Speaker 1If you use AI to write an op-ed, does that devalue your thoughts? Believe it or not, that's been maybe the biggest question the AI community has been discussing for the last couple of days. When one of finance's biggest voices dropped an op-ed in the Wall Street Journal critiquing the Treasury Secretary and current monetary policy, he probably didn't think that the big discourse was going to be about how he wrote the op-ed. And yet, that's what it was, because the piece you see was very, very clearly written by AI. People reacted like it was some big scandal. Except then both the famous financier and the opinion editor at the Wall Street Journal said, yeah, of course he used AI. So if we now live in a world where AI writing is totally normal, how can we make it actually good? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Blitzy, Agent, and HyperAgent. To get an ad-free version of the show, go to patreon.com.ai. And to learn more about sponsoring the show, or really to find out anything else about the show, go to ai.ai. There's sponsor information there. There's information about education programs we're doing. And of course, a full breakdown of every episode that will make it much easier to reference and share the specific parts that are most interesting to you. Again, you can find it all on ai.ai. Believe it or not, one of the biggest conversations around AI for the last couple of days has been around AI writing, specifically the legitimacy of it, disclosures around it, and how all of those things might be changing. The specific context for this was an op-ed published in the Wall Street Journal by investor Stanley Druckenmiller. The op-ed, called Let the Bond Market Speak, was a fairly full-throated critique of Treasury Secretary Scott Besant. Now, I'm not going to get into the substance of Druckenmiller's specific argument, as that's a totally different show than this one. But the things that are worth noting are, one, that Druckenmiller has a legendary status among finance folks, among other achievements, running a hedge fund that had zero losing years between 1981 and 2010, which is completely unprecedented before or afterwards. And two, that the focus of this piece is not some secondary or tertiary topic, but cuts to the quick of macroeconomic policy right now. In other words, the op-ed, from a substance perspective at least, matters. And yet, extremely quickly, the substance was not what people... were talking about. Instead, it was the fact that the piece is replete with just about every egregious AI-ism that exists today. It is, for example, full of it's not this, it's that. If the 30-year must trade at 5.5% to clear, that isn't a crisis, it's an invoice. This wasn't liquidity management, it was price management, and a mistake far larger than 4 billion suggests. AI detectors like Pangram quickly identified the chance that it was AI-generated at 100%, and some folks were not happy. BionX writes, I don't understand how people sniff this out as AI. I mean, apart from running it through Pangram. Why would Druckenmiller put his name on AI slop? He's fine with doing that, even though these allegedly aren't his words? Neil Sybart writes, Why did the Wall Street Journal not disclose that Druckenmiller used AI to write his piece? Such disclosures should be mandatory. Widely followed markets commentator Jesse Livermore writes, Presenting AI slop as if it were your own writing is plagiarism, full stop. You're taking credit for words and phrases you did not write. So Druckenmiller, having been caught out, admitted sheepishly that yes, indeed, he had used AI, and apologized and said he'd do better next time, right? Wrong. His response instead was, bro, of course I used AI. In addition to the literal line, of course I used AI, Druckenmiller added, there's a reason I moved from an English major to being an economics major. I'm not embarrassed by it. I write everything using AI now for the same reason I use a calculator when I do math problems. Wall Street opinion editor Paul Jai Goh backed Druckenmiller up, with Jai Goh saying, AI is a fact of modern life. People will use it to assist their work and their writing, including with research, checking grammar, editing, and more. The question for us is whether what we publish from contributors reflects an author's original argument, and if the author has the standing and credibility to make it. In Stan Druckenmiller's case, we have had a relationship with him for many years, and no one can doubt that his op-ed is his genuine opinion. So clear as day here, the journal's op-ed position is that what matters is the sincerity of the ideas, not the unique human flavor of the words used to express them. And from there, the conversation quickly moved, and while some of the critiquers might have been pretty loud, there were also a lot of folks pointing out this wasn't all that different from how leading figures had done op-eds for quite some time. Bloomberg's Joe Weisenthal wrote, Prior to AI, there were plenty of high-profile people publishing op-eds under their own name that were entirely written by some underling, and I don't think there was much controversy about it. Validating that, journalist Sharon Goldman writes, From 2015 to 2020, I wrote dozens of op-eds for top executives that ran in tech or trade publications. It was common practice and I never received or wanted a byline-slash-credit. It was a decent gig for a freelance writer who had to do lots of types of work to pay the bills. I usually had one conversation with the executive as well as some notes. Others picked up on Druckenmiller's argument that effectively this was just a tool. Reflecting on Druckenmiller saying that of course he used AI, investor Scott Phillips wrote, Good. Can you imagine having to apologize for using a typewriter or a computer or the internet for research rather than encyclopedias? AI is here to stay. We should be asking why people aren't using it, not criticizing them for doing so. When All In's Jason Calacanis wrote, If you write a public piece with AI, you should disclose that in the first sentence. AI wrote this for me. Unforgivable for a public figure to publish an AI-written piece without a clear disclosure in the first sentence. Undermines the entire premise of publishing a thought piece because we can't tell what's yours and what's the magic black box's thoughts. However, his co-host Chamath Palihapitiya bit back. This is a dumb new form of virtue signaling. Do you disclose every article you've ever read that gives you an opinion when you spout off an X? No. This is how AI also might as well disclaim the entire internet in every written word. If he puts his name behind it, it's his opinion. Separately, Chamath also posted, If Stan Druckenmiller isn't embarrassed to be a meat proxy, you shouldn't be either. It's like after Matches was invented, still celebrating the arduous time to rub two sticks together. Dumb. Even simpler, investor Andrew Steinwald wrote, People getting mad at others for writing with AI is boomer-coded. In like two years, no one will care if you use AI or not. Pointing out where value is shifting, Aria Denise writes, Who cares if Druckenmiller used AI to help his thoughts? AI is a tool. If it helps an investor take decades of experience, judgment, and pattern recognition and articulate those thoughts more clearly, that's a good thing. The value is in thinking. Get used to it. But for others, this was a much more nuanced conversation. Spectrum Markets' Brent Donnelly wrote, The AI writing debate comes down to one question. Is writing art or is it a functional communication tool like code? This is like asking if a car is to get from A to B or for excitement in good times. It's just a personal thing. To me, writing, even nonfiction, is art. And cars are for fun. But many don't agree. There's no right answer. David McMahon wrote, There's a legitimate debate here and Druckenmiller is a smart guy. I have no doubt the AI written editorial reflected his actual thinking. But too many people are using AI as a substitute for serious thinking. Just look at the epidemic of reply guys on this site flooding posts with AI-generated garbage. That type of use deserves nothing but mockery. Going deeper on the idea that part of the problem was not AI writing in general, but AI writing for op-eds, Dirty Texas Hedge posted, What Druck and Jai Go say is 100% true descriptively. The problem is not that they're wrong. The problem is that it represents an expression of contempt of the format of an op-ed and of its audience. And others started to reach for whether there was some variance here. That AI writing might be good for some things but not for others. CNBC's Deirdre Bosa wrote, I think about this often but can't remember who said it. AI is great for the middle. Beginning needs a human. What's the idea and what are you trying to say or build? Middle, let AI research, organize, draft, rewrite, tighten, poke holes in it. N needs to be human too. Is this any good? True? Do I buy it? I think that Druck and Miller didn't do the end part as well as he could have. Now for completeness, the thought that she was referencing came from Balaji Srinivasan, who back in June of 2025 posted, AI doesn't do end to end. It does middle to middle. The new bottlenecks are prompting and verifying. So with all of this in mind, and as someone who's thought about this quite a bit, I wanted to put together kind of a crib sheet on how I think about AI writing. For the sake of a catchy podcast title, it is called Five Rules for AI Writing. And in addition to those maxims, I also want to share about how I think about writing in a few different specific contexts. So the foundation for all of this is that I tend to agree that AI writing is now a fact of life. If you need any evidence of this, just go look at OpenAI's recently released enterprise data about how users are using ChatGPT. The preponderance of it across almost every department is writing. And that's not just for comms, that's for recruiting, marketing, customer support, legal, sales, policy, you name it. AI writing is here and it's here to stay. But that doesn't mean that all AI writing is the same. In fact, rule number one for me is that different types of writing mean different types of rules. An email is not the same as a strategy memo. A strategy memo is not the same as a post on LinkedIn. A post on LinkedIn is not the same as an op-ed. These different types of writing are trying to achieve different things, and the way that we engage with AI around those writings will consequently be different. My next maxim of AI writing is that I also agree that this purity test sort of phase will go the way of the dodo. I think the folks who are saying that this conversation is going to be quaint in a couple of years, are frankly probably right. But if the purity test will go, I think the quality test won't. In fact, I think if anything, the increased easiness of the inputs of writing will increase the burden on the outputs of writing even more. I don't think that just because people start to accept that people are using AI to write means that they'll accept bad writing as a consequence. To get specific on that is maxim three. For most people, quality is corresponding with effort, or the perception of effort. And effort or its absence tends to be pretty obvious I think a lot of the negative response to Druckenmiller's op-ed was that the AI-isms were so glaring and so fixable that it seemed like he didn't even care to take the time to change the syntax on a couple of sentences. And if he didn't care to take that time, do we really think that he put all that much thought into the point that he's trying to make? In that case, I think we have an example of people's perception of lack of effort on the writing itself being more broadly expressing of a lack of effort in the construction of the argument underneath. A new study from KPMG and the University of Texas at Austin found that when people work with AI, similar skills don't guarantee similar outcomes. Researchers studied more than 500 early career professionals and found that the best performers consistently amplified the value of AI by guiding, evaluating, and refining its outputs. These top performers, called AI amplifiers, weren't defined by what they knew alone, but by how they worked with AI. Learn more about what separates AI amplifiers from everyone else at kpmg.com slash us slash AI amplifiers. Here's why most legacy modernization projects fail. The AI doing the work can't understand code bases at scale. It sees a small slice of context, examines syntax, and misses years of decisions distributed across the global application ecosystem. Blitzy solves this the way it solves everything. Grounded in your code before any other machine, Blitzy solves it the way it solves everything. Grounded in your code before any other machine, Blitzy solves it the way it solves everything. Grounded in your code before any other machine, Blitzy solves it the way it solves everything. Grounded in your code before any other machine, Blitzy solves it the way it solves everything. Grounded in your code before any other machine, Blitzy solves it the way it solves everything. Grounded in your code before any other machine, Blitzy solves it the way it solves everything. Grounded in your code before any other machine, Blitzy solves it the way it solves everything. Grounded in your code before any other machine, Blitzy solves everything. Grounded in your code before any other machine, Blitzy solves everything. Grounded in your code before any other machine, Blitzy solves everything. Grounded in your code before any other machine, Blitzy solves everything. Grounded in your code before any other machine, Blitzy solves everything. Grounded in your code before any other machine, Blitzy solves everything. Grounded in your code before any other machine, Blitzy solves everything. Grown into a persistent knowledge graph. Every dependency, every constraint, every piece of tribal knowledge that used to live in one engineer's head. From that understanding, Blitzy autonomously executes language migrations, framework upgrades, and monolith-to-microservices transformations, all validated end-to-end. One Blitzy customer modernized a $10 million monolithic insurance stack in 16 weeks against a 137-week baseline with coding agents. That's 9x compression. Retire technical debt while accelerating your roadmap. See how at Blitzy.com. That's B-L-I-T-Z-Y dot com. Here's a harsh truth. Your company is probably spending thousands or millions of dollars on AI tools that are being massively underutilized. Half of companies have AI tools, but only 12% use them for business value. Most employees are still using AI to summarize meeting notes. If you're the one responsible for AI adoption at your company, you need Section. Section is a platform that helps you manage AI transformation across your entire organization. It coaches employees on real use cases, tracks who's using AI for business impact, and shows you exactly where AI is and isn't creating value. The result? You go from rolling out tools to driving measurable AI value. Your employees move from meeting summaries to solving actual business problems. And you can prove the ROI. Stop guessing if your AI investment is working. Check out Section at SectionAI.com. That's S-E-C-T-I-O-N-A-I dot com. This episode of the AI Daily Brief is brought to you by HyperAgent, where you run fleets of agents your team can manage together. New users get $1,000 in inference. Forget local agents and chat workflows waiting on your laptop to be prompted. HyperAgent deploys always-on agents in the cloud, doing real work across the tools your team already uses. Marketing's agent turns competitor moves into landing pages. Sales' agent enriches leads, drafts emails, and updates the CRM. Ops' agent chases the paperwork and tracks the budget. Every agent has access to shared context and follows your rules about scope and approvals. It's time you add agents that feel like teammates. Hire yours at HyperAgent, built by the team at Airtable. Claim your $1,000 in inference at HyperAgent.com. AI Daily Brief Maxim 4, and I think this one will be most acutely felt among enterprise listeners. When it comes to AI writing, longer is not better, and in fact, usually the opposite is true. The reason we have phrases like work slop now is that a lot of the AI models that we've used over the last couple of years have been much better at saying a lot than saying the right thing succinctly. And what's funny is that although this problem has been greatly exacerbated by the rise of AI writing, great writers have always known this to be true. There's a famous quote from Blaise Pascal in the 17th century that has been cribbed and updated by many authors since, where in a letter he wrote, I've made this longer than usual because I have not had time to make it shorter. The first wave of AI writing was defined by endlessly long, eye-bleeding length documents. The next generation will not be. And lastly, Maxim 5. As so many people have pointed out, in so many cases, writing is thinking. The process of deciding what your thesis is, is a process of thinking. It's a process of thinking. It's a process of what the supporting arguments are, narrative construction, even considering the impact of a turn of phrase. These are exercises whose main point is ultimately not the words that end up on the page, but the substance of the argument that runs underneath them. That does not mean a priori that using AI for writing means outsourcing your thinking, but it is the constant risk that needs to hover in your field of view when you are using AI writing because it is so easy to fall into. And I say this with absolutely no judgment. One of the places that I feel this tension is when I'm ideating on a new concept for an episode, like this one. I could have very easily told Claude or ChatGPT, I want to do an episode called Five Rules of AI Writing, and I know a couple of the rules, like writing is thinking, but you take and run with that idea and build me the asset that I can talk over in the show. That would have been a much faster process than what I did, which is sitting in Notion and actually thinking through each of these five, debating with myself around whether I wanted it to be six or whether I wanted to consider that first one, just the foundations. But for this particular context, I'm going to The more manual approach was the right one, which gets us to a crib sheet for the burden on different types of writing. Remember, maximum one is different types of writing, different types of rules. And what that means is can or should AI write this is probably in most cases the wrong question. So let's talk about a few different examples, starting with some of the most obvious, like emails. Emails are pretty clearly safe for AI in that for the vast, vast majority of emails, no one has ever judged them on the quality of the prose. And while yes, there could be the risk of outsourcing your thinking, with email writing, emails tend to focus on such small, concise units of thought that the risk that any of our inherent tendency towards intellectual outsourcing becomes prominent gets reduced. That said, I think in general, we're moving to a world where using AI for emails might be in many cases overkill. Now, maybe there are some types where it makes sense. For example, we have a standard response email for sponsorship requests that come through, but for interactive emails, the type of things that you might otherwise be doing via Slack or a messaging app, I tend to think that other ways of speeding up your process, like just dictating with something like WhisperFlow, are going to be even more effective than using AI. Next up, let's talk about meeting note summary, another extraordinarily common use of AI writing. This, once again, is very safe for AI, because in many ways, this isn't even really writing. It's about compression. It's about taking a bunch of other words and turning them into something succinct. Still, there is an AI failure mode even with meeting notes, which is the AI being too exhaustive. An AI that was sitting in a meeting and listened for 60 minutes and has the transcript of everything that everyone said, and then they're like, oh, I don't know what to do with this. It's probably going to be very good at summarizing what everyone said, but might not be as good as you as explaining the most important things that anyone said. And so I think even if you can outsource the vast majority of meeting note summarization and things like that to AI, it's worth coming in over the top with your clarification of the one or two things that are actually important as a complement to what it produces. What about an internal strategy memo? This, I think, can be sneakily bad for AI, although certainly you can use AI to assist. On the one hand, it seems like AI is a good way to help you out, but on the other hand, this would be a great area for AI because it's another one where the strength of the pros doesn't really matter that much. That said, this is a great example of where that fifth maxim that writing is thinking comes in. If an AI is tasked with writing some internal strategy memo, unless you are hyper-precise about exactly the strategy that it is communicating and the things that it is meant not to communicate, it is very, very easy for AI to go off the rails and add its own additions or simply to provide bad thinking in the form of generic advice that is based on its training data rather than the specific context of your actual organization. What that means is that internal strategy type writing is a great example of, I think, the middle that Deirdre was talking about. You're going to want to, on a personal level, exhaustively think through the strategy. And you might even want to think about your interaction with AI on two totally different levels. The first is using AI to support that thinking, bulleting, outlining, iteration, and then the final writing as an entirely separate process. I think if the outline is tight and contains all of that thinking effort, it will often be totally fine to use AI to support that let AI write the actual words. Now, what about a very common use case in social media copy? This one is interestingly mixed. On the one hand, I think that AI does a fine job on a lot of social copy, especially if you have tuned it away from the most egregious AI-isms like it's not this, it's that, the self-congratulation, and other things that are just going to be glaring and distract people from whatever it is that you're trying to say. I think in general, it does better the shorter the medium is. In other words, AI is probably going to have an easier time with an old single-line post on X than it is with a full LinkedIn essay, because that full LinkedIn essay just has more room for its AI-isms to shine through. The bigger problem with AI writing for social media is that to the extent that it ever did, social media no longer simply rewards pithy writing. In fact, social media demands engagement. It wants posters not just automating their posting to provide more content into the MAW, it wants those same posters and commenting on other people's posts, and generally providing engagement, the sweet, sweet human signal that allows those platforms to serve ads, because you spend more time on the platform. I think the way to do AI writing better for social media, then, is two parts. First, the more time you're willing to take to invest up front to tune the voice, the better your results are going to be, even with the AI-written part. The second part is simply to appreciate that AI writing, no matter how tuned it is, is only going to get you so far with social media. And this is one where I feel it acutely, as I've now built a full pipeline from the disaggregation of each episode that ends up on the AIdailybrief.ai website, to a full pipeline of episodes that ends up on the AIdailybrief.ai website, to a set of generated posts that go to Twitter and LinkedIn. It has been valuable, it's increased sharing, it's done a lot of the things that I wanted it to do, for social, especially considering I was doing exactly zero social before, but within weeks, so clearly slammed up against the walls of how much it can do for the AI Daily Brief on social media. So in some ways, the lesson of AI writing here is to just have realistic expectations. Two more to focus on before we get out of here. First, marketing copy, and then we'll end with op-eds, given that that's where the conversation started. I have found AI writing with marketing copy to be annoyingly bad. I say annoyingly bad because this is one of the areas where I want it to be good. Even as someone who obsesses over the words on the websites that describe my projects, I want it to be good enough to allow me to not do that sort of obsessing. And yet, marketing is the area where I find the most egregious AI-isms turn up. Maybe it's because marketing is an area where you want uniqueness and distinction, not sameness and commonality. And so the cost of those common AI-isms is even higher than in other contexts. I also find there are some specifically bad things that AI does when it comes to marketing copy, even as you are interacting with it. The one that I run across most often is when I give the model a correction. For example, something like, stop presuming so much about the reader. It tends to turn that correction itself into copy. So I'll look at the next version and find some headline somewhere that says, we assume nothing about the reader. It drives me absolutely batty, and it happens all the time. So when it comes to how to make AI writing better for marketing copy, my short answer, unfortunately, is that I genuinely haven't been able to. How I would try, however, if I had to, is that I would try to make AI writing better for marketing copy. One way to really dig down on this would be to extensively provide as many examples of past on-brand writing, as well as as many examples as possible of what I like and don't like for my brand specifically to the AI to try to tune it at least a little bit better. Still, I think for me right now, a better use of AI for marketing copy is as a rapid iterator. For example, if I'm trying to figure out a tagline that I really like, asking AI to come up with 20 is a really great way to spark something for myself. Lastly, on op-eds and convincing essays. Remember what Texas Hedge wrote about the Druckenmiller piece. He said, the problem is that it represents an expression of contempt of the format of an op-ed and of its audience. And this, I think, gets to the maxims about effort. The whole point of an op-ed is to convince someone of something. When readers can more or less instantly tell it was low effort, they inherently switch off. If you couldn't be bothered to take the time to make your argument convincingly, why should I care about your argument? As I said, no one actually cared, I don't think, about the quality of the writing in Druckenmiller's op-ed. The issue was that the perceived laziness of it, in their estimation, reduced the strength of the underlying argument. And this is perhaps the easiest example of thinking is writing. You just gotta do the thinking first. When you're trying to convince someone of something, you have to get extremely clear on both the thesis and the supporting points before you turn it over to AI. This is the five-paragraph essay from grade school, baby. You just gotta do it. And to complete the effort sandwich, after you have turned it over to the AI, go through and make sure it doesn't have obvious AI-isms. To sum up, AI writing is not going anywhere. And I think it will unlock a lot of opportunity. People who haven't used writing for communication before will start to. What it won't change is that laziness in work comes across as laziness in work. There are no shortcuts ultimately for doing things well, even if we can make great use of new tools that help us do that well faster and better. For now, that's going to do it for today's AI Daily Brief. Appreciate you listening or watching, as always. And until next time, peace! 🎵outro music plays🎵
PodcastTranscript AI