220: Alex Halliday: How to build content engineering systems that get cited and scale without slop

What's your take on how content engineering helps teams that don't want to just build more content at scale, but instead want to write things that can't be replicated from public data?

When we think about the construction of like a content system or the content engineer, a lot of that context is collected and gathered automatically. But then because on a call transcript, someone has started asking about a new regulation or a competitor feature, who internally is the right human to then go and get an updated response or to, or updated positioning on that? And so the goal of the engineer is not to remove humans out of the loop. It's actually to find more humans to put into the loop, but, but have them practice at the top of their license. So we spend a lot of time thinking about what unique context exists inside the business and what humans that we can leverage to get more knowledge out into the world. And that is a systems problem. That's a human problem. That's a strategy and taste problem.

What's up everyone? Today we have the pleasure of sitting down with Alex Halliday, founder and CEO of AirOps. In this conversation, we cover what led to AirOps pivoting to AI content engineering, what a content engineer does that a senior marketer doesn't, and what a marketing team looks like in 3 years. We'll also talk about what it actually takes to get AI content past a human editor and why enterprise CMS integration is harder than the content quality problem. To make the most of this conversation, joining me as co-host today is the mad marketing scientist, John Taylor. We'll dive in after a quick word from 2fer Awesome partners. This episode is brought to you by Atropishing. Everyone says multi-touch attribution is dead, but it's not dead. It's just been done wrong. Most tools are built for another era or by people who don't actually run growth. They lean on Google Analytics or whatever the ad platforms tell you. Both are wrong. That's why your CAC looks off and your spend doesn't line up with revenue. The team behind Attribution built growth data frameworks for companies like Dropbox, Replit, and Cargurus, feeding into the only MTA platform that delivers auditable revenue-grade accuracy across your entire funnel. It's trusted by over 1,000 B2B SaaS and e-commerce teams, and it's one of Twilio Segment's only two preferred partners for marketing auditability. Attribution connects to CRM, ad platforms, data warehouse, and e-commerce stack. With Attribution, you can pinpoint TAC by channel, audit funnels across every source, and tie every ad campaign and piece of content to real ROI and retention. That's auditable growth from first visit to repeat purchase. Visit attributionapp.com/humans to finally see what's really driving your growth. This episode is brought to you by Growthloop. Despite the seemingly infinite number of martech options, marketers today are still chasing that dream of one-to-one personalization. With the Growthloop Agentic Composable CDP, that dream is finally a reality. Built with a composable warehouse-native architecture, Growthloop replaces engineering tickets and SQL with no-code self-serve audience building and journey orchestration. Marketers get access to real-time customer data to build personalized campaigns quickly And technical teams can rest easy knowing that customer data never leaves the cloud. And for marketing teams who are eager to move from guessing to knowing which actions drive growth, Growthloop's composable AI decisioning continuously learns which marketing actions change customer outcomes and optimizes campaigns in real time. Teams can finally close the loop between action, measurement, and optimization, and marketers are empowered to finally build the dream of one-to-one personalization. Learn more at go.growthloop.com/hom. Alex, thank you so much for your time today, sir. Really excited to chat.

Guys, such a pleasure to be here. Excited to dive in.

So a lot of folks are probably familiar with AirOps. We are excited to get the founder on the show and, you know, be remiss by not getting a chance to at least chat about some of the founding story, the history behind the company. You started AirOps in 2022. The initial product was to help non-technical employees access company data, and you quickly saw that a lot of the marketers that were in your user base were showing up and a lot of them were taking that data and feeding it into content workflows. And in parallel to this, you were an early investor in OpenAI and Anthropic. And so you kind of had this like inside track on the crazy capability curve of, of LLMs. Talk to us about like at some point why AirOps became a content engineering platform for AI search. Like when did the turn actually happen? Uh, what did you see that kind of made you commit to it?

Yeah, for sure. I mean, I think as you pointed out, AirOps was formed pre-LM. So we were building for 6 months to a year before I had even touched the kind of early LMs from OpenAI. And I remember I was actually having a conversation with, of all people, Sam Altman in San Francisco, and this was like early 2022. And I ask him the same question every time I see him, which is something along the lines of what's getting you really excited these days, or kind of what's What's like, what's sparking your imagination? And he just looked at me and he said, the AI stuff's getting really good. And I, and I sort of dug in. I was like, what use case? Explain that. Like, what do you mean? Because up until that point, we didn't really have general purpose AI in the way that we do today, obviously. And so he said to me, you know, we're getting close to the AI being able to read all of your emails and tell you what, what to do for that week, which sounded back then totally insane to me. And, and I planted in the back of my head at that point, this idea that I should go play with them. Models and figure out what was happening, OpenAI, and create an account. And then on a flight to Atlanta, I decided when I was sandwiched between these two gigantic gentlemen and I was in the middle seat, I was like, how am I going to use this time? I was like, okay, I'm going to now go and finally play with this OpenAI thing. So I signed up and created an account. And I remember just having just almost an out-of-body experience as it started to generate perfect SQL and create copy and do all these insane things that really for me started to melt the laws of physics that had governed software for the last 20 years. And I, I like obviously had no idea of how this was going to unfold, except that I felt very clearly this was like a massive paradigm shift and AirOps should figure out how it should be part of our story. So we started experimenting in that direction and initially the product was very broad, but when a space is growing as quickly as, as the kind of LLM application space is, I felt very clearly that it was important to pick and own a focused slice of that universe and not try to be everything to everyone. And when we looked at the kind of heat map of usage and the kinds of people that were getting value from us, it was, there was this bright white heart kind of center of activity with people working nights and weekends and building crazy ambitious things. And it were these technical CMOs, like CMOs have a 50 to 100 person marketing orgs that were diving in and building with AirOps. And it was just one of those moments where you saw the market telling you, hey, we like what you're doing. And we, as a high taste user, have really strong opinions on what we want to do with this technology. And at the point we lent into that, it took a little while to make that decision, but we lent into that and never looked back and have just loved building for this customer ever since. And that was probably around the beginning of, of 2020, um, mid-2023, '24, that, that really happened. And since then, that's been our sole focus, who we talk to every day and who we build for.

Super cool. I feel like everyone has that like moment in history now, like the first time they spun up ChatGPT and asked their first question and started getting a taste of the potential here. And it's crazy that we're only just, uh, you know, less than half a decade away from, uh, when GPT-3 launched to the public and now some of the crazy stuff we were just chatting about before we press record. Um, but I, I do wanna like get, get, let you introduce or maybe get your definition of like what content engineering is. Like we, a lot of our listeners are in marketing and, and play around in the marketing sphere. We are familiar with like marketing engineering, GTM engineering, and like content engineer. Marketing seems to always have a need to create new terms in an attempt to kind of like redefine our jobs. But content engineering is interesting because as a term, it's existed for a decade. It, it comes, uh, it comes out of like technical documentation world. I feel like on LinkedIn there's like thousands of people with like technical content engineering roles, but you've kind of taken this framing and applied it to something that is way newer than, uh, than the old school way of thinking about this. So you're building, operating AI content systems at scale specifically to optimize AI visibility. A lot of people start having a negative connotation when they hear content systems and AI-powered content, cuz they immediately think of the generic sloppy content that looks the same everywhere that we see on LinkedIn, across social. What's your take on how content engineering helps teams that don't want to just build more content at scale, but instead wanna scale specificity and write things that can't be replicated from public data?

Yeah, I think it's a, it's a really interesting question and I think If we zoom out and look at what's happening at the moment, there is now a new priority audience for marketing and content teams, which are these AI agents that act as an intermediary between buyers and brands. And one of the behaviors of these AI agents is they're able to synthesize large amounts of content, form a consensus, and then give a single response in a very efficient, you know, single pane of glass back to a user. And when we look at that system, we start to see that the quality, depth, freshness, and information gain of all of this content really matters. And so you want to make sure as a brand and as a business that the content that you have on your own website, the content that you publish to partners, how you show up in Reddit, how you show up on third-party sites is as on-message as accurate and as accurate and as good at preempting real customer questions as possible. So In our view, this is a big challenge, and one that you can't simply throw more people at to try and meet this need. I think for a long time, content was seen as being kind of an incremental, you know, we want to publish X blog posts a month, or we want to refresh Y posts a month. In this day and age where the models are reaching into the long tail, increasingly pulling a lot of content, and synthesizing answers, the need to keep content fresh and accurate is, is really needs a systems response, not just a human response. But when we think about AI content in general, there's this like kind of sleight of hand, and you'll see it over LinkedIn all the time, where the headline— and actually one of our competitors has been publishing nonstop on this topic— the headline is always AI content is leading to bad outcomes for brands. And then you click on the study, and you scroll down. And it says solely— the definition then changes to content which is solely generated by AI with no original grounding or context. And I think that, that conflation is very dangerous. Because the details matter. The reason it takes us 3 to 4 weeks to onboard a customer is because We spend a lot of time thinking about what unique context exists inside the business and what humans exist inside the business that we can leverage to get more knowledge out into the world that can then be rewarded with visibility. And that is a systems problem. That's a human problem. That's a strategy and taste problem. And so when we think about who's successful in this era, it's typically, we're seeing more and more, it's these high-taste professionals that can augment themselves with a few technical skills to take advantage of the AI and ways, build systems, think creatively about applying these technologies. And that's why we've kind of lent into this term content engineer, because it's taking all of the best things that content and content marketers and marketers have been doing, telling their brand story, but augmenting it with a few additional skills to allow them to meet this moment and be successful. But the distinctions here between, you know, mass-produced, ungrounded content that was really prevalent in the first act of this. And what's happening in sophisticated brands that deeply care about their brand heritage now is like night and day and actually very misunderstood, I think, by a lot of people.

One of the things that I— and I want to just dive a little deeper here and maybe make sure that I've— I fully understand your definition and that we're on like kind of operating off the same page. But like, I love the idea of the engineer, right? Like, I've thought of myself as a marketing engineer, etc., in the past as well. But what I think you're kind of laying out here is this idea that content is, is an equation to some extent or another. I've been a writer for 15 years. Honestly, the way I assemble content 15 years ago is more or less the same way. Research it, brief it, check out what the competitors are saying. But now with AI tools, there's this entirely different layer where you can automate all the, the inputs, right? Like the basic inputs as they come in. It sounds like what you guys are doing are working with the brands to bring in their personalized context, right? Like the, I don't know if you guys handle sales transcripts, but as an SEO guy, I would take sales transcripts over a keyword data sheet 10 times out of 10. I'm curious if you start to see the engineer's role in this as being really twofold. Fold one, or part one is arranging the datasets, the inputs, making sure that everything's tuned in the system. And then the other side is turning out what the output is, the quality of the output, the direction of it. Like, content isn't just a blog, it's a landing page, it's a podcast, it's a LinkedIn post. Is that a right framing from your perspective? And, and, uh, what would you add to that?

I think the, the best content has a healthy tension, um, because ultimately content is operating in a competitive market. You're competing for attention in the SERP, in ChatGPT's response in TikTok. So there is an element of the, an element of creating content where what's going to perform well needs to be a voice in, in the output. Otherwise, you know, you're creating things that doesn't resonate, doesn't perform. And then the other is the brand voice as well and the, the brand story, the positioning, the things the brand wants to say to the world. Um, and so, so when we think about the role of the content engineer, they're actually bringing together And trying to meet those tensions with content that both performs well and is on brand and continues to tell the story of that brand. And the ingredients to that are obviously kind of AI as a part of that, but also, as you mentioned, the internal context, which is so often misunderstood. So call transcripts are amazing. Zendesk tickets are amazing. Loom videos the team are creating are amazing. Things that are sitting inside of Google Drive collecting dust that someone spent a lot of time on. All of this context internally can lead to new demand and eyeballs if it can be safely gotten outside of the company's firewall into the web. So those pieces are really important. But then I think when we think about the construction of like a content system or the content engineer, a lot of that context is collected and gathered automatically. But then when we're answering a specific question, because on a call transcript, someone has started asking about a new regulation or a competitor feature, who internally is the right human to then go and get an updated response or to an updated positioning on that. And so the, the, the goal of the engineer is not to remove humans out of the loop. It's actually to find more humans to put into the loop, but, but have them practice at the top of their license. So if I get pinged as the product owner of a feature I shipped by, by AirOps or another platform to, to contribute to a piece of content, that then allows that piece of content to be more successful when it goes out into the web. And so the idea that the goal is to have completely autonomous systems that just run is actually not I don't think it works, and I actually think it misses the point of what can be achieved when you're partitioning a lot of the, the, the automatable pieces off of the human's plate and focusing them on net new opinion, reactions, additional context that lives inside their heads.

Very cool. Alex, can you help us like paint a picture of what this new body of tasks looks like inside some of your customers that are coining some of these new roles like content engineers? Like, are they mapping out like fanout queries for content and like making sure that the pipelines are pulling from the CMS and refreshing process, like loading brand context? Like there's a bunch of stuff in this like system in the pipeline. There's a whole measurement layer also. Like walk us through some of the, the roles, paint that picture for us.

Yeah, I think we, we, we talk about it internally as the Fab Five. Um, because there's kind of four 5 core pillars. I won't talk about all of them now, and we'll be here for a while, but I think a couple that really stand out is one, thinking about content freshness. So if you've just spent the last 5 years as a business building a fantastic blog, your Webflow is one of our customers, for example, and they have this fantastic blog that's been a labor of love for the last few years. A lot of those posts age periodically, either because the Webflow product shifts or because the content comes out of date. Because trends shift a lot in web design and building. And so one of the things we have, they've built with us is a kind of workflow that periodically reviews their blog posts and looks for positioning updates in their product that they want to go and update or things that have become dated. So for example, the web, like web Pantone Colors of the Year or kind of web colors of the year and those trends shift. And so they're able to use Aira to keep those updated. So the mindset shift of content is published and done, on the one hand, to content is an ongoing gardening maintenance opportunity and actually a really big growth lever given how much AI models in particular love fresh content is an important one. And that the systems to power that are very, very important. And then another one I would, I would add is, is thinking about the maintenance of your core, like, context layer. So when we think about the context layer, it's kind of the fact base that any automation agent or workflow can run off of. And a lot of things happen in a business and outside of a business that affect that context layer. So for example, you know, a competitor launches updated pricing, and we need to have a competitive response to that stored somewhere in the context layer, or the business has some fantastic new proof points or wins an award, that all of the listening to that shift internally and externally and maintaining that context base is something that, in our view, the content engineer can own to make sure that when an automation runs to update documentation, for example, it understands the present state of the business and the product. And you can have— you can make sure what you're putting out there is grounded in a really, really strong fact base. So the sort of 5 pillars ladder up to an entire system. And to be honest, most, most companies are still in the early innings of layering those on. But the goal is to have kind of just increased impact of the same set of people within the marketing org, leveraging AI in the places it's really good and pushing to humans in the places where we need them to contribute.

Yeah, super smart approach to all of this. I think one thing that jumps out at me is this role of the engineer is also about the scale the scale of things, right? Like as an individual SEO years and years ago, maintaining something as simple as internal links, like we redesigned the homepage and now we've got new top nav links and guess what? You've got 10,000 pages on a site. Good luck finding these 50,000 links. It sounds like you guys are starting to solve that with having workflows in place and listening in place and the context to be able to say, uh, big brand shift. Now we're going into market.

I think there's one of the problems with the AI conversation is something I want to call out is that, you know, a lot of the content is around making people feel like they're behind, you know, like LinkedIn is full of like, I, these 5 prompts have changed, but changed our company trajectory. I think that that works really well on social, it's designed to kind of trigger an emotion. Obviously, the really important message here is that everybody is at some point on this journey. And most people are relatively early and starting to build their first automation. And so I don't want like people listening to this to feel like, oh my God, I'm so behind and everything's effed and what am I going to do? This is really a fabulously creative moment in this industry. And everybody needs to sort of chip away at that with just building their first thing, getting their hands dirty, using— going— we have a wonderful free training program. People can get their hands dirty in Claude Code, but just get your fingers on the keyboard and start building things. And, um, it's, it's, uh, it's a, it's a big spectrum of adoption, and everybody is at one point on that.

Yeah, thank you for saying that. It needs to be said at least once during the podcast, and I appreciate taking a little bit of the, the air out of that balloon. It's making LinkedIn insufferable for, for practitioners like myself. Now, what I love about, about this conversation is how pragmatic it is, right? Is, is these things are not novel things, right? The activity, the deliverable, refreshing content.

This is—

you talk to any SEO worth their salt, they'll tell you, I'll put $10 on this before I put it on new content because it's so valuable. Now there's also the LLM angle to this, and I want to get your, your take on this as well. It's like, within your own system and the learnings that you and your customers have had, like, the LLMs will often cite that old content as well. Do you find that these are the types of positioning opportunities that aren't just like We'll boost some SEO, we'll make sure positioning's good, but we're also feeding the LLMs really valuable information and not adding like an extra spark to the fire, uh, behind adoption of your own product.

Yeah, I think we, we, um, increasingly— I'm actually very interested in this. I think the idea that, um, the first wave of SEO and, sorry, AO conversation was about, we mentioned in the result, and now I think the conversation is shifting a little bit to what is the opinion the model is forming? What is the kind of consensus opinion? Do people think we have— well, for AirOps, for example, you know, people may think our pricing is complicated. Okay, which is a very real thing. And so me as an AIO platform, you know, I'm concerned about that opinion. I want to understand why do people think that and actually being able to drill into the source conversation, be that a Reddit conversation, a blog post, a Stack Overflow that someone wrote, allows me to go and then outreach to those people and, and address the core concern and shift that opinion. So LLMs present a lot of the similar challenges of building authority, having deep content, but they also slightly change, I'd say, the contrast on some of these things to make something slightly more important. And this idea of model sentiment, the fact-based and the opinion it forms for specific personas, for specific points in the buying journey, is kind of a new frontier for us. So we're shipping a lot to support those kinds of remediatory like workflows and agents that can help with those specific things. But it is a, it's a very important actor in the buyer journey now to influence.

It's funny how, how the market has shifted a little bit from last year of I got to be cited at all costs to now I'm seeing a much more, are the LLMs accurate? Like I think there's a recognition that these LLMs hallucinate all the time. I built like my own internal LLM visibility tracking tool. And like the amount of hallucinated URLs that you find in the content from brands you own and, and so on is, is actually really, it's, it's actually good news for people who think they're behind or like, wait, these LLMs are great, but they're not infallible.

Yep. And I think if you look at the average length of a question in ChatGPT and you add on things like Personal Intelligence from Google, you add on the increased context that ChatGPT has about you in its memory, you know, what you're seeing is these models are going more and more into the long tail to find kind of novel tidbits and things, and that's dredging up a lot of really old content or content that's inaccurate. And so that kind of long tail challenge is very real for brands and something we talk a lot about. But it does mean that you can end up with totally wild responses from stale opinion or things that reflect your product or service from, from 2 years ago. So it's, it's very— it's, it's in so many ways where we're moving really fast, and there's so many ways we're still very early in a lot of these things.

I, I— last one, and then Phil, I'll shut up for a second and give you a chance. But one of the things, again, coming back to this engineering perspective, like, engineering has been typically reserved for people who build products. But largely what I'm seeing from your framing is Your content's a product, and if you have vulnerabilities in the backend of your product, you know, it doesn't matter if you coded it 10 years ago or yesterday, it's still a vulnerability. And I feel like that's the gap your, your company's closing.

I'm really excited about this kind of collapsing of the talent stack, meaning that the same person who lived in marketing can now reach into more technical domain to perform their role and vice versa, right? We have engineers who are now you know, looking through call transcripts with our customers and prioritizing features. It's a very exciting time for the willing and the curious. And so attaching engineer to a content role doesn't mean we expect people to learn Python, but it means that in now there, people are able to go and be far more impactful and can scale their impact as a professional and actually have a path as the world changes pretty quickly. So, If we, if we can flip as AROPS, if we can flip people to feeling a little bit behind to feeling like they're participating in this shift and they get some new skills and they get, they get some confidence and excitement from this and that allows them to expand their agency or their business. I think we've done our job and that makes me feel really good about our role. So the content engineering thing I think is, is a, is an umbrella term for people getting more leverage in this moment, but still bringing a lot of their traditional skills, their taste. And their strategic thinking to, to this challenge as it shifts.

Alex, I wanna give you a chance to chat about how AirOps is helping some of the hurdles related to content engineering. Like when I chat about this to folks, a lot of folks think that content engineering sounds really easy and straightforward, but anyone that has done it, like I don't have the, the, the year plus experience that, uh, John does, but I've been like automating a lot of the post-production pipeline on the podcast in the last few months and, you know, getting past that, like, human editor, for me, that's like myself, but in a lot of cases, you're working with a content team and it's really hard. Like, that's like the, the final boss is like the, the human editor here. Um, the value of like institutional knowledge, that context layer that we talked about, it's a tall order to solve with a workflow. And maybe you like chat about like working with your customers. What are you seeing in terms of like how to cross those hurdles? Are we ever gonna attain this like production acceptable output? Should we kind of accept that the goals, the goalposts are always kind of changing and this is more of like a dynamic thing. What are your thoughts there, Alex?

Yeah, I mean, I think that it was really funny when we started AirOps and we, we had our first kind of create an article template and I think it was like 3 steps in the workflow. It was like very early. We didn't know anything, right? We were like, oh, let's just see how people use it. And then I woke up on a Sunday morning and we had workflows that were 100+ steps. And it was one of those moments where it's like the difference between an output and an outcome. Was just so starkly illustrated on this canvas. And I think in that is all of the, like, the learnings, the taste, the kind of nuance, the things that this professional who was building on our platform had accumulated over the years of how they want to work, how they want their brand to sound, and the kinds of things you actually need to do to get really good outputs from an LLM. And so I think that the kind of The interesting thing is decomposing the pipeline that you have today and thinking about which things are great candidates for AI, which things may have been impossible before that you can now go do. So trawling through historic transcripts, for example, to look for references you may want to include. And then, and then really keeping humans very front and center until you can get a little bit more confidence on some of those things to hand them off. So it is a labor of love that I completely miss— did not appreciate at the beginning of this journey, but have— but building against high-taste brands and people that really genuinely care is the best way to get good at something. Because as you alluded to, people will not put things on their website that they are not happy with, and they certainly won't pay for a platform like AirOps. So most AI pilots fail, but 98% of our pilots convert to people becoming annual customers with us who work with our team. And that's because the 3 to 4 week process we put people through, this sort of content engineering boot up where we like audit their contacts, we help them think about their tone and voice, we get them to agree how they want to speak, is, is very structured. It's a lot of detail in there. And the only way we were able to do that was by running into the fire for the first year and learning all these things. So again, like it's, it's so easy to do things with AI that gets conflated with actual business outcomes. And the two things are just wildly different. And in that in that space is the craft of content engineering and content marketing.

This episode is brought to you by our friends at Knack. I have the amazing pleasure of chatting with some of the greatest minds in marketing operations, but you'd be shocked at how many of them have the same issues when it comes to their campaign creation process. They're dealing with things like scaling, personalization, keeping assets on brand, render testing emails, working in rigid templates, and worst of all, having to hand code emails. But many of the leaders in the space are using Knack. To eliminate these headaches. Knack is a no-code email and landing page creation platform that covers initial campaign request right until the sync to your marketing automation platform. Knack improves the whole campaign creation process by bringing the building, optimization, testing, collaboration, and also translation functions all in a single spot. So you don't have to jump around and coordinate feedback between a bunch of different people or wait on developers for various simple edits. Cause at the end of the day, marketing ops folks want to focus on process and efficiency. Marketers want to focus on creativity, and Knack makes sure you get both and then some. Check them out at knack.com, that's K-N-A-K, and tell them we sent you. This episode is brought to you by MoEngage and their new warehouse-native CEP, which revolutionizes the way data should be seamlessly fetched, analyzed, and activated. Now you can use MoEngage Cloud Data Sync to sync your warehouse data with MoEngage and use it to segment audiences or personalize messages without any development, just set and forget. Use MoEngage warehouse segments to create segments on your warehouse data without moving underlying events, properties, or proprietary segmentation logic, and activate them directly on MoEngage. Using warehouse data analytics, brands can also directly generate insights from their warehouse data. Your long-search solution for seamless data sync, smarter segmentation, dynamic personalization, and delightful customer experiences ends here. Warehouse Native CEP for Mowingage, making it easy, smart, and secure. Learn more at mowingage.com.

One of the things that I've broke my back on a few times, it sounds like you guys have solved, so kudos to you. But like, whenever I've tried to build these automation content systems, very similar, I've arrived at the, uh, the conclusion of multiple workflow steps are required to clean up the data, put the brief together, get research, all that fun stuff. How do you end up designing a system that isn't defensive against the editor? Like the editor is, is going to end up for most brands going to be the, the main stumbling block, at least in my experience, especially if the editor has any bias towards AI-assisted content. Uh, and, and we immediately see, okay, if they think it's AI-assisted, they'll put a different complete level of, uh, heuristics and criticism against it. And so I've fallen into the trap in the past of building very defensive systems that ended up sitting on good amounts of data, just like your clients are, of all these transcripts and stuff like that. And it comes out so vanilla. And the only way I could solve it, I'm curious if this is like just baked into your platform as a workflow step, is that human element, that 20 seconds on Whisper Flow where I, hey, this is what we're actually trying to achieve here. Like, is, is that the right way to think about this? Is that how you guys solve the problem?

Or yeah, I, I think our view is that we should have as much, the starting point should always be as much human involvement as we need and we want. I don't think our goal is not to remove, as I said, the goal is not to remove humans from the loop, it's just to kind of move them to the top of their license. And so if, for example, the editor wants to get pitched 4 or 5 briefs from AirOps and give a 30-second whisper flow adjustment to the brief before the outline is even written, then we should include that step. If they want to see the finished product and just make like edits because they've built a lot of trust or it's a, it's a, maybe it's a long tail piece of content, it's relatively simple, then we can do that too. I think we are, we really want to bring people along and that means meeting people at their current level of comfort. I think the second thing I'll say is that the system has to get better over time. And actually we made this mistake with the first version of AirOps where it didn't really get better over time, the more you used it, as much as, as much as it needed to. We, with AirOps Next, which just launched, by the time this goes out, You know, one of the core design principles is that the more someone uses the system, the more it learns your style and starts to incorporate, um, your, your kind of rules and things, even if you haven't explicitly told us them upfront. Um, a lot of people can't articulate what they want. They have to react to things to be able to, to kind of tease those out. And so that's a core design goal. But, um, you know, even if, even if it's, even if the, the brief is completely, um, uh, human written upfront. But we just use AI to battle test that against the existing SERP to think about interesting references that can be included. And that's our starting point. That's great. That's, that's moving people forward and they'll layer on from there. So this is where the judgment is important and the incremental kind of layering of this is important.

I wonder if, uh, you know, as you're talking, I'm like, oh, I wonder what you think about the optimal places in these types of workflows to put the human in the loop. Because I think we're all on the same page here is that human-in-the-loop, AI-assisted, that's a path forward. It's not replacing people, it's augmenting. What do you see as being those opportunities where that human touch is a multiplicative factor? Just like AI can help us create 50 briefs at once, where does the human touch get the most bang for the buck in these, these new modern workflows?

Yeah, I think AI is not very good out the box at ideation, um, meaning that it, it's going to suggest a lot of things you could write about and a very small subset of those by default are generally good, like very good ideas. Um, so I think that's one where we are a long way from AI kind of calling its shot really well on, on, on great pitches for net new pieces of particularly editorial content. Like that's kind of, I think that's, you can see a lot of volume there, but actually a high taste user is going to have a lot of feedback there. So I think For sure, calling the shots on top-of-funnel content, I think is very, very hard and wants something that humans should be very involved in. I think when it comes to perspective as well and opinion, you know, the models are ultimately doing retrieval, like ChatGPT is doing retrieval to get the kind of frontier opinion and thought. And so there is a huge bias towards fresh content. If you refresh content, you update the freshness signals, your chances of being cited increase massively. So that means that when you're creating a piece of content in an AI workflow, figuring out who internally is the best person to add unique perspective to that and to provide some commentary and to provide a reaction, an attributed reaction to add net new perspective to something is another very, very important one. But, um, I think the goal here is to actually, um, try and step out of at least some of those things over time that you are very hands-on keyboard with initially. Um, once you get a little bit more confidence in, in, in how these things play out and, and are able to provide more guidance and context to the model. But certainly kind of top of funnel query, like ideation, I think is very hard. And then I think the, the kind of perspective and opinion and unique POV that's so important on top of all your internal ground funding again is something that I think if you don't have it in Loom videos or kind of internal resources is again something that, that should be added to most pieces to give it, to give it that authenticity.

So you said, Alex, that like in the early days it was just a couple of workflow steps and then you woke up at some point it was like 100 plus. Um, you said in a couple other interviews that most of the workflows you see with clients and customers have 30 to 50 steps to produce what you're kind of calling excellent AI-optimized content. Maybe take us through like what you think are the most important first steps, like the early steps as part of the process, and what's happening later on in the end, like the final checks. You just talked about like human in the loop and where that kind of falls in, but like even without the human steps in there, like what needs to happen at the start and what are you kind of seeing super important at the end?

Yeah, I, I will, I'll say that by the time this goes out to, to everyone, uh, we will have moved on from workflows to a new structure for building these things out, which will look a lot more like natural language playbooks that will then run for you. We're trying to lower the bar to building these things. But the structure remains pretty similar in the old world and the new world. I think the most important, the most important sort of headline is being thoughtful about context gathering and preparation. Even in a world where you're using Claude or you're using Airops or whatever platform it is your tool of choice, being very thoughtful about where the right information lives inside the business, outside the business is, I would say, 70% of the work. And connecting up the right connectors, making sure the context is prepared, thinking about how to query that and make sure that you are getting the right subset of that that internal knowledge is for a particular piece of content is very, very important. And often the content is internally is not ready for use by the LLMs. So there's often some prep work to do to make sure that, um, whether you're connecting directly to Google Drive, for example, or you're trying to get Gong transcripts, um, making sure those connectors work well, you're thoughtful about how you go and fetch from those services. And you sort of structure the retrieval process in a logical way is a huge amount of the work. So I would say the first column is getting information from inside the business. And I would start simply with just a Google Drive folder with as much information as relevant as you have available. Like start with that single source of information and go from there. The second is collecting information on the competitive dynamics of the piece of content you're creating, right? So content needs to win, in our view, a particular keyword, a search, a query. And so understanding what you're up against is important because you're not looking to create the same piece of content as everyone else. But there are certain clues in the content that's ranking that does tell you the kind of format that's expected, the sorts of like length, the kinds of topics that are probably expected to by the algorithms to have in a specific dynamic, a specific situation. And so doing a deep audit, understanding what's happening there and just really like getting a sense of the, what's included and what's not in the ranking content gives you clues on what you should cover and where there's opportunity as well. So that's your, those are your two, two of your biggest context sources internally. And then the sort of dynamic that you're competing competing with. And then how you bring that together is very much the next piece of, okay, do we want to start with pitches? Do we want to start with maybe an outline that we're providing ourselves? Do we want to actually start with an interview guide for the team? Because often you'll want to go and get a lot of information to add to an article, and Happee AI basically craft a really great interview guide for 1 or 3 people, 1, 3 people on your team. And then it goes through the creation process and a lot of folks are still going through the kind of brief to brief, brief, brief editing to article drafting to revisions to feedback to publish. And they can replicate that with kind of these human breakpoints along the way. And actually the interesting thing is Refresh works pretty similarly as well. A lot of folks are doing Refresh briefs. So they can be very specific about what's getting changed and why. And being thoughtful about designing for human input and the sort of procedure you want to run is a negotiation between members of the team participating in this.

One of the things that jumps out at me as we're talking is as your customers have become successful with this, right, you talk about a 3 to 4 week onboarding period. There's probably a lag period before they've got their internal workflows all tuned up so everybody's in the loop. And then comes the rush, right? We are now 10x more productive than we were 6 months ago. This is an experience that I've had in my own, my own small little shop here is like the capacity is, is exploded. One of the limiting factors, uh, that I want your take on is attention. I, I increasingly see my own attention as being the limiting factors for the amount of work that I can accomplish. Like, in theory, I could sit down and publish out 15 articles, 20 articles in a day, which would have been a month's worth of work 5 years ago. But what, what gets you is the drain, right?

The—

okay, context switching, gathering information, refining. How do we see the role of like a content marketer, content engineer evolving around this attention? Do you kind of agree that this becomes a limiting factor? I have a content calendar that I know I can execute against. I have the internal inputs. I have AirOps that does all the external research and validation. And now I need to sit my butt in the seat and do the work and manage all this stuff from briefs to even publishing it in a CMS.

Yeah, I think it's so interesting, right? Because every time AI kind of takes one of these leaps forward, we're very good at instantly recognizing all the things it made easier. So all the friction that gets removed is like, I get it. Like that, that job that used to take me 3 hours now goes to 5 minutes. What we're really bad at, I think, is understanding where the friction moves in the system and what's the new set of challenges that we need to solve. And so if you look at what's happened in engineering, traditional software engineering, review is now the bottleneck for so many things and just like literally just processing and focusing on is this good? Is this something we want to add to our codebase? So we don't spiral into like software slop. And so I think this idea that the friction has moved in content creation to different parts of the system is very real. And this idea of, of, of like the review surface being one of the most important parts of the process to help people accelerate, but also to make sure that it's an enjoyable experience, honestly. Um, if you got to— is a core design goal of ours. And with our, with our new platform, because we recognize that a lot of the work now is in thoughtfully, um, contributing, um, pushing back, augmenting, like adjusting, uh, work that's happening in these, in these AI systems to make sure that the brand isn't just getting diluted with noise. It's very important. One of the things with this AI search era is also that You know, if you're just creating content that's, that's just, that's just ungrounded, or it's off, or it's off message, or the positioning is weak, you're just polluting your brand story. And so actually, like, getting review right and being disciplined about it, and that being as an efficient process as possible, is I think one of the new great challenges of this era. Also, if everybody has the same tools— we're all very excited we can create more content, but if everybody has the same tools and it's a competitive Sport, you know, ultimately people have to reach deeper and get more creative to get the alpha and to really win. And so I think we are now beginning to understand how things like experiential content, video content, content that's like truly dis— differentiated and novel with great information gain is just getting more of a premium as I think the middle bar gets a little bit more accessible and commoditized to people irrespective of, of budget. So it's a very interesting time, but it's all— marketing's always involved, always have been evolving, and this is just feels like a period of acceleration on that front.

I want to give you a chance, Alex, to talk about some of the technical details, uh, that lets AirOps stand out from some of the other competitive tools on the market here. Um, like one of the last mile steps that I haven't taken yet with my own little system is connecting everything to my CMS, and that's really where I feel like you're unlocking a lot of the automation, right? Like outputting content in a text file or a markup file, that's one thing. And then having to like copy paste that in your CMS, it's a lot better than what we were doing before, but being able to have a direct MCP or API connection to our CMS and without having to log into it, refresh content and run pipelines on that. I feel like that CMS integration piece, it sounds really clean from the outside, but Anyone who's actually tried to automate that last mile, especially with like headless CMS, there's like a schema problem. There's structured fields and tables and cells that are manually mapped. Like it's, there's a lot of like things and not to mention like IT policies and API access. How, how is AirOps solving this last mile publishing problem at like an enterprise scale? A lot of your customers are enterprise, massive team, massive like, uh, content libraries where, you know, like there's technical constraints that are often more limiting than a content quality problem. What are your thoughts there?

Yeah, I'll know we've hit AGI when, uh, it can post an article to an enterprise CMS. It's like, because I, I'm as a tools person, like, tool native person, you know, I'm used to fantastic tools like Webflow, super easy to integrate with. But you, you work with enterprise and, and people have a 3 to 6 month lead time on adding a new page, and you're like, how on earth are you guys going to survive? We have to, we have to feed up We actually have 4 dedicated engineers who just integrate with enterprise CMSs. That's all they do. And God bless them. Like, it's, it's not for the faint-hearted. These are like incredibly complex systems and we build the pipes to integrate with them. So for the easier CMSs, it's all out of the box. For the ones that are more, you know, certainly have complex schemas and things, we, we, we build custom integrations, but both in terms of the data ingestion side. So the systems we can connect to for context, how we process that, and then how we push out final outputs. That's a big part of the value we bring to, to allow these workflows and agents to go, to go live. And in the middle of our platform, it's broadly split up into 3 layers. We have all the data we collect, 2 million answers a day from AI search engines. We analyze them, we understand sentiment drivers and citations, all that good stuff. We have our brand context layer, which is just added a lot of visual brand guidelines as well now. So you can create microsites and things off of it and it's now versioned and much more of a kind of governed source of truth. And then the third piece is the action layer, which we just relaunched to be agent-first and allows people to get, have more ambitious use cases. And what we've observed is that each of those pillars is valuable kind of on its own, right? Like some people want two of the three, or they're using another tool, or they'd like to work, you know, in their own setup. But also people are working increasingly inside of things like Cloud Code. And so every piece of AROps is available through an MCP. In fact, one of our customers, Ramp, actually runs all of their workflows in AirOps directly through an MCP, and they orchestrate it all from their own system, and they're super, super happy. And that's us kind of acknowledging that the top 5, 10, 15% of users now are really racing ahead through personal AI enablement and things like Claude CoWork. And we're excited to be part of that journey, but it's It's not all happening in a UI anymore. A lot of people are on CLI, a lot of people are building in different ways. So I'm thrilled by it, but it does present a reinvention moment for us and the role we want to play in that story.

I want to dig a little deeper on the Eugenics stuff that you mentioned. I think that's super fascinating. It's kind of an AI operator-first mentality. I can't remember the last time I've signed into some of the software that I use, but I pay a lot of API fees.

Yep.

I know modern SaaS is, you know, especially on LinkedIn, we shouldn't trash it too much, but on LinkedIn, you know, modern SaaS is dead, is dying, it's going away forever. But I think that AI operators like myself, Phil, yourself, we're challenging the, the normal operating model. How much do you think that SaaS is going to evolve to being like, hey, I signed up Airops. Now I have an API call that I can use with cloud code that does 50 other things that are specialized versus we still work with within the UIs and, and evolve that way. Like the, I guess maybe not so much the future of software, but like how much do you think the market goes towards AI operators using terminal, and then the software that either is built specifically for AI operators or is just agnostic and can be used in both cases?

This is a fantastic question, and it's honestly, it's an existential question for SaaS, to be honest. And it, so we probably need another hour to talk about it. I think to my point, my point about the friction moves in the system. I don't think anybody wants to review and approve feedback on a piece of content from a Claude code terminal necessarily. I think that would be like a pretty cumbersome experience. I also think that there's a lot of instances where, um, you know, specialized UIs can really help accelerate that review friction point that you, that you talked about. So, um, I think work is changing. I think the, the, if there is this cohort of users such as yourself, John, who are working with extreme leverage inside of their own setups. But then content is also a team sport that involves multiple people. There's a lot of kind of interesting new challenges that are emerging around governance of context. Where do skills live? How do they change? How do we know if they're good? Cost controls. All of these things that are sort of starting to become new problems to be solved, and SaaS needs to make sure it has its finger on the pulse on how the, how the, the challenges of getting to scale and getting to impact are evolving. For us, like, I think there's a big piece of this is having datasets that don't exist anywhere else, um, to make sure that the content that's created is very competitive and understands how the things it needs to hit on in order to be successful. That's our kind of our insights layer. Um, and then, uh, the brand context piece we talked about, but I think that's ballooned in the last 6 months as being something that's really concerning for CMOs because they're seeing content go out that talks about their product in 7 different ways. And then for the actual agent runtime, I think that there is an interesting question of where people will put their agents. And I think right now we're really in the sort of early innings of this with solo kind of local machine-based agents, some early cloud agents. Our bet is that we can build a better cloud agent runtime than anyone else for our use cases. And we've been working on that for 6 months. And I think buyers will become more sophisticated in how they think about where their agents are running, both from a cost standpoint, observability, multi-party, human in the loop, thinking about the data we could provide those agents and also how they get better over time. So we worked with Anthropic very closely on this, this new launch. And I think what, what emerged as we were going through that is that that the house that the agent lives in is actually going to be one of the most important drivers of impact that will emerge in the second half of this year. So TBD, but we're pretty bullish on that.

Fascinating. Fascinating.

Yeah, really cool to hear that, Alex. Got a couple more questions for you. I took a deep dive rabbit hole into content engineering in preparation for this conversation with you. There's one article that kept popping up in my research that was written by one of your competitors and it's called "I Wouldn't Hire a Content Engineer," basically the case against content engineering. His argument is like the role is hiring for skills we won't need for a long time. He chose— he would choose a great writer with no AI fluency over a middling writer who knows a lot of the agentic tools. What do you think he gets right in that article? I don't know if you had a chance to read it, but like where do you kind of disagree with that take?

Yeah, I mean, I think, I hope there is a role for great writers forever. And I think that there's a, you know, destination content that's thoughtful, that really is like deeply invested in is a huge growth lever for businesses still. I think in my mind, like the, there's a lot of area under the curve. First, as a starting point that your best writers are not going to have time to get after. And what that gives you is basically a marketing asset that's decaying over time. So I don't, I, I like, I, I actually haven't read this article. There's been a few recent potshots at content engineering. Some of them misunderstand it. Some of them are great critiques as we evolve this discipline. But my general view is like, if you have a really great writer, hold on to them at all costs and have them do work that is going to be the most impactful they can and go as deep as possible. I would prefer to pair a great writer with some content engineering and some agents in order to make sure they're really focusing on the right things. They're prepped to write, they understand the questions they should be answering, they understand the discourse going into a writing session, then have them just sort of sit on a blank page and figure everything out themselves. And so it kind of is again falling into the— just skimming it now, falls into that trap of thinking the goal here is to remove humans from the loop. I, I want more people in the loop in content, and I think that's I want them to be doing really great work to create content that's differentiated, that gets picked up and gets cited, not removing people and having the systems just spew out stuff that doesn't add anything and is just recycling the same talking points. Again, like the, the, the other piece is that I actually think that the most effective people in every discipline in knowledge work are learning some more systems and technical concepts. You can call that whatever you want. I do not have any opinion on changing your LinkedIn to content engineer as much as I love to see it. But the core principle here is, is that the mind of the writer is going to have really creative, interesting ways to use these tools that other people, other people probably wouldn't. And so our view is that the most successful folks will be a blend of the traditional art and taste and creativity and journalism with AI tools to, to, to really be able to leverage their impact. And that ratio will shift depending on the nature of the content task.

We're 3 years in the future, Alex Halliday, a team that got content engineering really right. What does their team look like? Like headcount structure, how the roles sit relative to each other. You kind of mentioned like great writers paired with content engineers. Map that out for us.

Yeah, so our view on this is that Marketing teams will onboard a new org, like a new, a new sub team. And that team will be captained by one primary agent with a team of sub agents that will help run the operations of the marketing team. And I think realistically, people will have multiple agent platforms working for them. The one that we are standing up is the, the kind of content engineering team, content operations team. And the goal is to give, to make sure that the humans and people in the loop are used as efficiently as possible. So they understand the shifts that are happening in the market. They understand competitor moves and the content that's being created. They understand where they're losing and gaining ground, and they understand the internal and external context shifts that should drive and prompt them to create content. And so my vision is that you have a really fantastic high, like extremely productive few hours inside of AirOps, where you come in and you are effectively given really great prompts and things to react to that can extract context. You can exercise your taste, you can add your, your Whisperflow note on, on something you have as an opinion. You can get educated on what's shifting in the business. And that allows content to become a kind of more proactive system over time. And, and because it's just, and hopefully it's really fun. And when it's really fun, I hope more people want to get involved. More people in the company want to contribute their POVs. More customers maybe want to contribute their perspective. And we can actually make the kind of the maintenance of our brand story and the content footprint we have something that is just, is just, is just more energetic than it, than it is today where a few people have to like really bear the bulk of the burdens. So, um, I think the big shift in AI systems we'll see within 3 years is that they become far more proactive. Trust increases as context and capabilities increase, and we learn how to work with them in a way that makes sure that the brand we spent years building can, can, can kind of really thrive. And, and we have far more coverage than we do today and the kinds of intents that we want to— I really think most people are doing like 10% of what they could be doing if they really could scale themselves. And that will take years to get to. But I think the future is really really, really bright and exciting.

Thinking of that content team, and I'll— my last question, I guess, uh, for the, for the podcast. But one of the things in my own experience is I've been learning on the battlefield, right? Getting that battlefield promotion because you're just figuring out like, okay, this works, this doesn't work. I know you guys are releasing some training, and I think that's very much needed in, in the market. Do you feel like the space is evolving so quick that that, you know, what you, what you're experiencing in 2026 won't be applicable in 2028? Do you think that things will carry forward that foundational, hey, I figured out this worked and didn't work, and then the model changes it? Like, how, like, how would somebody want to be on that team in 3 years from now, train themselves to be an effective member of it? It's just being on the tools every day. Is it hands-on? Is it reps? Is it taking a course? Combination?

Yeah. I mean, I think it's, this is a question everybody's, I think, asking in the kind of core knowledge work pillars right now. But, you know, I think you can doom scroll your way into getting into various headspaces that are ultimately unproductive. I think what is productive is treating experimentation and learning as a core, just role, a core duty that we all have now as professionals. And because this is such an interesting time, I think the more you can institutionalize play and experimentation in your week, the better. So we have play as a core operating principle at AirOps, and it doesn't mean go and like, you know, build, build sandcastles and, and Lego buildings, but it does mean to follow your curiosity and try things and share what works. Nothing like lowers my blood pressure more than just using tools and building and trying things. And that's what we try and push our customers to do too. The, the upside of that is that you really get used to testing and learning and feeding that into your playbook and iterating because it will, it will shift, it will shift quickly. But also, only you can really be the arbiter of good as a marketer, I think. And so the best way to get confidence and evolve your playbook is to have, you know, maybe a few people on your team, maybe one person, maybe it's you, just really carve out time to figure out where the value is in any given model generation, any given tool, and then work that into how you work. I will say that the spread of adoption has never been wider. Like, if you talk to the top 1% of marketers right now, and you talk to the median marketer, that gap has never been bigger. And there's going to be many years of people making their way up that kind of adoption curve or AI curve. And there's nothing wrong with that, right? I think it is, everyone's on their own journey, but making sure that we treat learning and sharing of learnings in teams as a core pillar, I think is super important.

I love it. Such a great answer, Alex. Really appreciate your time with us today. I got one last question for you. You're obviously a founder, CEO, speaker, investor, but you're also a home chef at home and active fitness fiend. You got a ton of other stuff going on. One question we ask everyone is, how do you decide what deserves your energy at any given moment, and what's your personal system for staying aligned with what actually makes you happy?

Um, I think you have to— for me, I have to split my time between what gives me energy and what I have to do in my role. Um, I think this is where I have had to, as the business has gotten significantly bigger, got much better at prioritization. I actually have a personal OpenClaw, that's primary role is to help me prioritize given all the noise around me. And I split the things I do in any given week between the, the, the shit, the sort of need to dos and the things that really give me energy and push my thinking. So back to the play point, I spend probably the, probably an entire weekend day per week just building things in the business using our tool, just kind of really following my curiosity, which gives me a lot of energy. And then in the week I'm spending time doing doing, you know, the other roles of a CEO. I do, I know everybody talks about kind of mind-body wellness community. I think just like the assault on your psyche that running a startup is, means that those things are so important. And I have had periods where I've neglected them much to my detriment. But this year, I'm trying to be a whole person, get outside. The West Side Highway is now open in New York as of, as of the weather improving 2 days ago. So I'm, I'm very excited for the summer. And hopefully that feeds my creativity and makes me a better leader. So it's a constant balancing act and I have my open clause to help me a little bit and it yells at me all the time when I don't do my things.

Yeah. So say we all.

Yeah. I love it. Alex, thanks so much for your time today. Really appreciate it. We'll obviously share out AirOps and all the new features that you kind of teased out for us today. But yeah, really appreciate your time. Thanks so much for doing this.

Such a pleasure, Phil, John. Nice to meet you both. This has been great.