. 00:05:26,280 --> 00:05:41,190 [Eric] You can think about, um, you know, sort of, uh, conventions with syntax, so as detailed as do you use a period, uh, in a bulleted or numbered list, right? 00:05:41,190 --> 00:05:41,220 [John] Sure. 00:05:41,220 --> 00:05:41,740 [Eric] Or what are the ru- 00:05:41,740 --> 00:05:42,160 [John] Mm-hmm 00:05:42,160 --> 00:05:42,720 [Eric] ... the rules or- 00:05:42,720 --> 00:05:42,900 [John] Right 00:05:42,900 --> 00:05:45,830 [Eric] ... the conventions that you use, right? When do you capitalize things? 00:05:46,900 --> 00:05:52,820 [Eric] So a lot of those types of conventions are encoded. A lot of core product knowledge is encoded- 00:05:52,820 --> 00:05:52,830 [John] Mm-hmm 00:05:52,830 --> 00:05:58,760 [Eric] ... or sort of, like, values or the way that we explain this. So you could think about that as product marketing input to- 00:05:58,760 --> 00:05:59,100 [John] Right 00:05:59,100 --> 00:06:01,720 [Eric] ... what we kind of call the core brain- 00:06:01,720 --> 00:06:01,900 [John] Right 00:06:01,900 --> 00:06:04,412 [Eric] ... um, of information. 00:06:04,412 --> 00:06:11,462 [Eric] And then there are sort of skills built on top of that. So one of those would be a content review. So that would be- 00:06:11,462 --> 00:06:11,462 [John] Right 00:06:11,462 --> 00:06:18,682 [Eric] ... we get, uh, the, the content engineering team gets up, uh, some content from another team. 00:06:18,682 --> 00:06:18,712 [John] Mm-hmm. 00:06:18,712 --> 00:06:36,941 [Eric] And this could be... It could be nearly anything. We review, we review... I'm not gonna say everything, because I'm sure that there are a lot, th- there are some things that are published that don't go through our full review process. But, but a vast majority of everything actually- 00:06:36,941 --> 00:06:36,941 [John] Right 00:06:36,941 --> 00:06:39,142 [Eric] ... goes through our team to review. Um, 00:06:40,332 --> 00:06:42,652 [Eric] and so it's an unreal volume. I mean, it- 00:06:42,652 --> 00:06:42,682 [John] Yeah 00:06:42,682 --> 00:06:47,152 [Eric] ... it really is an unreal volume, uh, which is part of why we use AI, because it- 00:06:47,152 --> 00:06:47,392 [John] Sure 00:06:47,392 --> 00:06:48,312 [Eric] ... it helps us accelerate- 00:06:48,312 --> 00:06:48,322 [John] Sure 00:06:48,322 --> 00:06:49,852 [Eric] ... that process. Um, 00:06:51,612 --> 00:06:58,732 [Eric] so we run a lot of content through a review agent. So there may be an email campaign that the growth team, you know, wants to send. So it's a series- 00:06:58,732 --> 00:06:58,742 [John] Mm-hmm 00:06:58,742 --> 00:07:34,762 [Eric] ... of emails. And the... This is a really great use for AI and for the way that we built it. So the lifecycle marketing lead does an incredible job of preparing documents that have a lot of detailed information about the segment, the performance, the conversion rate that we wanna move, how much we believe that we can move it, um, product usage data, et cetera. And then they will actually... I, you know, they use AI. We don't, we don't actually know. I mean, they, they probably just use AI to generate an email series. [clears throat] Right? 00:07:34,762 --> 00:07:34,762 [John] Right. 00:07:34,762 --> 00:07:36,672 [Eric] And they're not a, they're not a writer. And so- 00:07:36,672 --> 00:07:36,792 [John] Sure 00:07:36,792 --> 00:07:37,932 [Eric] ... they just sort of make a pass- 00:07:37,932 --> 00:07:38,012 [John] Mm-hmm 00:07:38,012 --> 00:07:45,772 [Eric] ... at it, right? And so we get this content, and it's an incredible, uh, it's... Each document has incredible context, 00:07:46,972 --> 00:07:53,432 [Eric] uh, for our content agent to understand all of the details- 00:07:53,432 --> 00:07:53,572 [John] Yeah 00:07:53,572 --> 00:08:00,852 [Eric] ... on top of the core brain knowledge that we have that are Vercel fundamentals. And then we have a skill that will review the content. 00:08:00,852 --> 00:08:01,152 [John] Right. 00:08:01,152 --> 00:08:08,292 [Eric] Right? And say, "Okay, this is way too long," or, um, you know, "The voice and tone is off." 00:08:08,292 --> 00:08:08,872 [John] Right. 00:08:08,872 --> 00:08:15,112 [Eric] Or, "It's way too informal," you know, or, "We don't refer to ourselves in the 1st person-" 00:08:15,112 --> 00:08:15,211 [John] Yeah 00:08:15,211 --> 00:08:16,452 [Eric] ... "you know, in this context." 00:08:16,452 --> 00:08:16,612 [John] Sure. 00:08:16,612 --> 00:08:16,992 [Eric] Right? 00:08:16,992 --> 00:08:17,052 [John] Yeah. 00:08:17,052 --> 00:08:21,852 [Eric] Um, "We refer to ourselves in the 3rd person in this context," et cetera. Right? And so 00:08:22,992 --> 00:08:28,822 [Eric] it will make a pass at, "Here are all of the issues with, you know, this particular- 00:08:28,822 --> 00:08:28,832 [John] Right 00:08:28,832 --> 00:08:29,392 [Eric] ... piece," or whatever. 00:08:29,392 --> 00:08:30,072 [John] Right. 00:08:30,072 --> 00:08:38,721 [Eric] And then from there, depending on what it is, we can... The content agent also has the ability to rewrite or, you know, to, to- 00:08:38,721 --> 00:08:38,721 [John] Right 00:08:38,721 --> 00:08:39,512 [Eric] ... do a rewrite. 00:08:40,652 --> 00:08:55,232 [Eric] Um, you know, or if you have... if it's made a review pass and you sort of see all of the issues, I'll, it can actually be way faster for a human just to look at that. And then, you know, if it's a 50-word, you know, email, you just- 00:08:55,232 --> 00:08:55,322 [John] Right 00:08:55,322 --> 00:08:56,452 [Eric] ... can rewrite it really quickly. 00:08:56,452 --> 00:08:56,592 [John] Right. 00:08:56,592 --> 00:08:59,422 [Eric] It just, it sort of, like, points out all of the issues, right? 00:09:00,572 --> 00:09:02,772 [Eric] You know, or you'll just have it do a rewrite, and you edit the rewrite. 00:09:02,772 --> 00:09:03,872 [John] Right. 00:09:03,872 --> 00:09:22,752 [Eric] Um, so it speeds that process up a lot. The, a- and, you know, we have landing pages that come through. We have, um, you know, assets for our demand generation team, you know, webinar descriptions, you know, sign-up forms, all sorts of stuff, um, 00:09:23,772 --> 00:09:25,572 [Eric] that comes through. So we use AI very heavily for that. 00:09:26,902 --> 00:09:32,352 [Eric] The experience on a, like for the individuals on my team, including- 00:09:32,352 --> 00:09:32,362 [John] Yeah 00:09:32,362 --> 00:09:38,262 [Eric] ... me, who use AI to actually do drafting and writing is where [laughs] things get very interesting. 00:09:38,262 --> 00:09:39,232 [John] Yeah. 00:09:39,232 --> 00:09:39,632 [Eric] Um, 00:09:40,872 --> 00:09:46,192 [Eric] because people, people have really different approaches to that, like very different styles. 00:09:46,192 --> 00:09:47,242 [John] Yeah. 00:09:47,242 --> 00:09:56,852 [Eric] Which, which when you step back, it makes a lot of sense. On the surface, you'd say, "Okay, well, we have a content agent," and so you get a blog post project, and you just sort of, you use the content agent, and it's- 00:09:56,852 --> 00:09:57,052 [John] Right 00:09:57,052 --> 00:10:08,732 [Eric] ... you know, um, and you follow the process, right? But the process is very different for each individual on the team, which is actually very reflective of a normal creative process. 00:10:08,732 --> 00:10:08,952 [John] Yeah. 00:10:08,952 --> 00:10:09,952 [Eric] Like differences in creative process. 00:10:09,952 --> 00:10:18,192 [John] And that's dictated not by their jobs, but how the, but by how they work, right? 'Cause they're working on similar work, but they're approaching it different ways. Is that what you're saying? 00:10:18,192 --> 00:10:42,492 [Eric] Exactly. And I'll give you a really specific example. Um, I'll give you a very specific, a specific example of that. It, it's also worth noting that the, the team, sort of the, the f- there are 2 technical writers on the team who focus on a very specific type of content, which would be, like, technical guides and, and other things like that, which we produce a very high volume of. And that's their area of expertise. They're- 00:10:42,492 --> 00:10:42,662 [John] Okay 00:10:42,662 --> 00:10:43,032 [Eric] ... involved in the process . 00:10:43,032 --> 00:10:44,912 [John] So there is some, like, specialization. 00:10:44,912 --> 00:10:53,832 [Eric] There's some specialization. So [smacks lips] there are 2 people who have that role. The other 4 people on the team, which that includes me, are generalists. And so- 00:10:53,832 --> 00:10:53,922 [John] Okay 00:10:53,922 --> 00:10:56,972 [Eric] ... we work on anything, including some of that type of work. 00:10:56,972 --> 00:10:58,162 [John] Right. 00:10:58,162 --> 00:11:08,392 [Eric] Um, but we will... A- and we intentionally, we intentionally hire for that profile, right? Highly technical, highly capable, excellent writer- 00:11:08,392 --> 00:11:08,692 [John] Right 00:11:08,692 --> 00:11:15,862 [Eric] ... experience across a wide variety, you know, a decade of n- you know, whatever experience across- 00:11:15,862 --> 00:11:15,862 [John] Yeah 00:11:15,862 --> 00:11:17,852 [Eric] ... a bunch of different functions, right? And so- 00:11:17,852 --> 00:11:17,862 [John] Right 00:11:17,862 --> 00:11:23,592 [Eric] ... you can sort of jump into any, any part of the business and provide value is really the target that we're hiring for. 00:11:23,592 --> 00:11:24,052 [John] Sure. 00:11:24,052 --> 00:11:24,352 [Eric] Um, 00:11:25,452 --> 00:11:41,312 [Eric] [smacks lips] and the... So let's take an example of a blog post, right? The standard for publishing content on the Vercel blog is extremely high. It goes through multiple rounds of review, even at a very high level in the company. Um, and 00:11:42,372 --> 00:11:42,692 [Eric] so 00:11:44,652 --> 00:11:51,862 [Eric] writing for the Vercel blog is a very intensive process. Um, and 00:11:53,052 --> 00:12:01,092 [Eric] the way that people approach that, like I said, can be very different. So I'll give you a, I'll give you an example of 2 workflows. So 00:12:02,592 --> 00:12:07,000 [Eric] there... So my workflow tends to be... 00:12:07,000 --> 00:12:14,700 [Eric] ... doing, using our content agent to do a significant amount of context gathering. 00:12:14,700 --> 00:12:15,720 [John] Yeah. 00:12:15,720 --> 00:12:16,739 [Eric] And so- 00:12:16,740 --> 00:12:18,059 [John] So what do you, what do you mean by that? 00:12:18,060 --> 00:12:23,660 [Eric] Um, a lot of times there will be very robust discussions in Slack about- 00:12:23,660 --> 00:12:23,670 [John] Oh 00:12:23,670 --> 00:12:24,440 [Eric] ... a certain topic- 00:12:24,440 --> 00:12:24,520 [John] Mm-hmm 00:12:25,620 --> 00:12:29,910 [Eric] ... that will sort of form into, "Hey, this is really interesting. We should write about this," right? 00:12:29,910 --> 00:12:30,020 [John] Right. 00:12:30,020 --> 00:12:42,260 [Eric] Or another great example is an engineer notices something or builds something that's, like, very interesting, you know? So, uh, an example would be, 00:12:43,340 --> 00:12:43,740 [Eric] um, 00:12:44,760 --> 00:12:57,860 [Eric] uh... Here's a really interesting example. We noticed, uh, we noticed in the past couple of months there was an attack on our system where someone was trying to steal inference. 00:12:57,860 --> 00:12:58,680 [John] Okay. 00:12:58,720 --> 00:12:59,080 [Eric] And so- 00:12:59,080 --> 00:12:59,580 [John] Oh, interesting 00:12:59,580 --> 00:13:05,540 [Eric] ... like, we have a, a help, uh, like a docs, like ask AI in our docs. 00:13:05,540 --> 00:13:05,560 [John] Like an agent, yeah. 00:13:05,560 --> 00:13:08,740 [Eric] Right? Like an agent, like a, a support agent, right? 00:13:10,220 --> 00:13:14,550 [Eric] And so anyone on the internet can go ask the agent a question, right? 00:13:14,550 --> 00:13:14,640 [John] Mm-hmm. 00:13:14,640 --> 00:13:31,260 [Eric] And this is actually interesting. I, I, this is... Probably a lot of people don't know this, but there's an entire, like, market around stealing inference because if you can figure out creative prompts to get the model to, to go around its guardrails that are built in- 00:13:31,260 --> 00:13:31,660 [John] Yeah 00:13:31,660 --> 00:13:34,910 [Eric] ... you can actually get it to just run normal jobs and, like- 00:13:34,910 --> 00:13:34,920 [John] Yeah. 00:13:34,920 --> 00:13:36,250 [Eric] And so people will literally do that- 00:13:36,250 --> 00:13:36,460 [John] Rephrase 00:13:36,460 --> 00:13:37,780 [Eric] ... and resell the inference, right? 00:13:37,980 --> 00:13:38,000 [John] Yeah. 00:13:38,000 --> 00:13:39,570 [Eric] Right, 'cause you're not paying to use the help- 00:13:39,570 --> 00:13:39,630 [John] Yeah 00:13:39,630 --> 00:13:40,240 [Eric] ... the, the support that . 00:13:40,240 --> 00:13:43,040 [John] Which, which the famous, one of the famous 1st ones was the Chipotle- 00:13:43,040 --> 00:13:43,250 [Eric] Exactly 00:13:43,250 --> 00:13:43,820 [John] ... support agent. 00:13:43,820 --> 00:13:47,340 [Eric] Yeah, yeah. So just people, like, building apps based, building apps- 00:13:47,340 --> 00:13:47,410 [John] Uh-huh 00:13:47,410 --> 00:13:48,960 [Eric] ... using inference from Chipotle's- 00:13:48,960 --> 00:13:48,970 [John] Yeah 00:13:48,970 --> 00:13:49,600 [Eric] ... support- 00:13:49,600 --> 00:13:49,700 [John] Right 00:13:49,700 --> 00:13:50,880 [Eric] ... bot [laughs] on their website. 00:13:50,880 --> 00:13:51,210 [John] Right. 00:13:51,210 --> 00:13:51,860 [Eric] Right? Which is- 00:13:51,860 --> 00:13:52,160 [John] Yeah 00:13:52,160 --> 00:13:52,410 [Eric] ... amazing. 00:13:52,410 --> 00:13:53,579 [John] It's crazy, yeah. 00:13:53,580 --> 00:13:53,900 [Eric] Um, 00:13:55,060 --> 00:14:01,140 [Eric] this is a great example of, like, whoa, what's happening to this endpoint, you know? And we had all these systems set up to, like, catch that and- 00:14:01,140 --> 00:14:01,240 [John] Yeah 00:14:01,240 --> 00:14:02,250 [Eric] ... you know, mitigate that- 00:14:02,250 --> 00:14:02,250 [John] Sure 00:14:02,250 --> 00:14:08,320 [Eric] ... or whatever. But we also learned a lot about the shape of these inference theft attacks, which is- 00:14:08,320 --> 00:14:08,450 [John] Sure 00:14:08,450 --> 00:14:10,210 [Eric] ... super interesting. And so that's, like, a very 00:14:11,340 --> 00:14:14,430 [Eric] rich subject. We have a lot of primary data around- 00:14:14,430 --> 00:14:14,430 [John] Yeah 00:14:14,430 --> 00:14:15,969 [Eric] ... like, what we observed and the telemetry- 00:14:15,969 --> 00:14:15,969 [John] Mm-hmm 00:14:15,969 --> 00:14:16,760 [Eric] ... around that internally. 00:14:18,440 --> 00:14:24,980 [Eric] Uh, there's a bunch of interesting stuff when you research, like, the Chipotle situation. And so really a bunch of context gathering. 00:14:24,980 --> 00:14:25,580 [John] Yeah. 00:14:25,580 --> 00:14:46,160 [Eric] And then that often looks like generating some sort of high-level report or summary of all of this information, um, that I can use a r- as a reference point. And then I'll actually, like, go back and forth with our content agent to develop a very clear understanding of really the entire picture. So it's an- 00:14:46,160 --> 00:14:46,270 [John] Right 00:14:46,270 --> 00:14:47,620 [Eric] ... accelerated research process, 00:14:48,840 --> 00:14:50,120 [Eric] which is, which is super fun, 00:14:51,380 --> 00:14:56,040 [Eric] uh, because you can consume a lot of really interesting information in a short amount of time. 00:14:56,040 --> 00:14:56,220 [John] Right. 00:14:56,220 --> 00:15:00,680 [Eric] Um, so I, it's really kind of accelerated research, but accelerated learning in many ways. 00:15:01,840 --> 00:15:16,220 [Eric] And throughout that process, just the way that my brain works, I tend to already be working on conviction around what I believe my core arguments and the core narrative- 00:15:16,220 --> 00:15:16,560 [John] Right 00:15:16,560 --> 00:15:19,550 [Eric] ... arc of the piece will be. And, 00:15:20,800 --> 00:15:21,120 [Eric] um, 00:15:22,660 --> 00:15:33,180 [Eric] I... So when I go into the drafting process, I generally... I actually don't use our content agent for outlining a lot. 00:15:33,180 --> 00:15:33,190 [John] Okay. 00:15:33,190 --> 00:15:40,140 [Eric] Um, I generally develop that sort of in my own head. And then when I go into the drafting process, 00:15:41,300 --> 00:15:48,140 [Eric] I will often write out an outline, like a fairly detailed sort of bullet point outline- 00:15:48,140 --> 00:15:48,400 [John] Right 00:15:48,400 --> 00:16:01,820 [Eric] ... and then feed that into our content agent to make a 1st pass at a draft. And it's usually detailed enough to get a pretty good 1st draft out of the gate from the content agent. 00:16:01,820 --> 00:16:02,160 [John] Right. 00:16:02,160 --> 00:16:08,200 [Eric] But at this point, I've been doing a huge amount of research and a, and just really a lot of thinking- 00:16:08,200 --> 00:16:08,380 [John] Right 00:16:08,380 --> 00:16:14,780 [Eric] ... to, like, settle the conviction in my own mind about what we're trying to communicate- 00:16:14,780 --> 00:16:14,860 [John] Right 00:16:14,860 --> 00:16:16,490 [Eric] ... and what I believe the narrative arc should be. 00:16:16,490 --> 00:16:20,100 [John] Yeah. Okay, so this is really interesting because from what you've described, 00:16:21,160 --> 00:16:23,399 [John] I'm imagining a pie chart, data person. 00:16:23,400 --> 00:16:23,620 [Eric] Mm-hmm. 00:16:23,620 --> 00:16:29,700 [John] So if you look at your pie chart, the amount of time you s- you spend on your team's AI tokens, let's frame it this way. 00:16:29,700 --> 00:16:30,260 [Eric] Mm-hmm. 00:16:30,260 --> 00:16:37,279 [John] So your team uses however many AI tokens a month. You spend a lot of tokens on review, it sounds right. 00:16:37,280 --> 00:16:37,440 [Eric] Yep. 00:16:37,440 --> 00:16:39,460 [John] You spend a lot of tokens on research. 00:16:39,460 --> 00:16:40,080 [Eric] Yep. 00:16:40,080 --> 00:16:40,090 [John] Um, 00:16:41,100 --> 00:16:50,040 [John] and I don't... Maybe you can give me rough percentages. Research, review, and then actual, like, it doing some kind of writing. What do you think the breakdown is between those 3? 00:16:50,040 --> 00:16:51,920 [Eric] I actually know this because I recently- 00:16:51,920 --> 00:16:52,080 [John] Oh 00:16:52,080 --> 00:16:52,950 [Eric] ... ran statistics- 00:16:52,950 --> 00:16:53,000 [John] Cool 00:16:53,000 --> 00:16:57,320 [Eric] ... on this for, uh... I wrote the keynotes for our- 00:16:57,320 --> 00:16:57,450 [John] Yeah 00:16:57,450 --> 00:16:58,640 [Eric] ... recent ship conferences- 00:16:58,640 --> 00:16:59,010 [John] Right, right 00:16:59,010 --> 00:16:59,740 [Eric] ... for, for the executives. 00:17:00,780 --> 00:17:02,490 [Eric] I mean, I say wrote. It, it's not like I- 00:17:02,490 --> 00:17:02,490 [John] Right 00:17:02,490 --> 00:17:04,660 [Eric] ... you know. That's a very collaborative process. 00:17:04,660 --> 00:17:06,120 [John] Right, sure. 00:17:06,120 --> 00:17:06,540 [Eric] Uh, 00:17:07,800 --> 00:17:08,240 [Eric] so 00:17:10,740 --> 00:17:12,810 [Eric] the... It was 00:17:14,400 --> 00:17:15,480 [Eric] 50%. 00:17:17,400 --> 00:17:19,580 [Eric] It was 40 to 50% research. 00:17:19,580 --> 00:17:20,110 [John] Yep. 00:17:20,110 --> 00:17:21,740 [Eric] And context. 00:17:21,740 --> 00:17:22,000 [John] Yep, gathering. 00:17:22,000 --> 00:17:23,260 [Eric] Like, processing context, yeah. 00:17:23,260 --> 00:17:23,760 [John] Right. 00:17:23,760 --> 00:17:31,320 [Eric] Like, burning tokens on... And I'll give you just a couple of examples, right? Oh, here's an interesting conversation that happened in Slack that's really relevant, right? 00:17:31,320 --> 00:17:31,840 [John] Right. 00:17:31,840 --> 00:17:34,330 [Eric] So let's pull that in, you know, sort of sanitize it. 00:17:34,330 --> 00:17:34,420 [John] Yeah. 00:17:34,420 --> 00:17:36,659 [Eric] You know, create a markdown file out of it. 00:17:36,660 --> 00:17:37,920 [John] Yeah, right. 00:17:37,920 --> 00:17:38,240 [Eric] Um, 00:17:39,520 --> 00:17:53,400 [Eric] the, uh... Or we have a call. I record the, like a, you know... I would, you know, for example, like, a call with our COO to go through their speech, and they would, like, do a practice run of their speech, and we would discuss all these specific points. 00:17:53,400 --> 00:17:53,800 [John] Right. 00:17:53,800 --> 00:18:01,320 [Eric] You know, and there are other people in the room too, and so we would go through a refinement process around, okay, you know, this is working, this isn't working. 00:18:01,320 --> 00:18:01,960 [John] Right. 00:18:01,960 --> 00:18:04,490 [Eric] These are points that I wanna make that aren't getting through, et cetera, right? 00:18:05,540 --> 00:18:15,452 [Eric] Um, and so that's a whole transcript. Uh, and so again, the same thing. So pull that in as context, right? But then there's all sorts of other interesting, you know- 00:18:15,452 --> 00:18:22,012 [Eric] ... competitive research, um, internal documentation on certain products. You know, there's all sorts of stuff, right? So- 00:18:22,012 --> 00:18:22,392 [John] Right. 00:18:22,392 --> 00:18:27,272 [Eric] And just to give you, like, a, a rough sizing here, um, 00:18:28,472 --> 00:18:28,872 [Eric] the... 00:18:29,912 --> 00:18:34,152 [Eric] we're talking hundreds and hundreds of documents. 00:18:34,152 --> 00:18:34,352 [John] Yeah. 00:18:34,352 --> 00:18:42,332 [Eric] So I mean, um, 3, probably 300 plus doc- like, distinct documents that are in the context for this project, right? 00:18:42,332 --> 00:18:43,432 [John] Okay. And, and it- 00:18:43,432 --> 00:18:43,632 [Eric] It's a lot 00:18:43,632 --> 00:18:48,312 [John] ... and in a document might be we took this Slack conversation, put it in a document. 00:18:48,312 --> 00:18:48,512 [Eric] Yep. 00:18:48,512 --> 00:18:50,752 [John] We have this actual Google Doc. Like, it could be- 00:18:50,752 --> 00:18:51,272 [Eric] This, yes. 00:18:51,272 --> 00:18:51,372 [John] Yeah. 00:18:51,372 --> 00:18:52,742 [Eric] This executive gave a presentation- 00:18:52,742 --> 00:18:52,822 [John] Right 00:18:52,822 --> 00:18:53,611 [Eric] ... at this conference. 00:18:53,612 --> 00:18:53,662 [John] Yeah. 00:18:53,662 --> 00:18:54,952 [Eric] So I pull that PDF in, right? 00:18:54,952 --> 00:18:55,442 [John] Perfect. 00:18:55,442 --> 00:18:59,372 [Eric] This is a screenshot of a chart. This is a report that I got. This is- 00:18:59,372 --> 00:18:59,932 [John] Yeah 00:18:59,932 --> 00:19:00,642 [Eric] ... a summary- 00:19:00,642 --> 00:19:00,642 [John] Yeah 00:19:00,642 --> 00:19:03,472 [Eric] ... because I had the agent go out and do a bunch of research and write a summary, right? 00:19:03,472 --> 00:19:03,732 [John] Right. 00:19:03,732 --> 00:19:05,712 [Eric] So we're talking hundreds of documents- 00:19:05,712 --> 00:19:05,842 [John] Okay. Yep 00:19:05,842 --> 00:19:07,832 [Eric] ... and hundreds of thousands of words. 00:19:07,832 --> 00:19:11,552 [John] Yeah. Okay, so that's 40, 40 or 50% of your budget is that. 00:19:11,552 --> 00:19:11,772 [Eric] Yep. 00:19:11,772 --> 00:19:17,792 [John] And then you have review left, and then you have the actual, like, writing. Like, what do you think the breakdown is for the- 00:19:17,792 --> 00:19:17,802 [Eric] Yeah 00:19:17,802 --> 00:19:19,672 [John] ... your last 50%? 00:19:19,672 --> 00:19:22,091 [Eric] So, um- 00:19:22,092 --> 00:19:23,552 [John] Let's call review, like, 00:19:24,712 --> 00:19:26,932 [John] e- editing and review, maybe we can put them together. 00:19:26,932 --> 00:19:32,272 [Eric] Yeah. I'll t- I'll be, I'll be specific about that. So in my process, I 00:19:33,332 --> 00:19:38,532 [Eric] generally, once I... So when I get the outline and I have the narrative arc and it, and it creates a draft, 00:19:39,572 --> 00:19:48,252 [Eric] I am basically going back and forth with the content agent on pretty specific edits of chunks, right? 00:19:48,252 --> 00:19:49,052 [John] Right. 00:19:49,052 --> 00:19:51,802 [Eric] So, like, this section isn't working. And this is, 00:19:52,852 --> 00:19:54,392 [Eric] this is one of the unique things about writing is, 00:19:55,572 --> 00:19:57,492 [Eric] uh, with AI, is that 00:19:59,172 --> 00:20:03,692 [Eric] if you make a significant change in one part of the document, 00:20:04,972 --> 00:20:09,212 [Eric] it, it impacts the entire piece as a whole. 00:20:09,212 --> 00:20:09,852 [John] Sure. 00:20:09,852 --> 00:20:17,072 [Eric] And AI is not great, in my experience, at, like, stepping back and understanding what those impacts are. 00:20:17,072 --> 00:20:17,452 [John] Mm-hmm. Sure. 00:20:17,452 --> 00:20:24,152 [Eric] So it's like, okay, we can make this paragraph or, like, this section. Like, let's change this, right? And it's like, well, it kinda changes the overall story- 00:20:24,152 --> 00:20:24,202 [John] Yeah, right 00:20:24,202 --> 00:20:27,821 [Eric] ... so we kinda have to go... we have to step back and, and sort of think about, 00:20:28,852 --> 00:20:31,132 [Eric] think about how that impacts the entire narrative arc. 00:20:31,132 --> 00:20:31,292 [John] Right. 00:20:31,292 --> 00:20:32,852 [Eric] Um, and so 00:20:34,832 --> 00:20:41,402 [Eric] th- I would say another, like, 25 to 30% is, is doing that, right? 00:20:41,402 --> 00:20:41,432 [John] Right. 00:20:41,432 --> 00:20:54,692 [Eric] Is, is actually working through the process of drafting and, like, okay, you know, I get a... Let's say I have a call. We do a table read. Um, we get a bunch of feedback. I process that feedback in a transcript. 00:20:54,692 --> 00:20:55,472 [John] Right. 00:20:55,472 --> 00:20:57,352 [Eric] And, okay, we need to change some things, right? 00:20:58,562 --> 00:21:09,821 [Eric] And from there, the process is really like, okay... And AI is great for, like, let's explore different ways to change this. Um, and I would ex- I would 00:21:11,292 --> 00:21:19,192 [Eric] probably explain it, it's most accurate in my workflow as really making the iteration process rapid. Like, trying- 00:21:19,192 --> 00:21:19,222 [John] Mm-hmm 00:21:19,222 --> 00:21:21,092 [Eric] ... a bunch of different things- 00:21:21,092 --> 00:21:21,312 [John] Yeah 00:21:21,312 --> 00:21:24,072 [Eric] ... while I, like, look at the entire piece- 00:21:24,072 --> 00:21:24,132 [John] Right 00:21:24,132 --> 00:21:26,582 [Eric] ... to see how it's going to be impacted by this, right? 00:21:26,582 --> 00:21:27,512 [John] Yeah, yeah. That makes sense. 00:21:27,512 --> 00:21:27,642 [Eric] Um, 00:21:28,892 --> 00:21:34,881 [Eric] and that works really well for me, um, to, like, go through a bunch of those iterations- 00:21:34,881 --> 00:21:34,881 [John] Right 00:21:34,881 --> 00:21:36,661 [Eric] ... and sort of, like, step and, you know, look back at that. 00:21:36,661 --> 00:21:36,661 [John] Right. 00:21:36,661 --> 00:21:42,232 [Eric] And one thing I will say generally is that every single person on my team 00:21:43,432 --> 00:21:48,212 [Eric] is frustrated with AI when they use it in the writing process. 00:21:48,212 --> 00:21:48,632 [John] Yeah. 00:21:48,632 --> 00:21:50,372 [Eric] And it's not because AI is not great. 00:21:50,372 --> 00:21:50,532 [John] Yeah. 00:21:50,532 --> 00:22:01,572 [Eric] But it's because writing is really hard in general, and you re- like, the amount of granular steering and everything, it, it, it can actually [laughs] just be kind of frustrating. 00:22:01,572 --> 00:22:05,072 [John] What, what's your favorite quote on writing? Write- like, writing is clearer thinking or something? 00:22:05,072 --> 00:22:08,012 [Eric] Writing is, writing is really just a great deal of hard thinking. 00:22:08,012 --> 00:22:08,682 [John] Yeah. Great deal of hard thinking. 00:22:08,682 --> 00:22:09,762 [Eric] David Brooks. Yeah. 00:22:09,762 --> 00:22:11,512 [John] Yeah, there you go. Um, but- 00:22:11,512 --> 00:22:13,441 [Eric] And then the last... To close it out- 00:22:13,441 --> 00:22:13,441 [John] Yeah 00:22:13,441 --> 00:22:19,592 [Eric] ... the last percentage points, which, I don't know, 10 or 15% or something like that, is, um, 00:22:21,412 --> 00:22:22,752 [Eric] just other random stuff. 00:22:22,752 --> 00:22:23,212 [John] Yeah. 00:22:23,212 --> 00:22:25,542 [Eric] Right? Like, I don't know, look at this image- 00:22:25,542 --> 00:22:25,542 [John] Yeah 00:22:25,542 --> 00:22:26,152 [Eric] ... you know, process this image. 00:22:26,152 --> 00:22:34,051 [John] But if you broke down, like, review from, like, active use in the writing process, like, what do you think that breakdown is? 00:22:34,052 --> 00:22:39,772 [Eric] Yeah, I mean, that's t- 20 to 25% of, like, actively- 00:22:39,772 --> 00:22:39,782 [John] In each bucket. Okay 00:22:39,782 --> 00:22:40,672 [Eric] ... like, I'm- 00:22:40,672 --> 00:22:40,862 [John] Yeah, yeah 00:22:40,862 --> 00:22:42,551 [Eric] ... writing and creating content. 00:22:42,552 --> 00:22:43,442 [John] Perfect. Okay, great. You're gonna make- 00:22:43,442 --> 00:22:43,442 [Eric] Yes 00:22:43,442 --> 00:22:44,892 [John] ... my point really well now. 00:22:44,892 --> 00:22:45,192 [Eric] Okay. 00:22:45,192 --> 00:22:45,751 [John] I, I was- [laughs] 00:22:45,752 --> 00:22:45,892 [Eric] Yeah 00:22:45,892 --> 00:22:46,952 [John] ... I wasn't sure for a minute. 00:22:46,952 --> 00:22:47,312 [Eric] Um- 00:22:47,312 --> 00:23:15,872 [John] But essentially what you're saying is 70% of your AI usage in this process, 70-plus percent, is not using the AI to write, to, like, write the words. Like, 50% is, like, context gathering and summary and all that other stuff. The other 30% is, is about, like, review and conformance to a standard and, and let's call it even formatting, spelling and grammar, whatever. And then a very small percentage is, like, active use, like actually 00:23:17,132 --> 00:23:17,632 [John] shaping 00:23:18,932 --> 00:23:19,952 [John] the document. Is that fair? 00:23:21,612 --> 00:23:26,832 [Eric] Roughly, yes. So the, uh, I w- the way that I would... 00:23:29,212 --> 00:23:33,472 [Eric] where, like, the actual percentage of generating net new content is- 00:23:33,472 --> 00:23:33,742 [John] Yes 00:23:33,742 --> 00:23:34,772 [Eric] ... is pretty small. 00:23:34,772 --> 00:23:35,472 [John] That, yeah. 00:23:35,472 --> 00:23:35,962 [Eric] Is, is pretty small. 00:23:35,962 --> 00:23:40,762 [John] But I think that's the exact opposite of the way the vast majority of people have used AI in writing. 00:23:40,762 --> 00:23:40,772 [Eric] Yes. 00:23:40,772 --> 00:23:41,752 [John] That's kind of my point. 00:23:41,752 --> 00:23:42,431 [Eric] Yes. 00:23:42,432 --> 00:23:43,232 [John] Um- 00:23:43,232 --> 00:23:44,332 [Eric] Yes, it is. It is. 00:23:44,332 --> 00:23:44,871 [John] Yeah. 00:23:44,872 --> 00:24:01,792 [Eric] I'll give you an example. Another, another guy on my team, um, who is, like, a, one of them, he is, he reminds me a lot of you, where it's like, man, you're using AI in, like, so many interesting ways, so much to learn. He will actually... He has a self, a self-optimizing loop. And so- 00:24:01,792 --> 00:24:01,801 [John] Okay 00:24:01,801 --> 00:24:02,492 [Eric] ... he'll run, 00:24:03,912 --> 00:24:07,932 [Eric] uh, he'll run our content agent on a loop for hours at a time- 00:24:08,972 --> 00:24:08,982 [John] Okay 00:24:08,982 --> 00:24:10,892 [Eric] ... um, to 00:24:12,232 --> 00:24:15,352 [Eric] get a draft really, really close. 00:24:15,352 --> 00:24:16,332 [John] Okay. Yeah. 00:24:17,400 --> 00:24:28,540 [Eric] And it's really... He has a really interesting way of... This is really cool, that the, our content agent, it, you can, um, like, well, w- we do it, it's like local markdown files is sort of how we- 00:24:28,540 --> 00:24:28,610 [John] Right 00:24:28,610 --> 00:24:29,790 [Eric] ... how we, uh, use it locally 00:24:30,860 --> 00:24:32,980 [Eric] in, in a harness that we use, you know, cloud code. 00:24:32,980 --> 00:24:33,240 [John] Mm-hmm. 00:24:33,240 --> 00:24:43,120 [Eric] Um, and he'll actually go in and write pretty detailed comments in the markdown file above, like, each section- 00:24:43,120 --> 00:24:43,150 [John] Oh, interesting 00:24:43,150 --> 00:24:44,700 [Eric] ... through the entire document and do- 00:24:44,700 --> 00:24:44,710 [John] Okay 00:24:44,710 --> 00:24:46,250 [Eric] ... like a very, very deep review- 00:24:47,340 --> 00:24:47,420 [John] Mm-hmm 00:24:47,420 --> 00:24:52,440 [Eric] ... and then describe what, like, what's wrong and, and- 00:24:52,440 --> 00:24:52,450 [John] Right 00:24:52,450 --> 00:24:57,450 [Eric] ... you know, the outcome that needs to be met and then run it on a loop for hours. And it will, like, regenerate, check its work, regenerate- 00:24:57,450 --> 00:24:57,600 [John] Sure 00:24:57,600 --> 00:24:58,160 [Eric] ... check its work- 00:24:58,160 --> 00:24:58,250 [John] Yeah, yeah 00:24:58,250 --> 00:24:58,980 [Eric] ... and sort of run it on a loop. 00:24:58,980 --> 00:24:59,960 [John] Sure. That's cool. 00:24:59,960 --> 00:25:04,460 [Eric] And so when he goes in to do his final edits, it's actually just almost pure human review- 00:25:04,460 --> 00:25:04,540 [John] Mm-hmm 00:25:04,540 --> 00:25:06,480 [Eric] ... and he's gotten it pretty close. 00:25:06,480 --> 00:25:06,960 [John] Yeah. 00:25:06,960 --> 00:25:08,740 [Eric] Um, but he runs it on a loop, whereas I- 00:25:08,740 --> 00:25:08,750 [John] Yeah 00:25:08,750 --> 00:25:10,540 [Eric] ... more am doing sort of like... 00:25:11,880 --> 00:25:15,420 [Eric] I wouldn't call it, like, straight line edits, but a little bit closer to that. 00:25:15,420 --> 00:25:16,160 [John] Yeah. 00:25:16,160 --> 00:25:19,850 [Eric] Um, you know, and we sort of both get to the same outcome, but we- 00:25:19,850 --> 00:25:19,850 [John] Right 00:25:19,850 --> 00:25:20,820 [Eric] ... use it very differently. 00:25:20,820 --> 00:25:20,960 [John] Right. 00:25:20,960 --> 00:25:23,130 [Eric] And that's okay. I think that's a great part about it. 00:25:23,130 --> 00:25:23,239 [John] Right. 00:25:23,240 --> 00:25:24,500 [Eric] Um... 00:25:24,500 --> 00:25:30,040 [John] Okay. So for the average person that's trying to write, one, I wanna come back to some writing rules- 00:25:30,040 --> 00:25:30,050 [Eric] Mm-hmm 00:25:30,050 --> 00:25:31,160 [John] ... I wanna get your opinion on. 00:25:31,160 --> 00:25:31,720 [Eric] Yep. 00:25:31,720 --> 00:25:40,080 [John] One of the hacks that's been going around, I'll call it a hack, um, the internet has been this, these writing rules that you can use to make your writing better. 00:25:40,080 --> 00:25:40,560 [Eric] Mm-hmm. 00:25:40,560 --> 00:25:46,060 [John] So we'll actually look at a list of the rules here in a 2nd, but I am curious, 00:25:47,140 --> 00:25:50,200 [John] like, as, as I'm reading the rules to you, tell me where, 00:25:51,480 --> 00:25:56,540 [John] where in the process you think they make sense. 'Cause we've gone over, I think, the most important part, which is the process- 00:25:56,540 --> 00:25:56,760 [Eric] Yep 00:25:56,760 --> 00:26:01,800 [John] ... right, of spending a lot of time on context gathering, like having really detailed reviews and standards and then- 00:26:01,800 --> 00:26:01,900 [Eric] Mm-hmm 00:26:01,900 --> 00:26:05,140 [John] ... editing. But here are the rules, so I'm curious where would you use these- 00:26:05,140 --> 00:26:05,420 [Eric] Okay 00:26:05,420 --> 00:26:10,180 [John] ... in a process? Okay. So the one that's been, um, popular this week 00:26:11,360 --> 00:26:13,220 [John] is Orwell's 6 rules- 00:26:13,220 --> 00:26:13,520 [Eric] Yep 00:26:13,520 --> 00:26:13,990 [John] ... for writing. 00:26:15,260 --> 00:26:23,940 [John] All right. Rule number one, I love this one. This is perfect for AI. N- And these are, you know, obviously not [laughs] written way before AI. 00:26:23,940 --> 00:26:24,550 [Eric] [laughs] Yes. 00:26:24,550 --> 00:26:30,150 [John] Um, although, I don't know, [laughs] based on, based on Orwell's writing, like maybe he knew something [laughs] that we didn't. [laughs] 00:26:30,150 --> 00:26:31,780 [Eric] That's, that is true. 00:26:31,780 --> 00:26:32,010 [John] All right. 00:26:32,010 --> 00:26:35,060 [Eric] Didn't he write, like, Down and Out in London and Paris or something? 00:26:35,060 --> 00:26:35,980 [John] Y- yeah. Yeah. 00:26:35,980 --> 00:26:37,280 [Eric] Yeah. He's... Yeah, he did- 00:26:37,280 --> 00:26:37,980 [John] In 1984 is- 00:26:37,980 --> 00:26:39,360 [Eric] Yeah, 1984 is the big one- 00:26:39,360 --> 00:26:39,370 [John] ... probably more known 00:26:39,370 --> 00:26:42,080 [Eric] ... but he actually is, his other writing is unbelievable. 00:26:42,080 --> 00:26:42,200 [John] Yeah. 00:26:42,200 --> 00:26:43,350 [Eric] I mean, he really is an incredible writer. 00:26:43,350 --> 00:26:54,080 [John] Yeah. All right. I love this 1st one. It is perfect for our current AI world. Okay. Never use a metaphor, simile, or other figure of speech which you're used to seeing in print. 00:26:54,080 --> 00:26:54,990 [Eric] Hmm. [laughs] 00:26:54,990 --> 00:26:57,120 [John] So this is his rules for politics in the English language. 00:26:57,120 --> 00:26:57,280 [Eric] Yeah. 00:26:58,540 --> 00:27:02,800 [John] Which we know, like, some of the worst AI writing is bad metaphors and similes. 00:27:02,800 --> 00:27:03,860 [Eric] Absolutely. 00:27:03,860 --> 00:27:04,070 [John] So that's rule- 00:27:04,070 --> 00:27:05,000 [Eric] You're absolutely right. 00:27:05,000 --> 00:27:10,740 [John] Rule number one. [laughs] Rule number one. All right, number 2, never use a long word where a short one will do. 00:27:12,180 --> 00:27:13,279 [Eric] Yep. 00:27:13,280 --> 00:27:15,840 [John] If it is possible to cut a word out, always cut it out. 00:27:17,100 --> 00:27:26,689 [John] That was number 3. Number 4, never use the passive where you can use the active. I feel like that's a strong English teacher, like, uh, my English teacher in my back of my mind is, is telling me that. 00:27:26,689 --> 00:27:27,540 [Eric] Absolutely. Yeah, totally. 00:27:27,540 --> 00:27:41,880 [John] All right, 5, never use a foreign phrase, a scientific word, or a jargon word if you can think of an everyday English equivalent. All right, that's number 5. And number 6, break any of these rules sooner than say anything outright, um, barbarous. 00:27:43,100 --> 00:27:43,880 [Eric] [laughs] 00:27:43,880 --> 00:27:45,820 [John] Which is great. Like, that's a great rule too. 00:27:45,820 --> 00:27:46,860 [Eric] Yeah. Yeah. 00:27:46,860 --> 00:27:56,040 [John] Yeah. So reaction to the rules and then, like, if you guys use those rules or used a, a version of those rules, like, where in your process do you think you'd use that type of shaping? 00:27:58,140 --> 00:28:03,160 [Eric] I, I think the... I mean, the rules are great. They are just really good rules. 00:28:03,160 --> 00:28:04,240 [John] Very simple. 00:28:04,240 --> 00:28:07,780 [Eric] And very simple. And the... 00:28:08,920 --> 00:28:13,560 [Eric] I, I mentioned David McCullough earlier, and 00:28:15,040 --> 00:28:19,770 [Eric] I, I, he is, sticks out in my mind as exemplary for, 00:28:21,600 --> 00:28:24,980 [Eric] you know, for following those rules and breaking them when it makes sense- 00:28:24,980 --> 00:28:24,990 [John] Mm-hmm 00:28:24,990 --> 00:28:29,800 [Eric] ... which I think is actually probably true of most good writers. Um, 00:28:31,440 --> 00:28:37,120 [Eric] a- another author that comes to mind, which I'll, I'll get back to the rules and AI in a second, but is Cormac McCarthy- 00:28:37,120 --> 00:28:37,200 [John] Okay 00:28:37,200 --> 00:28:46,990 [Eric] ... who is sort of a notorious rule breaker. So he doesn't use any, he, he doesn't use any markers for dialogue, so no quotes. And so- 00:28:46,990 --> 00:28:47,250 [John] Oh, interesting 00:28:47,250 --> 00:28:53,820 [Eric] ... it's, it's actually shocking for people to, like, try to read his stuff initially because there's no quotes. It's just like this person saying, this person- 00:28:53,820 --> 00:28:53,970 [John] Yeah, yeah 00:28:53,970 --> 00:28:54,920 [Eric] ... just like line, you know- 00:28:54,920 --> 00:28:55,030 [John] Right 00:28:55,030 --> 00:28:56,720 [Eric] ... line. Um, 00:28:58,020 --> 00:29:25,300 [Eric] but he also sticks out to me as someone who uses very plain English language to communicate these extremely deep and oftentimes dark, you know, realities about humanity. And it's, it's... So it, it... The language is very accessible, and it's almost deceptive because he just uses plain language. And, and a lot of times- 00:29:25,300 --> 00:29:25,470 [John] Yeah, yeah 00:29:25,470 --> 00:29:30,180 [Eric] ... his characters aren't necessarily, like, deeply educated people. Sometimes they are. 00:29:30,180 --> 00:29:30,600 [John] Right. 00:29:30,600 --> 00:29:34,220 [Eric] Um, but in a lot of cases, you have people who are in rural settings, right? 00:29:34,220 --> 00:29:35,060 [John] Yeah. 00:29:35,060 --> 00:29:35,360 [Eric] Um, 00:29:36,500 --> 00:29:37,860 [Eric] and so 00:29:39,300 --> 00:29:47,120 [Eric] the rules are great. I think that... I, I mean, it's ironic to me that we're sort of going back to... You know, it's like- 00:29:47,120 --> 00:29:47,680 [John] Yeah 00:29:47,680 --> 00:29:51,920 [Eric] ... the, we're going back to this to try to improve I- AI writing, right? 00:29:51,920 --> 00:29:52,020 [John] Yeah. 00:29:52,020 --> 00:29:54,020 [Eric] We're sort of going back to these timeless rules. 00:29:54,020 --> 00:29:54,760 [John] Yeah. 00:29:54,760 --> 00:29:55,120 [Eric] Um, 00:29:56,240 --> 00:29:56,640 [Eric] and 00:29:58,060 --> 00:30:06,300 [Eric] the... We do have rules like that, and they are used both in the review and drafting process. 00:30:06,300 --> 00:30:07,800 [John] Yes. Okay. That makes sense. 00:30:07,800 --> 00:30:17,359 [Eric] Um, you know, and so they sort of apply, uh, in both of those ways. But again, I would say, I mean, at Vercel, we don't... Nothing gets published without h- without- 00:30:17,360 --> 00:30:17,580 [John] Sure 00:30:17,580 --> 00:30:19,400 [Eric] ... a writer doing human review. 00:30:19,400 --> 00:30:20,160 [John] Yeah. 00:30:20,160 --> 00:30:22,232 [Eric] And- 00:30:22,232 --> 00:30:26,732 [Eric] ... the reason we hire really excellent writers is because 00:30:29,092 --> 00:30:32,972 [Eric] they know when to break the rules, or at least they have good intuition about that. 00:30:34,152 --> 00:30:34,952 [Eric] Because, 00:30:36,732 --> 00:30:37,392 [Eric] like he's, like, 00:30:38,852 --> 00:30:39,352 [Eric] Orwell- 00:30:39,352 --> 00:30:39,792 [John] So, r- rule 6, right? 00:30:39,792 --> 00:30:41,912 [Eric] Orwell said, "Break the rules, you know- 00:30:41,912 --> 00:30:42,112 [John] Right 00:30:42,112 --> 00:30:46,392 [Eric] ... before saying something, uh, barbarous." You know? Which is a wonderful word. 00:30:46,392 --> 00:30:46,552 [John] Yeah, sure. 00:30:46,552 --> 00:30:48,292 [Eric] Maybe he could've used a shorter word for that. 00:30:48,292 --> 00:30:48,492 [John] Yeah, fair. 00:30:48,492 --> 00:30:48,972 [Eric] Um, 00:30:50,412 --> 00:30:50,622 [Eric] but, 00:30:51,732 --> 00:30:56,012 [Eric] uh, one of the... If, if you think about writing that's, uh, 00:30:57,052 --> 00:30:58,232 [Eric] that you enjoy, 00:30:59,292 --> 00:31:04,112 [Eric] oftentimes it's so subtle, like, wh- why you actually enjoy it. 00:31:04,112 --> 00:31:04,232 [John] Right. 00:31:04,272 --> 00:31:10,372 [Eric] But it's because it's not systematic, right? It's not documentation. Um- 00:31:10,372 --> 00:31:11,392 [John] Yeah. Right. 00:31:12,532 --> 00:31:12,792 [Eric] And- 00:31:12,792 --> 00:31:15,212 [John] It's rhythmic, but breaks the rhythm every so often. 00:31:15,212 --> 00:31:15,742 [Eric] Exactly. 00:31:15,742 --> 00:31:16,292 [John] Like those types of things. Yeah. 00:31:16,292 --> 00:31:20,412 [Eric] So there are things like cadence that you can't really... 00:31:22,092 --> 00:31:28,252 [Eric] AI is very good at following a system, and it's pretty hard to encode surprising, like- 00:31:28,252 --> 00:31:28,512 [John] Right 00:31:28,512 --> 00:31:30,112 [Eric] ... differences in cadence, right? 00:31:30,112 --> 00:31:34,892 [John] Right. Murder mystery is one of the, um, one of the ones that people use to test writing. I don't know if you know that. 00:31:34,892 --> 00:31:35,382 [Eric] Oh, really? 00:31:35,382 --> 00:31:35,732 [John] Yeah. 00:31:35,732 --> 00:31:36,572 [Eric] Interesting. 00:31:36,572 --> 00:31:41,232 [John] 'Cause, uh, I think when GPT-5 6, that was, that was one of the ones everybody's like, "Okay- 00:31:41,232 --> 00:31:41,342 [Eric] Yeah 00:31:41,342 --> 00:31:44,132 [John] ... have it write a murder mystery. See what it is." And it, and it was better, but still not- 00:31:44,132 --> 00:31:44,512 [Eric] Right 00:31:44,512 --> 00:31:44,672 [John] ... you know, not great. 00:31:44,672 --> 00:31:52,392 [Eric] Yeah, yeah, yeah. Yeah, yeah, yeah. So, um, uh, I have 2... I'm of sort of 2 minds on... 00:31:53,532 --> 00:32:00,612 [Eric] I'm of 2 minds on this. I... You know, a lot of writing out there is bad. I don't say that in a demeaning way, but it's- 00:32:00,612 --> 00:32:00,702 [John] Yeah 00:32:00,702 --> 00:32:03,342 [Eric] ... a difficult skill. Just like writing software- 00:32:03,342 --> 00:32:03,382 [John] Right 00:32:03,382 --> 00:32:04,312 [Eric] ... is a difficult skill. 00:32:04,312 --> 00:32:04,792 [John] Right. 00:32:04,792 --> 00:32:16,672 [Eric] Right? And so, you know, people are like, "Oh, r- you know, AI is, is writing all this code." And it's like, "Well, you know, is it gonna be as good as a human?" It's like, oh, there was so much bad code before. Like- 00:32:16,672 --> 00:32:17,392 [John] Right, right 00:32:17,392 --> 00:32:19,472 [Eric] ... an, an unending amount of bad code. 00:32:19,472 --> 00:32:19,812 [John] Right. 00:32:19,812 --> 00:32:23,962 [Eric] And it's the same thing with writing. There's just a huge amount of poor writing. 00:32:23,962 --> 00:32:23,992 [John] Right. 00:32:23,992 --> 00:32:25,032 [Eric] And so- 00:32:25,032 --> 00:32:35,192 [John] But, but I would say the difference right now is, from what I know, all the best coders in the world are using AI to write code. All the best writers in the world are not using AI to write, in the same way. 00:32:35,192 --> 00:32:35,811 [Eric] Yes, I agree with that. 00:32:35,812 --> 00:32:40,052 [John] I think they're using it as part of their process, but it's, but it hasn't crossed that barrier. 00:32:40,052 --> 00:32:45,332 [Eric] Yes. And there's a very specific reason for that. I mean, multiple reasons for that. 00:32:45,332 --> 00:32:45,371 [John] Right. 00:32:45,372 --> 00:32:55,072 [Eric] But one really specific reason for that, that isn't necessarily obvious when you think about it, is that code is a means to an end. 00:32:55,072 --> 00:32:55,292 [John] Yeah, right. 00:32:55,292 --> 00:32:57,031 [Eric] Like, the code itself is not the product, right? 00:32:57,032 --> 00:32:58,032 [John] Exactly, yeah. 00:32:58,032 --> 00:33:02,402 [Eric] It, it... You run code under the hood of some sort of product- 00:33:02,402 --> 00:33:02,402 [John] Yeah 00:33:02,402 --> 00:33:04,992 [Eric] ... or experience or agent that's running a job, et cetera. 00:33:04,992 --> 00:33:05,792 [John] Right. 00:33:05,792 --> 00:33:10,512 [Eric] When you're writing, the output, the words themselves, are the product. 00:33:10,512 --> 00:33:11,132 [John] Yeah. Yeah. 00:33:11,132 --> 00:33:11,932 [Eric] And, 00:33:13,592 --> 00:33:17,702 [Eric] um, as long... With code, as long as the product works and there aren't security issues- 00:33:17,702 --> 00:33:17,932 [John] Yeah 00:33:17,932 --> 00:33:18,292 [Eric] ... and, you know- 00:33:18,292 --> 00:33:19,212 [John] Mm-hmm 00:33:19,212 --> 00:33:25,412 [Eric] ... whatever, it, it meets some threshold of like, "Okay, this works and it's secure and it's fast and it sort of meets all of our criteria." 00:33:25,412 --> 00:33:25,572 [John] Yeah. 00:33:25,572 --> 00:33:26,072 [Eric] And so it's like, "Okay, good." 00:33:26,072 --> 00:33:29,312 [John] It's fast enough, it works enough, it's secure enough. 00:33:29,312 --> 00:33:29,772 [Eric] Exactly. 00:33:29,772 --> 00:33:30,612 [John] Yeah. 00:33:30,612 --> 00:33:32,092 [Eric] Uh, and there's a... 00:33:33,372 --> 00:33:37,572 [Eric] There's absolutely a point with code where the, like, 00:33:38,872 --> 00:33:44,102 [Eric] practical optimization curve i- it's logarithmic, right? There's a point at which- 00:33:44,102 --> 00:33:44,102 [John] Yeah 00:33:44,102 --> 00:33:50,082 [Eric] ... like, making it better doesn't actually create any additional marginal value. 00:33:50,082 --> 00:33:50,112 [John] The- 00:33:50,112 --> 00:33:50,371 [Eric] Right? 00:33:50,372 --> 00:33:50,572 [John] Yeah. 00:33:51,612 --> 00:33:53,412 [John] The excellence is invisible. 00:33:53,412 --> 00:33:55,492 [Eric] Yeah. And I mean- 00:33:55,492 --> 00:33:57,792 [John] And it's not, doesn't mean it's not valuable, it's just invisible. 00:33:57,792 --> 00:34:06,972 [Eric] 100%. And in fact, I would say, we've talked about this on the show a bu- a bunch, right? Like, handwriting truly excellent code will make you way better at coding with AI- 00:34:06,972 --> 00:34:06,982 [John] [laughs] Yeah 00:34:06,982 --> 00:34:08,371 [Eric] ... because you're building your craft. 00:34:08,372 --> 00:34:08,592 [John] True. 00:34:08,592 --> 00:34:08,772 [Eric] Right? 00:34:08,772 --> 00:34:09,512 [John] True, right. 00:34:09,512 --> 00:34:16,732 [Eric] But, like, literally writing everything perfect in a pragmatic business sense, like for a software company- 00:34:16,732 --> 00:34:16,912 [John] Mm-hmm 00:34:16,912 --> 00:34:18,472 [Eric] ... it, it doesn't actually make sense 'cause you're- 00:34:18,472 --> 00:34:18,552 [John] Right 00:34:18,552 --> 00:34:19,752 [Eric] ... not getting value for- 00:34:19,752 --> 00:34:19,762 [John] Yeah 00:34:19,762 --> 00:34:21,332 [Eric] ... you know, whatever. Um, 00:34:22,352 --> 00:34:25,812 [Eric] but yeah, when you are writing, the words are the product themselves. 00:34:25,812 --> 00:34:25,852 [John] Yeah. 00:34:25,852 --> 00:34:33,572 [Eric] Right? And so I would agree with you. The, the... All the people who I really respect 00:34:34,612 --> 00:34:40,122 [Eric] in, and for whom... Really respect in terms of writing and for whom I know something about their process- 00:34:40,122 --> 00:34:40,132 [John] Right 00:34:40,132 --> 00:34:41,552 [Eric] ... which is not a huge amount- 00:34:41,552 --> 00:34:42,172 [John] Right 00:34:42,172 --> 00:34:42,531 [Eric] ... um, 00:34:43,612 --> 00:34:44,052 [Eric] all 00:34:45,712 --> 00:34:46,792 [Eric] love AI, 00:34:47,832 --> 00:34:48,092 [Eric] you know- 00:34:48,092 --> 00:34:48,212 [John] Mm-hmm 00:34:48,212 --> 00:34:51,362 [Eric] ... because it's so useful for so many things. 00:34:51,362 --> 00:34:51,372 [John] Mm-hmm. 00:34:51,372 --> 00:34:59,412 [Eric] But fight... Will describe using AI in the writing process as somewhat adversarial- 00:34:59,412 --> 00:34:59,522 [John] Yeah 00:34:59,522 --> 00:35:00,152 [Eric] ... and frustrating. 00:35:00,152 --> 00:35:00,432 [John] For sure. 00:35:00,432 --> 00:35:00,712 [Eric] Right? 00:35:00,712 --> 00:35:01,432 [John] Yeah. 00:35:01,432 --> 00:35:05,362 [Eric] It's not like I was running in water before and now I'm not, right? 00:35:05,362 --> 00:35:05,412 [John] No, no. 00:35:05,412 --> 00:35:05,752 [Eric] It's like... 00:35:06,772 --> 00:35:21,852 [Eric] I mean, I'll give you a very, just a very normal, everyday example of how this plays out on our team. It is so common for someone on our team channel or for someone to DM and just say, like, "I hate AI." Or like, you know- 00:35:21,852 --> 00:35:22,032 [John] Right 00:35:22,032 --> 00:35:23,592 [Eric] ... "Opus is so dumb." 00:35:23,592 --> 00:35:23,952 [John] Right. 00:35:23,952 --> 00:35:24,382 [Eric] Right? 00:35:24,382 --> 00:35:24,412 [John] Right. 00:35:24,412 --> 00:35:27,972 [Eric] Or like, "Fable is trying way too hard." 00:35:27,972 --> 00:35:28,632 [John] Right, right. 00:35:28,632 --> 00:35:34,032 [Eric] And that's just exhaust of, like, writing is a hard process. 00:35:34,032 --> 00:35:34,512 [John] Right. 00:35:34,512 --> 00:35:39,112 [Eric] And because words are the final output and because 00:35:40,452 --> 00:35:48,912 [Eric] our standards are such that we need to imbue our product with, like, deep human conviction. Like, it's- 00:35:48,912 --> 00:35:48,922 [John] Right 00:35:48,922 --> 00:35:56,332 [Eric] ... actually, you know, it is. It's just a, it's a tool that is very, very useful, but it's not like it just solves all the, all the writing problems- 00:35:56,332 --> 00:35:56,342 [John] Right 00:35:56,342 --> 00:35:56,852 [Eric] ... for you. 00:35:56,852 --> 00:35:57,672 [John] Yeah. 00:35:57,672 --> 00:35:57,972 [Eric] Um, 00:35:59,032 --> 00:36:01,732 [Eric] so okay, the other set of rules. 00:36:01,732 --> 00:36:16,772 [John] Yes. Okay, we've got one more set of rules. This one, this one's a, um... I remember it from school, but not well, [laughs] 'cause I di- I didn't do a lot of this. I didn't do a lot of writing. I did some writing in school. All right, so Strunk and White's The Elements of Style. 00:36:16,772 --> 00:36:16,852 [Eric] Yes. 00:36:16,852 --> 00:36:17,912 [John] Very popular. Um- 00:36:17,912 --> 00:36:18,432 [Eric] Classic. 00:36:19,692 --> 00:36:20,092 [John] Yeah. 00:36:20,092 --> 00:36:23,432 [Eric] Most writers, you will see it, like, on the shelf behind them on a Zoom call. 00:36:23,432 --> 00:36:25,320 [John] There you go. Yep. 00:36:25,320 --> 00:36:29,960 [John] All right, so these are just... There's a whole list here. This is my truncated list. 00:36:29,960 --> 00:36:30,040 [Eric] Yep. 00:36:30,040 --> 00:36:30,740 [John] Um, okay. 00:36:31,880 --> 00:36:32,920 [John] Omit needless words. 00:36:34,640 --> 00:36:37,260 [John] Okay? Use active voice. We kind of already talked about that. 00:36:37,260 --> 00:36:37,840 [Eric] Yep. 00:36:37,840 --> 00:36:42,060 [John] Write with nouns and verbs. Use definitive, specific, concrete language. 00:36:43,940 --> 00:36:48,280 [John] Put statements in positive form. Say what is rather than what is not. 00:36:48,280 --> 00:36:48,720 [Eric] Yep. 00:36:48,720 --> 00:36:49,460 [John] That's a, a good one. 00:36:50,500 --> 00:36:54,290 [John] Make the paragraph the unit of composition. That's an interesting one. 00:36:54,290 --> 00:36:54,720 [Eric] Hmm. 00:36:54,720 --> 00:37:05,590 [John] Um, keep related words together. This is a good one. Subjects and verbs or modifiers and the words they modify should not be separated by unnecessary distance. I really like that one. 00:37:05,590 --> 00:37:06,580 [Eric] So great. 00:37:06,580 --> 00:37:06,650 [John] Um, 00:37:08,070 --> 00:37:08,120 [John] but- 00:37:08,120 --> 00:37:09,880 [Eric] That's actually a cadence. That's a- 00:37:09,880 --> 00:37:11,060 [John] That's a- 00:37:11,060 --> 00:37:13,840 [Eric] ... gig- that has a significant influence on cadence. 00:37:13,840 --> 00:37:22,120 [John] And place emphatic words at the end. Um, most important or impactful words of a sentence should sit at the end, um, to leave a lasting impression. 00:37:22,120 --> 00:37:23,260 [Eric] Yep. 00:37:23,260 --> 00:37:24,429 [John] Um, so there's a whole- 00:37:24,429 --> 00:37:24,940 [Eric] But, yeah, the... Yes. Yeah. 00:37:24,940 --> 00:37:28,760 [John] There's a whole, like, longer list. I think there's, like, 26, um, but those are just kind of snippets- 00:37:28,760 --> 00:37:28,850 [Eric] Yeah 00:37:28,850 --> 00:37:31,380 [John] ... pulled from it. Um, but, okay. So 00:37:32,500 --> 00:37:35,920 [John] from... A, would you have anything to add to that list, like, from your personal, 00:37:37,020 --> 00:37:39,170 [John] um, mental model of how you should- 00:37:39,170 --> 00:37:39,170 [Eric] Hmm 00:37:39,170 --> 00:37:41,560 [John] ... write? And then B, like, what's your favorite one, 00:37:42,880 --> 00:37:43,410 [John] kind of, uh, 00:37:44,900 --> 00:37:46,320 [John] on that list? 00:37:46,320 --> 00:37:46,760 [Eric] Uh, 00:37:48,500 --> 00:37:51,060 [Eric] the... Okay, anything to add, 00:37:52,560 --> 00:37:52,740 [Eric] um- 00:37:52,740 --> 00:37:54,380 [John] And that's abbreviated. Like I said, there's- 00:37:54,440 --> 00:37:55,480 [Eric] Yeah, there's a lot of them 00:37:55,480 --> 00:37:55,530 [John] ... 26 of them, so. 00:37:55,530 --> 00:37:58,250 [Eric] Yeah. I mean, I don't know if I... Adding... That's a pretty solid- 00:37:58,250 --> 00:37:58,870 [John] Yeah, I mean [laughs]- 00:37:58,870 --> 00:37:58,950 [Eric] ... list to add to the- 00:37:58,950 --> 00:37:59,930 [John] ... it is a solid- 00:37:59,930 --> 00:38:00,770 [Eric] ... to The Elements of Style. 00:38:00,770 --> 00:38:01,680 [John] What... Okay, what do- 00:38:01,680 --> 00:38:01,850 [Eric] But- 00:38:01,850 --> 00:38:05,980 [John] W- what... This is a better question. When you get stuff in from people- 00:38:05,980 --> 00:38:06,680 [Eric] Mm-hmm 00:38:06,680 --> 00:38:14,820 [John] ... what rule are you editing for the most, do you think? Where you're like, "If I could just train people on this one rule, I have to edit for this one all the time when people send me stuff." 00:38:18,880 --> 00:38:21,540 [Eric] I'll speak to that on 2 levels. 00:38:22,620 --> 00:38:27,640 [Eric] Pr- a- and I, I'd, I'd need to think about... Th- but let's just say these are probably both 00:38:29,200 --> 00:38:30,840 [Eric] equal in quantity. 00:38:30,840 --> 00:38:31,419 [John] Okay. 00:38:31,420 --> 00:38:34,699 [Eric] Um, but one of them is the, is the deeper problem. 00:38:34,699 --> 00:38:35,900 [John] The... Yes. 00:38:35,900 --> 00:38:36,230 [Eric] And 00:38:38,160 --> 00:38:40,410 [Eric] this is going [laughs] to sound so simple, but 00:38:43,660 --> 00:38:52,700 [Eric] I don't even... I don't know how many times per week, but I say all the time, "What are we actually trying to say here?" 00:38:52,700 --> 00:38:53,420 [John] Yeah. Yeah. 00:38:53,420 --> 00:38:59,620 [Eric] And one thing that I was gonna point out with both Orwell's rules, but especially The Elements of Style, 00:39:01,120 --> 00:39:08,400 [Eric] is the, the... they are inherently related. 00:39:08,400 --> 00:39:09,180 [John] Right. 00:39:09,180 --> 00:39:15,540 [Eric] So if you think about th- if... Let's just combine a couple of these as a- 00:39:15,540 --> 00:39:15,700 [John] Right 00:39:15,700 --> 00:39:17,560 [Eric] ... as a very rudimentary- 00:39:17,560 --> 00:39:18,120 [John] Right 00:39:18,120 --> 00:39:25,050 [Eric] ... like, everyday thing that a writer's, you know, everyday task that a writer's gonna perform. Actually, I just committed one of the sins that I'll talk about next, which is saying- 00:39:25,050 --> 00:39:25,050 [John] Okay 00:39:25,050 --> 00:39:26,080 [Eric] ... "everyday thing," right? 00:39:26,080 --> 00:39:26,760 [John] Okay. 00:39:26,760 --> 00:39:29,680 [Eric] Everyday writing task that a writer- 00:39:29,680 --> 00:39:29,690 [John] Mm-hmm 00:39:29,690 --> 00:39:35,380 [Eric] ... is going to perform. Okay. The paragraph as a unit of composition. 00:39:35,380 --> 00:39:35,640 [John] Okay. 00:39:35,640 --> 00:39:40,900 [Eric] 100%, right? What's interesting about that is you have multiple paragraphs in a piece, and those form- 00:39:40,900 --> 00:39:40,920 [John] Mm-hmm 00:39:40,920 --> 00:39:42,140 [Eric] ... the narrative arc. And so this goes- 00:39:42,140 --> 00:39:42,330 [John] Yeah 00:39:42,330 --> 00:39:49,429 [Eric] ... back to what I was saying earlier in that then my job, a big part of my job is stepping back and looking at the composition- 00:39:49,429 --> 00:39:49,440 [John] Right 00:39:49,440 --> 00:39:55,340 [Eric] ... of those units of composition, right? [smacks lips] But within a, within an individual unit of composition, 00:39:56,700 --> 00:40:04,530 [Eric] the... We'll, we'll just kind of... Let's try to run the gamut here really quickly. So 1st of all, using simple words 00:40:06,480 --> 00:40:07,620 [Eric] wherever you can. 00:40:07,620 --> 00:40:07,740 [John] Yeah. 00:40:07,740 --> 00:40:10,440 [Eric] Right? Plain English, understandable words, right? 00:40:10,440 --> 00:40:11,060 [John] Right. 00:40:11,060 --> 00:40:18,200 [Eric] Not using an unnecessarily simile or metaphor to try to get fancy when you are trying to explain something. 00:40:18,200 --> 00:40:18,520 [John] Mm-hmm. 00:40:18,520 --> 00:40:18,780 [Eric] Right? 00:40:19,940 --> 00:40:20,260 [Eric] Um, 00:40:22,880 --> 00:40:24,740 [Eric] keeping, um, 00:40:27,520 --> 00:40:29,820 [Eric] keeping 2 units close together- 00:40:29,820 --> 00:40:30,010 [John] Yeah 00:40:30,010 --> 00:40:31,080 [Eric] ... as opposed to spreading them out, right? Like- 00:40:31,080 --> 00:40:34,400 [John] I feel like that's one that I do and that I see, like- 00:40:34,400 --> 00:40:34,870 [Eric] Absolutely 00:40:34,870 --> 00:40:35,060 [John] ... done. 00:40:36,100 --> 00:40:52,590 [Eric] Right? So, and then s- uh, saving sort of the, the highest, uh, le- leaving a lasting impression by saving, like, the highest impact word or phrase for the end of a paragraph, right? So really what makes writing difficult is that you're balancing all of these different things- 00:40:52,590 --> 00:40:52,600 [John] Right 00:40:52,600 --> 00:40:57,019 [Eric] ... to try to create a great paragraph, right? And 00:40:58,220 --> 00:41:03,800 [Eric] when you make mistakes on multiple fronts, it... what happens is you lose clarity. 00:41:03,800 --> 00:41:04,000 [John] Right. 00:41:04,000 --> 00:41:21,400 [Eric] Um, and that shows up in, in, in multiple ways, right? But you use a metaphor, and it's like, "Okay, like, I kinda get what you're saying there." And then there's, like, a word that's too fancy, and then there's a lack of clarity on, you know, say, like, 00:41:22,920 --> 00:41:24,020 [Eric] very commonly, like, 00:41:25,100 --> 00:41:25,420 [Eric] uh, 00:41:27,440 --> 00:41:30,120 [Eric] not defining terms- 00:41:30,120 --> 00:41:30,190 [John] Hmm 00:41:30,190 --> 00:41:30,960 [Eric] ... that you're using. 00:41:30,960 --> 00:41:32,080 [John] Right. 00:41:32,080 --> 00:41:41,380 [Eric] You know, or using ambiguous terms, right? And so when multiple of... when multiple rules are broken and there's sort of a collision there, the ultimate result is a lack of clarity. 00:41:41,380 --> 00:41:42,160 [John] Yeah. Right. 00:41:42,160 --> 00:41:49,510 [Eric] And so I... it w- it is so common for me to say, "L- just tell me like we were talking in person." 00:41:49,510 --> 00:41:49,600 [John] Mm-hmm. 00:41:49,600 --> 00:41:53,100 [Eric] Like, "What are you really... What is the main- 00:41:53,100 --> 00:41:53,740 [John] Right 00:41:53,740 --> 00:41:57,300 [Eric] ... point that you're trying to make? And say it as succinctly as possible." 00:41:57,300 --> 00:41:57,920 [John] Right. 00:41:57,920 --> 00:42:04,800 [Eric] And oftentimes, like, if people say that e- even on a call, it's like, that is what... Just write that, you know? 00:42:04,800 --> 00:42:05,020 [John] Yeah, yeah. 00:42:05,020 --> 00:42:05,520 [Eric] Just- 00:42:05,520 --> 00:42:05,780 [John] Right 00:42:05,780 --> 00:42:07,850 [Eric] ... write that, you know? That's, that's what you need to do. 00:42:07,850 --> 00:42:15,759 [John] So you're... So it seems like if you had to pick, like, one thing in summary, you are trying to focus and clarify the argument? 00:42:15,760 --> 00:42:15,990 [Eric] Yes. 00:42:15,990 --> 00:42:28,360 [John] Like, what are we trying to say? What's the argument? And then there's a lot of work on the arc, to use another A word, as far as the flow and the paragraphs and the, you know- 00:42:28,360 --> 00:42:28,540 [Eric] Yep 00:42:28,540 --> 00:42:36,068 [John] ... like, those 2 things feel like 2 of the hardest things, hardest problems that- ... that you guys are working on, on a, on a regular basis. Is that fair? 00:42:36,068 --> 00:42:41,198 [Eric] That is, that is the hardest problem, for sure. Because, a- and I'll, I'll, I'll say this. 00:42:42,748 --> 00:42:51,388 [Eric] If your argument is sound and you have a lot of conviction around, uh, a- and around the narrative arc, conviction doesn't mean it's good, but like- 00:42:51,388 --> 00:42:51,568 [John] Right 00:42:51,568 --> 00:42:53,228 [Eric] ... if, if you're a good writer and you've- 00:42:53,228 --> 00:42:53,608 [John] Mm-hmm 00:42:53,608 --> 00:42:56,308 [Eric] ... done the work of developing conviction around the narrative arc- 00:42:56,308 --> 00:42:56,528 [John] Mm-hmm 00:42:56,528 --> 00:42:56,528 [Eric] ... 00:42:58,268 --> 00:43:01,508 [Eric] it, AI can generate a pretty good draft. 00:43:01,508 --> 00:43:01,528 [John] Right. 00:43:01,528 --> 00:43:03,128 [Eric] But you've already done the hard work. 00:43:03,128 --> 00:43:03,508 [John] Right. 00:43:03,508 --> 00:43:06,628 [Eric] Right? And then you edit, and you need to be a good editor and follow- 00:43:06,628 --> 00:43:06,798 [John] Yeah 00:43:06,798 --> 00:43:07,468 [Eric] ... Orwell's rules and- 00:43:07,468 --> 00:43:07,808 [John] Sure 00:43:07,808 --> 00:43:08,368 [Eric] ... you know, um- 00:43:08,368 --> 00:43:11,708 [John] But that's wh- but AI is more- ... helpful right now with that process. 00:43:11,708 --> 00:43:21,988 [Eric] Yes, exactly. Exactly. The, and then one very specific thing that we run across all the time- ... all the time, all the time. And I mean, I'm guilty, everyone's guilty of this, right? But, [smacks lips] uh, 00:43:24,268 --> 00:43:29,778 [Eric] not stating things explicitly. So, uh, n- not, 00:43:31,308 --> 00:43:34,678 [Eric] that in itself is an ambiguous phrase, so, you know, the cobbler's shoes here. But- 00:43:34,678 --> 00:43:35,218 [John] [laughs] 00:43:35,218 --> 00:43:38,958 [Eric] ... uh, when I ed- I do a lot of editing, especially now that the team is larger. 00:43:40,398 --> 00:43:46,808 [Eric] [smacks lips] But there'll be a phrase of, you know, or there'll be a paragraph that has a couple of sentences, and so you'll say, um, 00:43:48,968 --> 00:43:49,228 [Eric] you know, 00:43:50,448 --> 00:43:56,948 [Eric] "Team used product XYZ to solve the, to solve, you know, problem ABC-" 00:43:56,948 --> 00:43:57,068 [John] Right 00:43:59,108 --> 00:43:59,528 [Eric] ... uh, 00:44:00,808 --> 00:44:02,368 [Eric] for whatever reason. 00:44:02,368 --> 00:44:03,408 [John] Right. 00:44:03,408 --> 00:44:04,508 [Eric] And then the very next sentence, 00:44:05,588 --> 00:44:05,848 [Eric] uh, 00:44:06,888 --> 00:44:09,988 [Eric] let's say starts with, "What this unlocked for them." 00:44:09,988 --> 00:44:10,447 [John] Mm-hmm. 00:44:10,448 --> 00:44:11,188 [Eric] Right? And it's like- 00:44:11,188 --> 00:44:11,228 [John] Yeah 00:44:11,228 --> 00:44:12,368 [Eric] ... "Well, hold on." Like- 00:44:12,368 --> 00:44:12,508 [John] Right 00:44:12,508 --> 00:44:15,508 [Eric] ... "What do you mean by this?" Because- 00:44:15,508 --> 00:44:15,598 [John] Right 00:44:15,598 --> 00:44:18,818 [Eric] ... we loaded a bunch of stuff into the 1st sentence. 00:44:18,818 --> 00:44:18,848 [John] Right. Right. 00:44:18,848 --> 00:44:22,478 [Eric] We talked about a product they used. We talked about their reasons for using it. 00:44:22,478 --> 00:44:22,528 [John] Mm-hmm. 00:44:22,528 --> 00:44:24,748 [Eric] We talked about the pain points that they had- 00:44:24,748 --> 00:44:24,798 [John] Mm-hmm 00:44:24,798 --> 00:44:30,368 [Eric] ... whatever. Like, we loaded up that 1st sentence, and then so you start on the next one and you say, "This." And it's like- 00:44:30,368 --> 00:44:30,378 [John] Mm-hmm 00:44:30,378 --> 00:44:30,568 [Eric] ... "Well, 00:44:31,928 --> 00:44:35,368 [Eric] what specifically do you mean by this?" 00:44:35,368 --> 00:44:35,968 [John] Right. 00:44:35,968 --> 00:44:38,888 [Eric] And it's actually surprisingly hard for people to nail that down- 00:44:38,888 --> 00:44:38,898 [John] Sure 00:44:38,898 --> 00:44:44,068 [Eric] ... because it's just a catchall phrase that makes it easy to not write with- 00:44:44,068 --> 00:44:44,208 [John] Yeah 00:44:44,208 --> 00:44:45,068 [Eric] ... extreme clarity. 00:44:45,068 --> 00:44:46,028 [John] Yeah, yeah. Yeah. 00:44:46,028 --> 00:45:07,877 [Eric] And that, I would say, is the most common very specific thing that I, like, edit for and try to catch, is lack of being specific, because it is just way easier to pa- to just load up a, a generic word like this with a bunch of ambiguous meaning. And aga- a lot of times is that, 00:45:08,888 --> 00:45:16,468 [Eric] are you going to lose the entire meaning of the paragraph because of that? Probably not. Is that excellent writing- 00:45:16,468 --> 00:45:16,568 [John] Right 00:45:16,568 --> 00:45:20,877 [Eric] ... and does it meet our standard? It absolutely doesn't at Vercel. And so that's sort of- 00:45:20,877 --> 00:45:20,877 [John] Right 00:45:20,877 --> 00:45:24,648 [Eric] ... the, you know, human element that we, you know, 00:45:25,668 --> 00:45:27,508 [Eric] that we apply to everything that we review. 00:45:27,508 --> 00:45:27,648 [John] Yep. 00:45:28,668 --> 00:45:35,488 [John] Awesome. This has been great. I look forward to you running this episode through AI and getting your new set of rules for writing. 00:45:35,488 --> 00:45:37,778 [Eric] [laughs] We should do an episode on h- an episode on how I do that- 00:45:37,778 --> 00:45:37,808 [John] [laughs] 00:45:37,808 --> 00:45:39,567 [Eric] ... 'cause I use AI heavily for, for . 00:45:39,568 --> 00:45:49,597 [John] No, but, but in, but in all seriousness, like, you know, 45 minutes of content on describing this, like, is really good s- starting point on- 00:45:49,597 --> 00:45:50,088 [Eric] [laughs] It is 00:45:50,088 --> 00:45:52,148 [John] ... creating, you know, skills for writing. 00:45:52,148 --> 00:45:52,248 [Eric] Yeah. 00:45:52,248 --> 00:45:53,568 [John] Like, so there you go. 00:45:53,568 --> 00:46:05,428 [Eric] Absolutely. All right. Thanks for joining. We'll catch you on the next one. [instrumental music]