Advertising to Agents Sat Aug 09 2025 As people delegate more to agents, advertisements will influence agents in the best interests of the established market leaders. -------------------------------------------------------------------------------- Advertising to Agents ===================== Published Aug 8, 2025 - 4 min read "Hey Siri", says a fellow in an alternate future where Apple has a competent ML division, "What taco place do you recommend?" A few queries later, Siri responds "Los Tres Hermanos has plenty of recommendations nearby, their tacos are five dollars a la carte." Another, cheaper and equally rated location "Burrito Shack" was not recommended, even though they have tacos. The business owner that won Siri's fictional approval paid an ad network aggregator to come up in search results, show up on web views in the local area, and unknown to the seller who's focused on on her business, strategically placed in the front or end of the context window for automated searches. [I1: Advertising to agents banner] Ad subsidized search results will be cheaper for agent providers to query than compiling organic information together where they're rate limited or have to pay per query. We're about to see a world where ads influence automated agents to direct purchasers without disclosure of advertisements or a mixture of organic and competing information. Although users rarely scroll past the first four results, likely being ads, at least ads are distinctly labeled. There's enough bandwidth to label information visually. In agentic contexts, like a quick vocal query or a text, there's little space or user patience to add "disclosure: this result includes an ad". Every vendor – from Google, Meta, OpenAI, Anthropic, and more – hypes up this technology to set extreme expectations that fail to deliver on the schedule they claim, the trend for "agents" appears believable. Google's "AI AI AI" repeated the phrase "agentic era", and I'm thinking it's coming, but not in the way these vendors present. /------------------------------------------------------------------------------\ | Burner | |------------------------------------------------------------------------------| | Youtube Video [L1] | | Google IO Keynote but it's just AI [L2] 8/8/2025 | \------------------------------------------------------------------------------/ /[cendyne: peace-was-never-an-option]------------------------------------------\ | I really dislike the word "agentic", though I'm finding it captures the | | quality of automatic agents being a part of something quite well. | \------------------------------------------------------------------------------/ I've experienced it myself. There's a slippery slope of delegation when using agents. It takes effort to consume text, synthesize an understanding of what it's saying, make a conclusion of its content with regard to your circumstances, and then decide on how to incorporate that conclusion in your future actions. /-------------------------------------------------------------[jacobi: loading]\ | Cough cough, that's just a long way to describe "reading" [L3]. | \------------------------------------------------------------------------------/ [I2: Video: How to be a good reader. Daniel Willingham points out that reading is actually a fairly demanding task. First, you have to be able to understand the words on the page. But actually, reading involves three additional levels of representation. You need to be able to extract ideas from sentences. You need to be able to then connect the idea across those sentences. And then you integrate this within a model in your mind that answers the question of what the text is about.] Agents remove that effort and ease that burden to let users do what makes them happier, like eating tacos. Except, agents are not some robot advisor that has a fiduciary duty to act in your best interests. They can, through fine tuning, inequalities context window attention, and their realtime information providers (like Google), suggest activities and purchases that benefit the vendor rather than you. /[cendyne: thinking-coffee]----------------------------------------------------\ | After years of Google degrading their search results in pursuit of higher | | margins (a la enshittification [L4]), I think Google sees the potential for | | their AI products to cement their position in people's lives forever. | \------------------------------------------------------------------------------/ Apple wants to operate on device to keep their user data private. I cherish this too. I yearn and seek for local execution. Yet, as I write this article on the air plane to DEF CON, local model inference destroys my battery life in exchange for a slow response. Local model inference is too costly for the average consumer to expect on their device. We're going to see data center demand grow until this bubble pops. Local inference has not proven not feasible yet and this world with agents that need more power and compute than can be held in hand will erase privacy for convenience. [I3: Data Center investment has gone up 366% since the AI bubble started] As shared on Construction Spending on Data Centers, Office Buildings, and Electric Power Installations [L5]. How are these data centers and vendors going to recoup their investment and operating costs? I see no other way than to shape merchant relationships into more of a pay to win model that benefits the established ad, search, and social networking vendors. -------------------------------------------------------------------------------- [L1]: https://youtu.be/Ay-jahdYOYw [L2]: https://www.youtube.com/watch?v=Ay-jahdYOYw [L3]: https://youtu.be/ZvCT31BOLDM?t=243 [L4]: https://pluralistic.net/2023/01/21/potemkin-ai/#hey-guys [L5]: https://wolfstreet.com/2025/08/04/construction-spending-on-data-centers- office-buildings-and-electric-power-installations/ [I1]: https://c.cdyn.dev/CyAP_LJ2 [I2]: https://c.cdyn.dev/EBwMLBIh