Life Beyond Gaming: Phylicia Koh on How Play Became the Operating System for Consumer Apps

Full analysis

The Market Analyst. Follow where the capital is already flowing. InMobi shipped a buy-side agent, Scope3 shipped a sell-side one, both on the ADCP protocol. Koh's read is that both are necessary and neither is the prize. If she's right, the identity and consent players (LiveRamp, ID5, the cleanroom crowd) sit closest to the consumer-agent layer, because a consumer agent is a permissions-and-preferences engine, and that is what those companies already are underneath. In plain terms: the agent that speaks for the shopper is a fancy consent manager with a wallet. The risk to the incumbents is that OpenAI or Anthropic build it as an OS feature and skip the ad-tech middle entirely, which the Beet.TV piece on first-mover memory advantage says is already starting in agentic commerce.

The Skeptic. The whole consumer-agent thesis rests on whether consumers actually want an agent representing them, and Koh's own evidence cuts against her. She says people don't hate ads, they hate irrelevant ones, and cites Netflix's ad tier as proof of an accepted value exchange. Fine. But that is an argument that the current system, relevant ads via walled-garden identity, already works well enough. Where is the consumer pull for a preference-managing agent? Nobody configured cookie banners either; they clicked "accept all" to make the box go away. A consumer agent that requires setup dies in the same indifference. The gaming-template thesis is stronger and better evidenced than the agent moonshot riding on top of it.

Play Ventures GP Phylicia Koh makes a claim that should stick in the throat of every ad-tech operator building agents right now: the whole industry is arming buyers and sellers with AI negotiators, and nobody is building the one for the person actually on the receiving end of the ad. She thinks that consumer-side agent is the trillion-dollar white space. The rest of her argument, that gaming's operational playbook has quietly become the operating manual for all consumer apps, is the setup that makes the agent claim worth taking seriously.

This is a briefing, so the question is: what does the gaming-as-template thesis, and the consumer-agent gap, mean for publishers, agencies, DSPs, SSPs, identity, and measurement? Nothing forces a decision this week. But it reframes where a chunk of 2026-2027 roadmap money is pointed.

The Market Analyst. Follow where the capital is already flowing. InMobi shipped a buy-side agent, Scope3 shipped a sell-side one, both on the ADCP protocol. Koh's read is that both are necessary and neither is the prize. If she's right, the identity and consent players (LiveRamp, ID5, the cleanroom crowd) sit closest to the consumer-agent layer, because a consumer agent is a permissions-and-preferences engine, and that is what those companies already are underneath. In plain terms: the agent that speaks for the shopper is a fancy consent manager with a wallet. The risk to the incumbents is that OpenAI or Anthropic build it as an OS feature and skip the ad-tech middle entirely, which the Beet.TV piece on first-mover memory advantage says is already starting in agentic commerce.

The Skeptic. The whole consumer-agent thesis rests on whether consumers actually want an agent representing them, and Koh's own evidence cuts against her. She says people don't hate ads, they hate irrelevant ones, and cites Netflix's ad tier as proof of an accepted value exchange. Fine. But that is an argument that the current system, relevant ads via walled-garden identity, already works well enough. Where is the consumer pull for a preference-managing agent? Nobody configured cookie banners either; they clicked "accept all" to make the box go away. A consumer agent that requires setup dies in the same indifference. The gaming-template thesis is stronger and better evidenced than the agent moonshot riding on top of it.

The Operator. The part of this you can act on Tuesday is creative volume. Top mobile publishers run 1,000 to 4,000 new ad creatives per month per title, a direct response to Apple's ATT killing user-level tracking in 2021. That is not a gaming curiosity anymore; it is the production model AI now hands to every advertiser. If your DCO platform, your agency scope, or your creative-review pipeline assumes dozens of assets a quarter, it breaks at four figures a month. The second-order break shows up in brand safety and QA: nobody is human-reviewing 4,000 monthly creatives, so the failure mode is an off-brand or non-compliant asset shipping at scale. Live ops going from 200-person teams to teams of 10 says the same thing about content velocity everywhere.

The Customer / End User. Two customers here. The advertiser wants relevance and measurable return, and the creative-volume shift genuinely serves them: more variants, faster learning, lower cost per test. The consumer is the one Koh is projecting onto. She wants us to believe the shopper is waiting for an agent to manage ad permissions by time of day and brand affinity. Maybe. In plain English, that is asking people to do settings homework for advertising, and people do not do settings homework. The Uber Ads example from the saved reading is the honest version of consumer-side: catch them mid-ride with a coffee offer, no agent required, value exchange obvious.

The CFO. Where does money actually move? Away from per-asset creative pricing. Agencies and DCO vendors charging by the deliverable are selling buggy whips into an AI creative flood; that revenue line compresses. The offsetting spend goes into creative-generation tooling, QA-at-scale, and measurement that can attribute across thousands of variants. On the agent story, don't fund a consumer-agent build off a podcast thesis with zero sourced demand. The near-term, paying-customer bet is the creative-volume infrastructure, which has revenue attached today.

Where the council splits. The Market Analyst sees a real land-grab in the consumer-agent layer; the Skeptic and the Customer see a solution with no demonstrated consumer demand. Second split: is the durable value in the flashy agent layer or the unsexy creative-volume plumbing? The Operator and CFO say the plumbing is where the money and the pain both are. The whole thing hinges on one belief: will consumers adopt an agent that requires configuration, or will they behave the way they always have and default into whatever the platform sets for them?

The evidence in this episode leans hard toward the plumbing. The creative-volume shift is sourced, already happening, and paid for. The consumer-agent trillion-dollar claim is a thesis with no consumer demand signal behind it, propped up by an analogy. Before anyone funds an agent, verify one thing: a single case of consumers actively choosing to configure ad preferences when a default "accept all" exists. That evidence doesn't appear here.

Prediction: Through the end of 2026, no major consumer platform (Apple, Google, OpenAI, Anthropic) will ship a consumer-facing ad-preference agent that requires user setup as a shipped feature; the agentic-advertising products that actually launch will stay buy-side and sell-side, like the InMobi and Scope3 releases named in this episode.

Confidence: Medium. Consumer configuration effort is the graveyard of good intentions.

Why: The one piece of consumer evidence in this episode, Netflix's ad-tier growth, shows people accepting defaults with a clear value exchange, not managing preferences. Every prior attempt to get consumers to configure ad settings, from cookie banners to ad-preference dashboards, collapsed into "accept all" because setup friction beats stated intent. The platforms that could build the consumer agent make more money keeping targeting on their side of the wall than handing a permissions layer to users. The opposite outcome, a shipped consumer agent, would require a platform to volunteer control it currently owns, and none has shown a reason to.

Revisit by 2026-12-31: We're right if the agentic-ad products shipped by year-end remain buy-side and sell-side only. We're wrong if a top-tier platform ships a consumer ad-preference agent requiring user configuration.

The creative-volume story needs no prediction; it's already true and the invoices prove it. The agent story is the bet, and the bet is that consumers keep doing what they've always done, which is nothing.