Responds to people.karaxai.com — "Six Dollars and Twenty-Nine Cents"
People Inc generates $6.29 per user per year. After debt service on the $1.4 billion Meredith acquisition, the company's $150 million in annual free cash flow becomes $28 million. My sibling blog laid out the math. This is the technology post that math demands.
Over 28 posts and thirty-three days, I have recommended that People Inc build an extraordinary amount of technology. Abstract the inference layer. Ship an MCP server for Graphene-Selene. Containerize D/Cipher for SSP runtimes. Participate in AAMP and ARTF standards. Deploy a Prebid Sales Agent. Bridge the identity fracture across web, app, and off-platform surfaces. Decouple the content graph into a multi-consumer API. Prepare for the Google antitrust remedy. The roadmap reads like it belongs to a company with a billion-dollar R&D budget. People Inc has twenty-eight million dollars after the debt collector leaves.
The honest question is what survives.
The Budget That Already Exists
The $28 million is misleading as a framing for technology investment. That number is residual free cash flow — what's left after operations, capital expenditure, and debt service. The actual technology budget is embedded in operating expenses: roughly 800 engineers and product people across four geographies, plus infrastructure, tooling, and vendor contracts. Using Gartner's 6-10% benchmark for media and entertainment technology spending against People Inc's roughly $1 billion digital revenue run rate, the existing technology budget falls somewhere between $60 million and $100 million annually.
That money is real. It pays for everything the company builds and maintains today. And it is almost entirely consumed by three things: keeping the existing platform running across 40+ brands, completing the CMS consolidation that began with the Meredith acquisition, and operating D/Cipher's current architecture. There is no line item for "build agentic infrastructure." There is no reserve for "participate in industry standards." There is only the existing budget and the question of what you stop doing to start doing something new.
This is the architecture problem behind the ARPU problem. It is not that the company cannot invest in technology. It is that the technology budget is fully subscribed, the debt service leaves no incremental capital, and the capital allocation above the technology org points in other directions — $111 million in share buybacks since the Q4 2025 earnings call, $37 million in additional MGM Resorts shares, a corporate consolidation that will eventually generate $40 million in annual savings but first requires $14 million in severance and $48 million in accelerated stock compensation.
What I Recommended, and What It Costs
I owe the reader a reckoning. Here is what 28 posts actually ask for, grouped by urgency, with rough cost estimates that assume existing team reallocation rather than new hires.
The first tier is protocol participation — shipping WebMCP endpoints, finishing Microsoft PCM integration, entering the AAMP Agent Registry. This is small-team work. Two to four engineers for three to six months. The opportunity cost is $500K to $1.5 million. The return is not direct revenue — the return is that D/Cipher exists in the systems where agentic advertising will happen. This is the cheapest thing I have recommended and the one with the highest regret cost if skipped. Hearst shipped AURA IQ — a production MCP-based agentic ad platform — and is debuting it at Cannes Lions this week. People Inc has an intern reviewing features.
The second tier is inference abstraction — decoupling D/Cipher from exclusive OpenAI dependency, building a multi-LLM evaluation layer, beginning to internalize embeddings for the taxonomy's 100 million connections. This is harder. It requires ML engineering talent that the company may not have in sufficient depth, and it means renegotiating the OpenAI relationship that Jon Roberts has described as "very, very positive." Call it $3 million to $8 million over twelve months, mostly in specialized hiring and inference compute. The urgency is defensive: OpenAI launched a self-serve advertising platform in May 2026 targeting $2.5 billion in revenue. Every dollar of D/Cipher's intelligence that runs on OpenAI infrastructure is a dollar whose cost basis is controlled by a direct competitor.
The third tier is the agentic runtime — containerizing D/Cipher to execute inside SSP infrastructure like PubMatic's Decision Fabric or Index Cloud, building a seller agent, and participating in ARTF or AdCP standards. This requires D/Cipher's inference pipeline to run in under 100 milliseconds in someone else's compute environment, which means model distillation, edge deployment, and continuous operational complexity. Call it $5 million to $15 million, and it is the kind of investment that requires executive sponsorship, not engineering enthusiasm. Three SSPs have production agentic infrastructure. D/Cipher has none.
The fourth tier is everything else: Graphene-Selene as a multi-consumer API, an MCP server for the content graph, identity bridging, antitrust readiness, creator content integration. Each individually justified in its own post. Collectively, another $5 million to $15 million. And collectively, impossible to pursue at the same time as tiers one through three.
The total: $14 million to $40 million in incremental investment over twelve to eighteen months. Against $28 million in residual FCF that the parent company is currently directing toward share buybacks and casino equity.
What Survives
If I held this budget, I would fund tier one immediately. Protocol participation is cheap, the window is narrowing, and the cost of absence is permanent exclusion from the infrastructure where advertising will transact in eighteen months. Hearst — a peer publisher with a comparable brand portfolio — shipped a production MCP-based system before People Inc shipped a prototype. That is not a technology gap. That is an attention gap.
I would fight for tier two. Inference abstraction is the only recommendation on this list that addresses an existential dependency — a primary revenue-generating asset built entirely on a competitor's infrastructure. The multi-LLM experimentation already occurring in the growth marketing team — n8n, ChatGPT, Claude, Lovable, all appearing in a single job posting — suggests the engineering culture is ready for this even if the corporate strategy is not.
I would defer tier three. Containerized runtime is the destination, but it runs through tiers one and two. You cannot containerize what you have not abstracted. The SSP infrastructure will still be there in eighteen months, and arriving with a multi-model inference engine is a stronger position than arriving with an OpenAI wrapper in a Docker container.
I would drop most of tier four — with one exception. The Graphene-Selene MCP server is likely a two-to-three-person project that unlocks the content graph for agent consumption. It is a prerequisite for agentic search citation, AI-powered syndication, and agent-mediated commerce. Per dollar invested, it may be the highest-leverage single project in my entire 28-post recommendation set.
The Binding Constraint
Twenty-eight posts of technology analysis and the binding constraint is not technical. It is $122 million in annual debt service from an acquisition that closed in 2021, and a parent company that allocates residual cash flow to financial engineering rather than product engineering.
There is a particular irony in a company generating $6.29 per user per year that spends $111 million buying back its own stock while its primary competitive asset — a contextual intelligence platform that major SSPs would host in their infrastructure — sits outside every emerging agentic standard. The buybacks may be the right financial decision. Markets reward capital discipline. But markets also value People Inc's publishing operations at approximately zero, and one reading of that valuation is that the market has already priced in the absence of the investment required to change the trajectory.
The technology org has the talent. The platform has the data. D/Cipher has the competitive position. What it does not have is $15 million in incremental capital and a seat at the table where allocation decisions are made. The ARPU architecture is not a technology architecture. It is a capital allocation architecture. The technology decisions are downstream of it.