• LinkedIn
• Twitter
• Instagram
• YouTube
• RSS Feed

Apple •Spotify• Pocket Casts •Youtube •Overcast •RSS

What’s up everyone, today we have the pleasure of sitting down with Alex Halliday, Founder and CEO at AirOps.

Summary: Alex breaks down what content engineering actually means in 2026: building the systems infrastructure to maintain quality, freshness, and brand accuracy across everything a company has ever put online. He makes the counterintuitive case that great content engineering puts more humans into the content process, and explains why 98% of AirOps’s pilots convert to annual customers while most AI content pilots fail. If you think AI content is just a faster way to publish more, this episode will for sure change how you think about it.

In this Episode…

• How AirOps Pivoted to AI Content Engineering
• The Real Definition of Content Engineering and Why It’s Not About Publishing More
• What a Content Engineer Does That a Senior Content Marketer Does Not
• What It Actually Takes to Get AI Content Past a Human Editor
• Why Review Becomes the Bottleneck After You Automate Content Production
• Why Enterprise CMS Integration Is Harder Than the Content Quality Problem
• Why the Agent Runtime Is the Next Competitive Battleground for Content Teams
• What the Case Against Content Engineering Gets Wrong About the Role
• What a Content Engineering Team Looks Like in 3 Years

About Alex

Alex Halliday is the Founder and CEO of AirOps, where he leads the development of AI content engineering systems that help brands build visibility in AI search. Before founding AirOps in 2022, he served as Head of Product at MasterClass, where he was the company’s first product hire and helped scale revenue 10x. As a Venture Partner at SparkLabs Global Accelerator, he’s made early investments in OpenAI, Anthropic, Groq, and Discord.

How AirOps Pivoted to AI Content Engineering

In early 2022, the LLM moment hadn’t really happened yet, at least not for most of us. GPT-3 barely existed and wasn’t really on anyone’s radar in marketing. Most “AI for marketing” convos were still about sentiment analysis tools and basic chatbots. The prevailing assumption at the time was still that software had rules, rules had limits, and those limits were sort of the floor you designed around.

But Alex saw into the future. He saw a different future with AI in marketing, but to be fair, he had an unusual vantage point. As a venture partner at SparkLabs Global Accelerator with early investments in OpenAI and Anthropic (nbd), he was closer to what was actually happening than almost anyone on this planet. He still wasn’t ready for what came next.

It started with a conversation. He was in San Francisco with Sam Altman, something he made a habit of: whenever they crossed paths, Alex asked the same question: what’s sparking your imagination these days? On this particular occasion, Sam’s answer was a bit different. The “AI stuff” was getting really good, he said. When Alex pushed for specifics, Sam told him they were getting close to AI that could read all your emails and tell you what to do for the week. It’s kind of funny to reflect on that and how today that sounds super vanilla… but at the time, it sounded completely insane.

Alex filed it away. Then, a few weeks later, he was on a flight to Atlanta, sandwiched in the middle seat between two dudes with nowhere to go and nothing else to do. So he finally opened an OpenAI account and started building.

“I remember just having 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.”

That experience in a cramped middle seat sent AirOps in a new direction. The company had been founded to help non-technical employees access company data, a broad product with no obvious north star. But there was traction. And knowing the paradigm was shifting and knowing what your company should actually do about it are totally different problems. Alex had to translate that conviction into a focus, which meant making a hard call. When a space is growing as fast as LLM applications were in 2022 and 2023, trying to be everything to everyone is a trap.

The answer came from the data. When the team looked at their heat map of usage, one cluster burned hotter than others: technical marketers, leaders of 50 to 100 person marketing orgs, working nights and weekends inside AirOps working on early versions of what would become known as content systems. They were high-taste users with strong opinions and no patience for tools that couldn’t meet their standard. The market was doing what markets do when they find something they want, and it was insisting.

The Signals That Pointed AirOps Toward Content Engineering

• Technical marketers were already using the product heavily
• They were working nights and weekends inside the tool
• They had high standards and very specific feedback
• Their use cases were about systems, not one-off content generation
• The usage data showed a clearer customer than the original company story did

By mid-2023, AirOps had committed fully. The customer was the high-taste marketing professional who wanted to build content systems at scale, not just generate more content. Every decision since has been built around that person. The most important pivots happen when you actually use the thing, look at the data honestly, and trust what the market is telling you over the story you had planned to tell.

Key takeaway: Look at your usage data and find the cluster of users who are working hardest and complaining most specifically. They are telling you who your product is actually for. Make time to try the tools reshaping your industry with your own hands. Alex’s pivot started in a cramped middle seat he couldn’t escape. Any open hour will do.

Back to the top ⬆️

The Real Definition of Content Engineering and Why It’s Not About Publishing More

Marketing teams have been chasing the wrong metric since LLMs went mainstream. The race defaulted to volume: how many posts, how fast, how much can you automate. That framing made sense in an era where more content meant more crawlable pages, more keywords, more surface area for Google to index. The era has changed.

AI agents now sit between buyers and brands. When someone asks ChatGPT or Perplexity a question about your product category, an agent synthesizes content from across the web: your owned pages, third-party publications, Reddit threads, review platforms, then returns a single answer. That agent is not counting pages. It’s evaluating quality, depth, freshness, and what Alex describes as information gain: the degree to which any given piece of content adds something new to what the model already knows.

That’s a meaningfully different standard. A 2022 blog post with outdated product language, stale statistics, and broken links doesn’t rank lower in AI search. It’s absent from it entirely.

What AI Search Seems to Reward

• Fresh, accurate information
• Original context from inside the business
• Specific examples and proof points
• Clear attribution to real people or sources
• Content that adds something new to the existing answer set
• Consistent product and positioning language
• Pages that stay maintained after publication

Webflow, one of AirOps’s customers, saw what investing in content refresh workflows does to those outcomes: 42% more traffic and AI-attributed conversions performing 6x better than standard organic. That’s a maintenance story, not a content production story.

“The reason it takes us three to four 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, but can then be rewarded with visibility.”

There’s also a conflation doing a lot of damage in this conversation. Content written with AI assistance gets lumped together with content generated by AI with no original grounding or context. The studies that say “AI content performs poorly” tend to define AI content as the second category, and the conflation goes unexamined in most LinkedIn commentary. The distinction matters enormously. Content that draws on real interviews, proprietary data, internal expertise, and company-specific context performs differently from content that’s a model recombining what already exists on the internet.

Old Content Marketing	Content Engineering
Publish more net-new content	Maintain a living content system
Optimize for keywords and rankings	Optimize for freshness, citations, and information gain
Treat old posts as an archive	Refresh old posts as a growth lever
Use AI to draft faster	Use AI to surface, structure, and route context
Measure traffic only	Measure AI visibility, citations, and conversions

The brands performing well in AI search right now are treating their content library as a living system with real quality standards, maintained like a garden rather than stored as a publishing archive. They’re building workflows to keep content fresh, surface internal knowledge that’s been sitting in Google Drive unused, and maintain what Alex calls the context layer: the fact base that any automation can run against. That’s the infrastructure for AI visibility. More volume built on top of a stale or ungrounded foundation just adds more content that ages out of the picture.

Content engineering, in this framing, is a new discipline that adds the systems layer content marketing never had.

Key takeaway: Before building any new content, audit what you already have for freshness and accuracy. Run your brand name through ChatGPT and Perplexity and note what comes back: what’s cited, what’s missing, what’s wrong. Content not updated in the past 6 to 12 months is significantly less likely to earn AI citations. Start the refresh queue before the production queue.

Back to the top ⬆️

What a Content Engineer Does That a Senior Content Marketer Does Not

For 15 years, content creation followed roughly the same process: research a topic, brief a writer, check the SERP, publish, and move on. AI tools have changed some of those steps. What they haven’t changed, according to Alex, is the fundamental question content teams keep getting wrong: what is the actual job?

The best content has a healthy tension baked in. It has to perform: compete for attention in search results, in AI answers, in social feeds. And it has to tell the brand’s story accurately, in the brand’s voice, with the brand’s actual positioning. Bringing those 2 forces together is the content engineer’s job. Pure performance optimization without brand grounding produces generic content that looks like everyone else’s. Pure brand storytelling without performance thinking gives you content nobody discovers.

What makes the role distinct from senior content marketing is the internal context piece. Alex is specific about what he means. Sales call transcripts, Zendesk tickets, Loom videos team members created and forgot about, documents sitting in Google Drive that took someone a week to write: all of this is source material that can be turned into demand if it gets out of the firewall and into the world. The content engineer builds the system to do that automatically, and then brings in the right humans to add perspective that only they have.

“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 have them practice at the top of their license.”

When a product owner gets pinged by an AI workflow to add their perspective on a feature they shipped, they’re contributing knowledge that’s not available anywhere else on the internet. That’s the information gain AI search rewards. The content engineer designs the system that routes those requests to the right people, at the right time, with the right questions, and then builds that knowledge into the content layer.

The talent stack is collapsing in both directions. Engineers are reading through call transcripts to understand customer priorities. Marketers are building workflows, connecting APIs, designing automated pipelines. Alex frames this as a leverage story. People who lived in marketing can now reach into more technical domains to perform their role, and that expanded range is worth something real. Attaching “engineer” to a content role is an umbrella term for getting more done with traditional skills intact, not a career change into Python.

Senior Content Marketer	Content Engineer
Writes and edits content	Designs systems that help content get made, refreshed, and improved
Researches topics manually	Connects internal and external knowledge sources
Works with SMEs through interviews	Routes questions to the right experts at the right time
Owns quality and voice	Owns quality, voice, context, and workflow logic
Publishes and moves on	Maintains the content library over time

The marketers who understand both what to make and how to build the systems to make it at scale are going to have a growing advantage over those who specialize in only one half.

Key takeaway: Map your internal knowledge assets before you design any content workflow. Sales call transcripts, support tickets, expert Loom recordings, product documentation: list what your company knows that nobody else does. That’s your content differentiation. Then design your workflow to route that knowledge into your content rather than recycling what’s already on the internet.

Back to the top ⬆️

How Content Engineers Maintain a Library That AI Search Actually Cites

Most marketing teams are in early innings. They’ve started experimenting with AI for content creation, they’ve seen what’s possible, and then they’ve realized how much infrastructure actually needs to exist before the creation step matters. Alex calls his internal framework the Fab 5: 5 pillars that ladder up to a full content engineering system.

2 of them stand out as the foundation everything else builds on.

Content Freshness

The first is content freshness. Webflow built a workflow with AirOps that periodically reviews their blog posts and looks for positioning updates: product changes, dated references, trend shifts in web design and building.

What a Freshness Workflow Should Check

• Product changes
• New positioning
• Outdated stats
• Broken or weak links
• Old screenshots
• Dated trend references
• Competitor movement
• New customer proof points
• Missing AI-search citations
• Pages that still get traffic but no longer reflect the business

The Pantone colors of the year change. Webflow’s product changes. A post from 2022 that was accurate then may be quietly misleading now, and an AI agent reading it won’t flag it as outdated. It’ll just use it.

“Content is an ongoing gardening and maintenance opportunity and actually a really big growth lever, given how much AI models in particular love fresh content.”

Context Layer

The second pillar is the context layer: the fact base that any workflow or agent can run against. Businesses are constantly generating information that should live in this layer but doesn’t: a competitor updates pricing, a company wins an award, a product gets repositioned, a new case study closes.

Each of those shifts should update the context layer automatically. Without that, the next time an automation runs to update a piece of documentation, it runs against a fact base that’s months behind. What goes out reflects a version of the company that no longer exists.

What Belongs in the Context Layer

• Current positioning
• Product facts
• Feature descriptions
• Customer proof points
• Case studies
• Competitive notes
• Brand voice guidelines
• Approved claims
• Recent launches
• Pricing or packaging notes
• Industry POVs
• Executive-approved narratives

The content engineer owns that layer. Building it, maintaining it, listening for shifts internally and externally, routing the right updates to the right people. That’s the job that makes everything else work.

Key takeaway: Build a simple context layer before you build any content automation. Start with a shared Google Drive folder containing your current positioning, key proof points, competitor notes, and recent product updates. Every workflow you build should pull from this folder as a source of truth. Update it whenever something material changes, because the quality of your AI content is only as good as the context you feed it.

Back to the top ⬆️

What It Actually Takes to Get AI Content Past a Human Editor

Anyone who has actually tried to build an AI content system for a brand with real editorial standards has had the same experience: it’s harder than it looks. The demos are clean. The first test outputs seem promising. And then you take it to a real editor, or a real CMO, and you discover how much work is still ahead of you.

Alex lived through this from the platform side. When AirOps first launched a basic article template, the workflow had 3 steps. A few months later, he woke up on a Sunday morning to find a customer had built something with more than 100 steps.

“I woke up on a Sunday morning and we had workflows that were a hundred plus steps. And it was one of those moments where the difference between an output and an outcome was just so starkly illustrated on this canvas.”

Every step in that 100-step workflow was a judgment call someone made over months of iteration: how to pull context, which source to trust, how to format a brief, where to insert a human review checkpoint, how to handle tone drift, what to do when the brand voice slips. The workflow was documentation of everything that professional had learned about getting AI content to production quality. It looked like complexity from the outside. From the inside, it was craft.

The Hidden Judgment Calls Inside an AI Content Workflow

• Which sources to trust
• What context to include
• What context to ignore
• Where to add human review
• How to handle tone drift
• When to ask an SME for input
• How to flag risky claims
• What the editor should review first
• What must be rewritten manually
• When the draft is good enough to publish

AirOps now runs a 3 to 4 week onboarding process for new customers. They audit context, work through tone and voice with the content team, and get alignment on quality standards before any automation runs. It’s structured, and it’s detailed. Most AI pilots fail because they skip this work: you get the tool running in a day and assume the outputs will sell themselves. They don’t. 98% of AirOps’s pilots convert to annual customers, and Alex is direct about why: the onboarding process does the real work before anyone has to convince a skeptical editor.

The editors are not the enemy. The system has to meet them where they are. Some editors want to see 4 or 5 briefs and give a 30-second voice memo before the outline is written. Others, once they’ve built trust, are comfortable reviewing a finished draft. Neither is wrong.

The content engineering system has to be flexible enough to accommodate both, and smart enough to learn from their feedback over time. AirOps Next, the platform’s most recent release, is built around this: the more a team uses the system, the more it learns their style and incorporates their rules, even from feedback they never articulate explicitly.

Getting to production-acceptable output is a process of building trust incrementally, with the system improving as the team and the tools get used to each other.

What to Align on Before the First AI Draft

• Brand voice
• Product positioning
• Claims that need approval
• Editorial quality bar
• Review checkpoints
• SME involvement
• Source-of-truth documents
• Examples of good and bad output
• Rules for attribution
• What should never be automated

Key takeaway: Plan for a 3 to 4 week ramp period before your AI content system produces anything you’d actually publish. Use that time to audit your brand context, align on voice with your editorial team, and build quality checkpoints into the workflow before you hit the review stage. Treating it as a quick setup is the reason most pilots fail.

Back to the top ⬆️

Where Human Judgment Has the Highest Leverage in AI Content Workflows

Once a content engineering system is running, a new question surfaces: where should humans actually spend their time? Misplacing human effort is one of the main ways teams end up with outputs that feel simultaneously expensive and mediocre. Alex identifies 2 areas where human judgment compounds the most.

The first is ideation at the top of funnel. AI can suggest hundreds of content ideas. It can analyze the SERP, identify content gaps, and generate angles by the dozen. But the percentage of those ideas that are genuinely good is low, and high-taste users have a lot of critical feedback when they actually look at AI-generated pitches.

“AI is not very good out of the box at ideation. It’s going to suggest a lot of things you could write about, and a very small subset of those by default are genuinely good ideas.”

Where Humans Still Make the Biggest Calls

• Which ideas deserve to exist
• Which angle is actually differentiated
• Which expert should contribute
• Which claim needs evidence
• Which story fits the brand
• Which draft has real taste
• Which content gap matters commercially
• Which piece should be killed before it wastes time

Calling the shots on what gets made, especially for editorial and thought-leadership content, is still very much a human job. The second area is perspective and attribution. AI models do retrieval, synthesizing from what already exists, which means there’s a strong bias toward fresh, specific, attributed content.

If you can identify who internally has the most relevant expertise on a topic and route the right questions to them, the answer they give you is information that doesn’t exist anywhere else on the internet yet. That’s the information gain that gets content cited. The content engineer designs the workflow to capture it. The human provides the perspective. The combination is what performs.

Key takeaway: Identify your top subject-matter experts by topic and build intake workflows that route specific questions to them at the brief stage. Even 2 to 3 sentences of attributed perspective from the right person inside your company can separate a piece from the rest of the content competing for the same AI citation. Build the routing, then protect the experts’ time by making the ask small and specific.

Back to the top ⬆️

Why Context Gathering Is 70 Percent of an AI Content Workflow

The most common mistake in building AI content workflows is spending most of the design effort on the generation step. The prompt engineering, the model selection, the output formatting: these are where attention goes, partly because they’re visible and partly because they feel like AI. The actual leverage is earlier.

“Being very thoughtful about where the right information lives inside the business and outside the business is, I would say, 70% of the work.”

In practice, this means deciding where the right information lives inside and outside the business, setting up the connectors to pull from it, making sure internal content is in a format LLMs can actually use, and structuring the retrieval process so the right subset of knowledge gets attached to the right piece of content. Often the content that exists internally is not ready for use by the models, and needs preparation before the workflow can touch it.

Start with internal sources: a Google Drive folder, Gong transcripts, a shared Notion wiki, whatever the company actually uses to store knowledge. Get that working first, then layer on competitive research. Understanding the ranking landscape for a target query tells you what format is expected, what topics are covered, and where there’s a gap worth owning. Competitive intelligence at the query level is a built-in step in most mature workflows.

Internal Sources to Connect First

• Google Drive
• Notion
• Gong or call recording tools
• Support tickets
• CRM notes
• Product documentation
• Sales enablement docs
• Customer research notes
• Webinar transcripts
• SME Loom recordings
• Slack threads with durable insight
• Existing high-performing blog posts

Then comes the creation process: brief, outline, draft, revisions, feedback, publish. That looks a lot like the old process, but with AI assistance at each step and human review checkpoints built in. Refresh workflows follow the same structure, shorter: a refresh brief specifies exactly what’s changing and why, so the editor knows what to look for.

A Basic AI Content Workflow

• Gather internal context
• Pull competitive research
• Build the brief
• Route questions to SMEs
• Generate the outline
• Review the angle
• Draft the piece
• Check claims and positioning
• Edit for voice and quality
• Publish
• Monitor citations, traffic, and conversions
• Refresh when the facts change

Key takeaway: Before you touch a prompt or choose a model, spend a week mapping your internal knowledge sources. List every place your company’s expertise actually lives: call recordings, support tickets, product docs, internal wikis, expert Loom recordings. Set up connectors to the 2 or 3 most valuable ones and get the retrieval working cleanly. The generation step will take care of itself once the context is right.

Back to the top ⬆️

Why Review Becomes the Bottleneck After You Automate Content Production

There’s a pattern that repeats itself when marketing teams get AI content workflows running. The first few months are exhilarating. Output that used to take a week takes a day. Backlogs clear. Capacity feels unlimited. And then, somewhere around the 3 to 6 month mark, things slow down again.

Alex draws the comparison to software engineering, where AI code generation created a review bottleneck as output volumes climbed. Review is now the constraint for engineering teams: the work of processing output, evaluating quality, and deciding what actually belongs in the codebase. Content teams are discovering the same thing. You can generate 15 articles in a day that would have taken a month 5 years ago. But reading 15 articles critically, evaluating their accuracy, catching positioning drift, and deciding what’s ready to publish is a significant cognitive load. Attention becomes the limiting factor.

“If you’re just creating content that’s ungrounded or it’s off-message or the positioning is weak, you’re just polluting your brand story.”

That’s the risk that doesn’t get enough attention. Brands that scale content production with AI and don’t invest equally in the review layer end up with a content library that’s large and inconsistent, with content that talks about the product 7 different ways, uses stale positioning, or presents the brand differently depending on which workflow produced it. The scale advantage becomes a liability when the brand voice frays under the volume.

AirOps built its new platform around this problem. The review surface, how content gets evaluated, adjusted, and approved before it goes out, is as important to them as the generation layer. Making the review experience efficient and engaging enough that people actually do it, rather than rubber-stamping outputs to move on, is one of the harder design problems in the space.

What Your Review Layer Needs to Define

• What “good” looks like
• Who reviews which type of content
• Which claims require approval
• Which tone issues are unacceptable
• Which sources are allowed
• Which product terms are current
• What gets checked by AI
• What gets checked by humans
• What blocks publication
• What can be fixed after publication

The competitive implications run in parallel. When everyone has access to the same tools, volume stops being the moat. What remains is differentiation: experiential content, original research, video, content with genuine information gain that can’t be replicated from public sources. The teams that invest in review and quality while scaling production are building the output that still performs when the middle tier of AI-generated content gets commoditized.

Key takeaway: Build your review workflow before you scale your production workflow. Define what good looks like for your brand in writing, specific enough that an AI system can flag when outputs drift from it, and specific enough that a human reviewer knows what to look for. Instrument the review step so it’s fast and low-friction. The brands that get this right will compound; the ones that skip it will publish more and perform less.

Back to the top ⬆️

Why Enterprise CMS Integration Is Harder Than the Content Quality Problem

Building a content engineering system that produces great outputs is one problem. Getting those outputs into the places where they actually live: the website, the knowledge base, the partner portal. That’s a different problem entirely. Most people who haven’t tried it underestimate it significantly.

Manual Publishing Steps to Audit

• Copying drafts into the CMS
• Formatting headers and tables
• Adding metadata
• Selecting images
• Adding internal links
• Mapping structured fields
• Updating author information
• Adding schema
• Routing for legal or brand approval
• Scheduling publication
• QAing the live page

Alex has a line for this: “I’ll know we’ve hit AGI when it can post an article to an enterprise CMS.” It’s a joke, but barely. He’s built the tool side of this long enough to know that Webflow integrations are straightforward. Enterprise CMS integrations are not. There are companies with systems that have a 3 to 6 month lead time on adding a new page, organizations where IT policy governs API access, where structured fields are mapped manually, where a headless CMS schema means a human has to think carefully about where every element of the content goes before automation can touch it.

“We actually have 4 dedicated engineers who just integrate with enterprise CMSs. That’s all they do. God bless them.”

The AirOps platform is built in 3 layers to handle the full pipeline. The AI search monitoring layer collects 2 million answers a day from AI search engines, analyzing them for sentiment, citation drivers, and what’s moving in a given topic area. The brand context layer stores visual guidelines, positioning, product information, and proof points, versioned and governed, so content automation runs against a current and accurate fact base. The action layer, recently relaunched as agent-first, is what connects those inputs to actual published outputs, including the last-mile CMS step.

For teams that want to run AirOps entirely through their own infrastructure, the platform exposes every piece through an MCP. Ramp, the financial automation company, runs all of their AirOps workflows through an MCP from their own orchestration system. That’s an acknowledgment that the top 10 to 15% of users are operating in Claude Code, in their own terminals, and aren’t always going to interact through a UI. AirOps has to work in both worlds.

The enterprise CMS problem is a good proxy for the overall operational maturity of a content engineering team. Teams that have solved it have usually built serious infrastructure around their content. The distance between “copy-paste from a text file” and fully automated publishing is a few months of focused work, not a fundamental capability gap.

Key takeaway: Map your content publishing pipeline from generation to live before you build any automation. Identify every manual step that happens between an output being created and it going live on your site. Those are the integration points your workflow needs to eventually handle. Start with the steps that happen most frequently and solve them one at a time before assuming the full pipeline can run autonomously.

Back to the top ⬆️

Why the Agent Runtime Is the Next Competitive Battleground for Content Teams

The conversation around AI operators and the future of SaaS has a tendency to collapse into 2 extreme positions. Either SaaS as we know it is dying because everyone will run agents from the terminal and never log into a UI again, or SaaS will evolve incrementally and the UI stays central. Alex’s position is more nuanced than either, and in some ways more interesting.

He agrees the cohort of AI operators is real. People who run their entire AirOps workflow through an MCP in Claude Code, who pay significant API fees and rarely log into a dashboard, who build their own orchestration on top of platform primitives: that group exists and is growing fast. But content is also a team sport. It involves editors, CMOs, legal reviewers, subject-matter experts. Most of those people are not working from a terminal. The review friction that teams are now discovering as their main bottleneck is not a problem that a CLI experience solves well.

“Our bet is that we can build a better cloud agent runtime than anyone else for our use cases. We’ve been working on that for 6 months.”

What’s emerging as the real question is where agents run, and the answer has real consequences. Cloud agent runtime versus local machine-based agents has meaningful differences: cost structures, observability, multi-party human-in-the-loop capabilities, how agents improve over time, and what data they can access. AirOps has been working closely with Anthropic on this and is making a specific bet: a purpose-built cloud agent runtime for content engineering use cases will outperform a generic one. The fact base, the brand context layer, the 2 million daily answers from AI search engines: these are data assets that a specialized runtime can incorporate in ways a general-purpose runtime cannot.

There’s also a governance problem surfacing among CMOs. When content is being generated by multiple workflows, some in-house and some via platform, and the brand context isn’t governed, you get content that describes the product 7 different ways. Purpose-built SaaS with a structured, versioned brand context layer is positioned to solve this better than a collection of API calls assembled from scratch.

SaaS built for AI operators, with the infrastructure and observability that enterprise content teams actually need underneath the automation, is where this is heading. The teams thinking about agent runtime now will have a structural advantage when the rest of the market catches up.

Key takeaway: Set up a governed brand context source of truth before scaling any content automation: a versioned, shared document that defines your positioning, product facts, tone guidelines, and current proof points. Every workflow you build should pull from it. When the positioning changes, update it once and every downstream automation updates with it. That’s the governance layer most teams are missing, and the one that will matter most as agent runtimes mature.

Back to the top ⬆️

What the Case Against Content Engineering Gets Wrong About the Role

There’s a credible critique of content engineering making the rounds. Ryan Law at Ahrefs published a piece called “I Wouldn’t Hire a Content Engineer,” arguing that the role is hiring for skills that won’t be needed long, and that a great writer with no AI fluency beats a middling writer who knows all the agentic tools. It’s a position worth engaging with directly.

Alex hasn’t read the specific piece, but he’s read the genre: critiques that treat content engineering as automation-for-its-own-sake, as if the goal were to reduce writer involvement. His actual belief about great writers is that they’re invaluable and should be protected. The whole point of content engineering is to create conditions where great writers can do their most impactful work, instead of spending their hours on the parts of the job that automation handles better.

A great writer who goes into a session prepped, knowing which questions they should be answering, having done the competitive research, having context on the current discourse, produces something different from a great writer starting from a blank page.

“I want more people in the loop in content, and 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.”

The deeper issue is what happens to the content library when the best writers are focused only on their best work. There’s a lot of area under the curve that doesn’t get covered. Existing content ages out. Documentation goes stale. The long tail of questions buyers are asking never gets answered. That unaddressed surface area is a marketing liability, with content decaying in place while the writers work on the flagship pieces. Content engineering is the system that addresses the library, not the system that replaces the people writing the flagship pieces.

Humans Should Own	Systems Should Support
Point of view	Research collection
Taste	Content inventory tracking
Original ideas	Refresh monitoring
Narrative judgment	Brief assembly
SME interviews	Question routing
Final editorial calls	Claim checking and consistency flags

The most effective people in every knowledge work discipline right now are adding some systems and technical fluency to what they already do. The mind of an experienced writer has genuinely creative ways to use these tools that a non-writer probably wouldn’t think of: the brand instincts, the judgment about what’s worth saying, the feel for what will actually resonate. Those skills compound when combined with the technical layer. The right answer is pairing, not replacing: great writers doing the work that requires their judgment, with content engineering systems handling the infrastructure around them.

Key takeaway: Evaluate your content team’s current workflow and identify which tasks require real editorial judgment and which are logistical. Research, brief writing, competitor analysis, content inventory management, refresh tracking: these are strong candidates for automation. Reserve editorial decision-making, top-of-funnel ideation, and point-of-view development for humans. The split changes the output quality of both.

Back to the top ⬆️

What a Content Engineering Team Looks Like in 3 Years

Forecasting 3 years in this space feels like guesswork. But Alex’s prediction is grounded in what’s technically possible now, not in speculation about breakthroughs that haven’t happened yet.

His view: marketing teams will onboard a new subteam within the next few years, and that subteam will be captained by a primary agent with a team of specialized subagents underneath it. AirOps is building that team for the content engineering and content operations use case. The human members will be there to react, exercise taste, contribute perspective, and make calls that require judgment, with the logistical work handled by the agent layer.

The working session he describes: a few highly productive hours inside the platform where you arrive and find a set of prompts ready for you. Some ask for your opinion on a positioning call. Some flag a shift in what competitors are doing in a particular topic area. Some surface a content gap that opened since last week. You exercise taste, add a voice note, approve or redirect. The agent layer does the rest.

What the Agent-Led Content Team Might Surface

• A competitor gaining citations in AI answers
• A stale page that still gets traffic
• A product claim that no longer matches positioning
• A new customer question worth answering
• A missing comparison page
• An expert who should contribute to a topic
• A content gap opened by a market shift
• A refresh opportunity tied to conversion data
• A draft that needs human taste before publishing

“I really think most people are doing like 10% of what they could be doing if they really could scale themselves.”

The goal is to make that experience genuinely engaging, because when it’s engaging, more people inside the company want to participate. Product owners contribute perspective. Customer success shares anecdotes. Customers contribute their own points of view. The content footprint becomes a collaborative, living thing with energy and distributed ownership behind it, rather than a burden carried by a small team.

What changes in 3 years is the degree to which these systems are proactive. Most content engineering today is reactive, with teams building workflows that respond to prompts and assignments. The next phase will produce systems that surface the right opportunities before anyone has to ask for them. Trust increases as context and capabilities increase, and the coverage gap between what teams could theoretically do and what they’re actually doing starts to close.

Key takeaway: Pilot a regular content intelligence session with your team now, even if it’s manual. Spend 1 hour per week reviewing: what content is being cited in AI answers, what competitor content is ranking, what customer questions are going unanswered, what owned content is going stale. Build the habit before you automate it. The teams that already have this operating rhythm will adapt to the agent layer much faster when it arrives.

Back to the top ⬆️

How to Stay Current When the Tools Shift Every Few Months

Knowing how to use the tools that exist right now is only part of the challenge. The tools are changing fast enough that what works in 2026 may not work the same way in 2027. That creates a real question for anyone trying to build a durable skill set in content engineering: what’s worth learning, and how do you stay current without going in circles?

Alex’s answer is less about specific tools and more about the practice of experimentation. Doom scrolling industry commentary about which model is best or which workflow is now deprecated feels like learning and produces anxiety. What’s actually productive is treating experimentation as a professional duty, the way other disciplines treat staying current with their field.

“The spread of adoption has never been wider. If you talk to the top 1% of marketers right now and you talk to the median marketer, that gap has never been bigger.”

AirOps has institutionalized this internally with “play” as a core operating principle. Not building sandcastles: following curiosity, trying things, and sharing what works. Alex spends roughly one full weekend day per week building inside the platform, following whatever thread his curiosity is pulling on. That time gives him energy, feeds his thinking, and keeps him current in a way that reading about the tools never would.

The gap between the top marketers and the median marketer is wider now than it’s ever been. The gap comes down to hands-on experimentation. The people at the top are building things, trying things, accumulating reps. Reps are how you stay current when the tools shift, because you develop the pattern recognition to learn new tools faster, to evaluate what matters, and to make judgment calls that no course or tutorial can teach. The specific skills matter less than the practice of building them.

Key takeaway: Carve out one dedicated hour per week to experiment with a tool or technique you’ve been curious about but haven’t tried. Set a specific question to answer with it, something connected to your actual work. Document what you learn. After 4 weeks, share it with someone on your team. The experimenting is the training.

Back to the top ⬆️

How Alex Decides What Deserves His Energy

Most people who run a startup discover at some point that the job is designed to consume everything. The decisions don’t stop, the priorities compete, and the weeks blur together in ways that are hard to account for. Alex has built a few intentional structures around this.

The first is a personal Claude whose job is prioritization. As the business has grown significantly, having a system that’s been given context about his goals, his time constraints, and where he tends to get pulled off course has become a genuine tool for staying on track.

“I actually have a personal Claude whose primary role is to help me prioritize, given all the noise around me.”

He splits his week into 2 explicit categories: need-to-dos and things that give him energy. The energy-giving category gets protected time: specifically, 1 full weekend day per week spent building inside AirOps, following whatever he’s genuinely curious about. That time has a dual function: it keeps him close to the product and it gives him something that isn’t transactional. Following curiosity in a creative, low-stakes environment is how he recovers enough to do the high-stakes work well.

The physical and community side is harder to be consistent about, and he’s honest about it. There have been periods where he’s neglected it, and he’s direct about what that cost him. Running a startup takes a sustained toll in ways that are easy to underestimate until you’re already in deficit. This year he’s being more intentional about it, getting outside when the weather allows, treating wellness as part of the job rather than a reward for finishing it. The West Side Highway in New York just reopened as the weather improved, which helps more than it probably should. The Claude yells at him when he skips his things. The balancing act continues.

Key takeaway: Identify the 1 or 2 activities in your work week that consistently give you energy rather than drain it. Block time for those the same way you block for meetings. Then identify what tends to consume your attention without proportional results, and get something, a system, a person, or a Claude, to help you push back on it. Energy management is the meta-skill that makes every other skill more sustainable.

Back to the top ⬆️

Episode Recap

Alex’s central argument is that content engineering is a discipline built around putting more humans into the content process. The goal of the content engineer is to find the right people inside an organization, route them the right questions, and build their knowledge into a content system that can maintain and scale it. Automated output that recycles what already exists on the internet is exactly what content engineering is built to avoid. The brands performing well in AI search have figured this out. The ones generating volume without that infrastructure are quietly building a liability.

The tactical thread running through the episode is the systems layer that content marketing has never had. A content library treated as a publishing archive decays in place: posts go stale, positioning drifts, the fact base running underneath any automation falls behind the business. The teams winning in AI search have built context layers, refresh workflows, and integration pipelines that treat their content library as a living asset. AirOps’s work with Webflow illustrates what that investment actually returns: 42% more traffic and 6x better AI-attributed conversions, all from refresh rather than new production.

The conversation also surfaced something that doesn’t get enough attention: the friction doesn’t disappear when you automate content production. It moves. Review becomes the bottleneck. The attention of the people doing the review becomes the limiting factor. Brands that scale production without equally scaling their review layer end up with large, inconsistent content libraries that dilute their brand voice at scale. Building the review infrastructure is as important as building the generation infrastructure, and in Alex’s experience, it’s the part teams consistently underinvest in.

There are honest tensions in the episode worth noting. The question of where agents should run, cloud runtimes versus local machines, SaaS platforms versus API-first operator setups, is genuinely unresolved, and Alex is direct that it’s an existential question for companies like AirOps. The enterprise CMS integration problem remains hard in ways that are difficult to automate around.

And the skills question, what junior content marketers should build toward and whether content engineering is a durable role or a transition role, doesn’t have a clean answer. Alex’s bet is that the blend of traditional writing craft and systems fluency will compound over time, but the ratio is still being worked out in the market.

You can follow Alex on LinkedIn and learn more about AirOps on their site.

Full episode ⬇️ or Back to the top ⬆️

✌️

—
Intro music by Wowa via Unminus
Cover art created with Midjourney (check out how)

Apple •Spotify• Pocket Casts •Youtube •Overcast •RSS

Related tags

See all episodes

Future-proofing the humans behind the tech

Apple •Pocket Casts•Google •Overcast •Spotify •Breaker •Castro •RSS

• LinkedIn
• Twitter
• Instagram
• YouTube
• RSS Feed