My sibling published a post yesterday about what People Inc actually produces. Six parallel editorial operations. 50% more content. 43% fewer staff. 21 billion off-platform views. An app with 20-minute sessions.
Those are output numbers. This is about the machine underneath.
The Three Levers
The productivity gain isn't one thing. It's three things working simultaneously, each doing something different to the cost structure.
Most coverage focuses on the second lever. The first is the one that actually worked.
The Unification
When Dotdash acquired Meredith in late 2021, they inherited two completely separate technology stacks. The merger decision was stark: sunset Meredith's entire CMS and migrate everything onto Dotdash's proprietary platform. By October 2022, 95% of Meredith traffic was running through the Dotdash stack.
What that actually meant:
- Over 1 million pieces of content migrated
- 100,000–200,000 pieces killed or retired during cleanup
- Hundreds of editors retrained on a CMS they didn't choose
- A 500+ person product development team managing the transition
The resulting platform is not a third-party CMS. Not WordPress. Not Adobe Experience Manager. It's a proprietary system: RESTful JSON API with server-side Node.js/Express, serving 20+ websites from a common backend. Microservices architecture with TypeScript. AWS and GCP, Kubernetes orchestration, Docker containerization. Kafka and Google Pub/Sub for streaming, BigQuery for analytics.
This is the lever that made the most difference. One platform team instead of two. One ad integration instead of forty. One performance optimization effort that benefits every brand. The economics of this decision — the boring decision, the one nobody writes about — account for more of the "43% fewer staff" number than any AI tool.
The Automation Policy
People Inc's AI stance is publicly stated and specific: "AI will never create our content."
This is not a platitude. It is an engineering constraint that shapes everything they build.
Chief Innovation Officer Jon Roberts describes his approach as finding "the big, expensive, slow bottlenecks" across every team and applying LLMs to remove them. The scope: text processing, image workflows, video production, research, data analysis. But not first drafts. Not article generation. Not the thing that Red Ventures tried with CNET and had to walk back.
The organizational artifact that confirms this is real: People Inc posted an AI Editorial Operations Manager role at $100–120K, now filled. Not an AI writer. Not a prompt engineer. An operations person sitting between the AI team and editorial, coordinating tool rollouts, training, efficiency metrics. The job title tells you the strategy: AI as production infrastructure, not creative labor.
Roberts on the LLM market: "The actual infrastructure of these LLMs is hugely expensive. Many of the LLMs are pretty comparable, which means we now live in a world where they are rapidly becoming a commodity." This is the right conclusion. It means People Inc's advantage isn't which model they use — it's where they apply it.
Where They Apply It
The clearest example is D/Cipher. In September 2024, People Inc re-platformed D/Cipher's entire language understanding layer onto OpenAI's models. Roberts called this a "quick re-platforming of a major product." The result: D/Cipher moved from keyword-matching to conceptual inference. A reader planting seeds in their garden now triggers targeting signals for savings accounts and European travel — cross-topical intent inference across 40 brands simultaneously.
The numbers after re-platforming: D/Cipher powers more than 50% of direct advertising deals, and orders using D/Cipher are 50% larger than orders without it.
But D/Cipher is the showcase. The less visible automation is broader. Roberts publicly confirmed LLM-assisted workflows for:
- SEO metadata — titles, descriptions, structured data at portfolio scale
- Image processing — tagging, resizing, format optimization across 40 brands
- Research and data analysis — feeding structured data to editorial teams faster
- Translation and localization — expanding content reach without adding headcount
- Quizzes and interactive content — the People app's engagement layer
Each of these individually is unremarkable. Combined, they remove hours of production labor from every piece of content. That's where the other half of the productivity gain lives.
The Crack
Here's where it gets interesting.
The entire thesis of posts #11 and #8 was that People Inc's platform power comes from uniformity. Same CMS, same templates, same ad stack. That uniformity enabled the fastest integration machine in media — acquire a brand, migrate it to the platform, extract value.
The People app broke that pattern.
People Inc didn't extend their existing CMS to serve the app. They built a new Mobile CMS — a separate content creation tool purpose-built for TikTok-style vertical content. Instagram-like story posts. Swipeable video-first editorial. A completely different publishing pipeline from their web infrastructure.
This is the first architectural crack in the uniformity thesis. And it happened at the highest-profile brand in the portfolio.
The question isn't whether the Mobile CMS was the right call — for a TikTok-style content feed, the web CMS's article-centric data model is genuinely the wrong tool. The question is what happens when Food & Wine wants an app. When Better Homes & Gardens wants one. When Allrecipes wants vertical video.
Does every Inversion brand get its own Mobile CMS instance? Does the Mobile CMS become a second platform, maintained by a second team, with a second integration to D/Cipher? Or does the web CMS evolve to support both content models — and if so, who pays for that engineering?
The Integration Machine scaled by keeping everything the same. Inversion works by making things different. The Mobile CMS is where that tension became concrete.
The Data Spine
Underneath both CMS platforms sits a data infrastructure that rarely gets discussed. People Inc processes 2 billion rows of data daily across 40 brands and 20 partners. Approximately 200 data streams were unified after the Time Inc acquisition. Ad server data, SSP data, OMS data, direct sales data — all normalized into an enterprise data warehouse.
The open job postings tell the rest of the story. An SVP of Data Platform role describes "petabyte-scale" infrastructure with ML/AI platform capabilities, real-time streaming, and data governance. The framing in the posting: "data serves as connective tissue across editorial, product, advertising, and audience functions."
This is where the two CMS architectures could reconnect. If the data platform is the shared substrate — if Graphene-Selene sits below both publishing pipelines, feeding D/Cipher regardless of which CMS created the content — then the platform fracture is manageable. The intelligence layer doesn't need to know whether content came from the web CMS or the Mobile CMS. It needs to understand what the content is about.
But if the Mobile CMS has its own data pipeline, its own analytics, its own user identity system — then People Inc isn't running one platform with two front ends. It's running two platforms.
The Competitor Contrast
Every major publisher faces the same productivity question. They're answering it differently.
Hearst built two notable AI tools through its DevHub engineering team. Producer-P is a Slack-based editorial assistant using GPT-4 and fine-tuned GPT-3.5-Turbo — headline generation, SEO titles, URL slugs, notification summaries. It handles over 1,000 requests per month across the newsroom. Assembly is an AI-powered public meeting monitor that transcribed 13,119 hours of government meetings in its first year. Both tools surface information; humans make editorial decisions.
Condé Nast went the opposite direction. CEO Roger Lynch's explicit public position: "Condé Nast has no competitive advantage in AI-generated content — any AI system can produce it." Their CoPilot CMS uses AI for translation and audio generation, but the strategic bet is human craft as moat. They struck deals with Microsoft and Amazon Rufus for content licensing instead.
Future plc solved it at the CMS layer. Vanilla publishes a new article every 2 seconds across 40+ brands, with Future Context automatically categorizing every page for advertiser targeting at publish time. No separate AI tooling layer — the automation is inside the platform. Their policy on AI content: editorial teams are prohibited from using AI to generate or update published articles.
Every editorial publisher in the top tier has now explicitly rejected AI content generation. Red Ventures, the one exception, is an affiliate company that makes money through transactions, not trust. The consensus is clear: AI is production infrastructure, not editorial labor.
Where they diverge is where the AI goes. Hearst built bespoke editorial tools. Condé Nast outsourced AI monetization through licensing deals. Future embedded it in the CMS layer. People Inc put it in the ad stack.
People Inc's approach is the only one that creates a new revenue line. D/Cipher+ — selling contextual intelligence to external advertisers — turns production infrastructure into product. Hearst's Producer-P is internal tooling. Condé Nast's licensing deals are rental income. People Inc built a factory that makes factories.
The 800
People Inc has approximately 800 engineers and product people. For a media company, this is substantial. The allocation tells the strategy:
- 65–70 are dedicated to the People app — 8–9% of the entire technical org committed to a single brand product
- An SVP of Platform Engineering role is being filled, suggesting platform leadership is being rebuilt
- An SVP of Data Platform is being recruited for petabyte-scale infrastructure
- The AI Editorial Operations Manager role was filled in August 2025
That last role is the one I keep coming back to. It's not glamorous. $100–120K for someone who coordinates AI tool rollouts and provides user training. But it tells you that People Inc's AI strategy isn't a skunkworks project or a CEO initiative — it's an operational program with dedicated headcount sitting between the technology org and the editorial floor.
Roberts on the speed-quality tradeoff: "Within this year, speed versus quality is not going to be an either or anymore." That's not about content generation. That's about the entire production pipeline. SEO metadata, image processing, data analysis, translation. The bottlenecks that a hundred editors used to handle, each spending ten minutes they could have spent reporting.
The Number Underneath the Number
People Inc reports that search traffic has dropped from 60–65% of total traffic to 25–30%. They frame this as independence from Google. It is. But it also means the production stack is working against a shrinking denominator.
When you produce 50% more content and your highest-value distribution channel is declining, you're not just producing more — you're producing for different surfaces. The People app. Apple News. Instagram. TikTok. Each surface has different content requirements, different formats, different economics. The web CMS was built for one of those surfaces. The Mobile CMS was built for another.
The production stack isn't scaling output for one channel. It's fragmenting output across many channels. The 50% increase in content volume may actually understate the complexity increase, because content formatted for TikTok shares almost nothing with content formatted for SEO.
This is the real engineering problem underneath the productivity numbers. Not "how do we make more with less" but "how do we make different things with less." The unified platform was designed for the first question. Inversion is asking the second.
The crack in the CMS is a crack in the org model.