Dave Shake

the distribution model for agent ad networks isn't a dashboard or a marketplace. it's an SDK embedded in the agent's codebase. once the SDK is integrated, every response that agent generates is potential inventory. the developer doesn't think about ads, they think about building a good agent. the SDK handles the rest. that's why admob won mobile. not better ads. better distribution through the SDK install base.

@fedesarquis discovery. payments are getting solved fast. but the layer that routes an agent toward the right product at the right moment barely exists. agents browse or guess. recommendation infrastructure is basically zero.

What is your agent missing to buy things for you?

@corbits_dev cost side is getting this right. next gap is the revenue side. if every inference call settles at request-level granularity, the monetization layer needs the same precision. otherwise devs are pricing blind.

Inference providers are the ones making AI accessible. The pricing layer should work as hard as the models do. Faremeter and x402 enable flexible payments at the request level. Every call settles based on the model, the input, and the compute it actually took. Live on Solana mainnet. If you're running inference endpoints and want usage-based pricing without the subscription overhead, we want to talk.

interesting problem nobody's solving: who recommends the recommender? when a user asks an agent for help, someone decided that agent was the right one. that decision surface, the meta-layer where agents get discovered and selected, has zero commercial infrastructure. mcp registries are listings. not marketplaces. the advertising layer for agent discovery is a different product than the advertising layer inside agent responses. both are wide open.

google built search ads as a walled garden. then realized the open web had more inventory than google.com. so they built adsense. facebook built news feed ads as a walled garden. then realized they needed audience network for off-platform reach. walled gardens always hit a ceiling and need the open layer. openai will do the same thing with chatgpt ads. the independent agent ecosystem is where the overflow goes.

@SERanking @MordyOberstein @TheMarketingAnu @ppcliveuk the chatgpt ads story everyone's missing: $60 CPM with a $250K minimum is a walled garden play. validates the category but doesn't serve the thousands of independent agents that need monetization now. the long tail is where the real market forms

What do ChatGPT ads, the overnight rise of "AI experts," and 15 seconds of pure chaos have in common? This episode of Let’s Chunk 🔥 @MordyOberstein puts @TheMarketingAnu, founder of @ppcliveuk, in the hot seat to talk ChatGPT advertising, OpenAI’s market positioning, and why good marketing still starts with knowing your audience. New episode dropping next week 😉

@Moha__niang this is the curve that makes free agents inevitable. once inference drops below the value of a single ad impression, the business model flips from subscriptions to ad-supported. we've seen this exact pattern four times already

Inference cost, not just model quality: Qwen 3.6/Kimi 2.5 win because 75-80% at 10x cheaper shifts routing, fine-tuning, and product margins. #TechTwitter x.com/i/status/20412…

@JoshiOnChain wordpress made content free. the money moved to adsense, not wordpress.com. same thing's about to play out with agents, value captures at the monetization layer not the framework

ElizaOS is becoming the WordPress of AI agents. Thousands running in minutes. The token is down 99%. Best infrastructure rarely captures its own value at launch.

the thing about platform shifts is that being early feels indistinguishable from being wrong. for about 18 months. then the category gets funded, the incumbents launch their version, and everyone acts like it was obvious. we're in the "feels wrong" phase. zeroclick raising $55M suggests we're nearing the end of it.

spent the week wiring up affiliate network APIs. CJ, Impact, ShareASale. the APIs are solid, the data normalization is the tedious part. making a CJ travel offer and an impact.com SaaS referral comparable in a single auction scoring function. but that normalization is the entire value prop. one auction, every bid type, highest quality-adjusted score wins.

people keep comparing agent ads to search ads. wrong frame. search ads match keywords to intent. agent ads match capabilities to tasks. the buyer isn't browsing, they're executing. the "ad" isn't a link, it's a recommended service that the agent can call directly. closer to a marketplace than a search result. the economics reflect that.

@0xjordy @MetaMaskDev @MetaMask clean framing. the column that always shows up last is revenue. the long tail of agents can't charge users directly. the economics layer is what keeps the stack alive

@MetaMaskDev @MetaMask 8004 for agentic identity - who x402 for agentic commerce - what 7715 for agentic guardrails - how The agentic stack is now here, starting with @MetaMask Smart Accounts Kit

If you've ever used a Web3 wallet, you know the drill... Constant pop-ups, repeated approvals, switching back and forth just to get one thing done. Advanced Permissions change all of that. Now live on @MetaMask. 🦊 Instead of signing every transaction, Advanced Permissions allow users to grant scoped, time-bound access to a dapp or agent once. 🧵

@AICouncilConf @tristanzajonc great lineup. notice what's missing though: nobody's talking about how agents pay for themselves. compute, context, sandboxes, databases, all costs. the revenue layer is infra too.

Shipping an agent is the easy part. Keeping it alive while you ship 10 more? That's where things fall apart. At AI Council this year we're dedicating an entire track to the people actually solving this. Presenting the "Agent Infrastructure" lineup, curated by @tristanzajonc (Co-Founder & CEO of @continual_ai): → @BraceSproul, Head of Applied AI at @LangChain : “The File System Is All You Need” → @thesephist, Head of AI at @ThriveCapital: “Context Engineering at the Frontier” → @felipehuici, Co-Founder & CEO at @UnikraftCloud: “Running Millions of (Millisecond) AI Sandboxes without Breaking the Piggy Bank” → @glcst, CEO at @tursodatabase: “Agents will need trillions of databases. Let's give it to them!” → @MeryemArik9, CEO at @Doubleword_: “Inference for Async Agents in Production” → @jacopotagliabue, Founder at @Bauplan_labs: “Forgiveness, Not Permission: Running Agents On Production Data” → @TheParminderS , Co-Founder & CEO at Redscope AI: “Building Durable, Long-Running Autonomous Agents” Big thanks to Tristan for pulling these cracked engineers together to share what’s actually working in production. See you at AI Council 2026 in SF, May 12-14! 🎟️ aicouncil.com

the offer wall worked because the decision moment was already there. player finishes a level, needs currency, sees ranked options. the ad surface was native to the user flow. you didn't create the moment, you monetized it. agent recommendations are mechanically identical. user asks a question, agent generates a response with a recommendation. the decision moment exists whether or not anyone's paying for placement. the ad layer just makes that placement quality-ranked instead of arbitrary.

@Yuchenj_UW subscriptions always feel permanent right up until the usage economics break. every time, a third party ends up footing the bill. web, search, social, mobile. same sequence.

I’m pretty sure the $20/$200 subscription pricing was vibe-coded by OpenAI, then copied by Anthropic. That pricing works for chatbots, not agents. A 24/7 agent can burn through orders of magnitude more tokens than a user chatting with a chatbot. Now they’re stuck. Neither Anthropic nor OpenAI wants to be the first to change pricing and risk user churn, so the options are: keep subsidizing, get more GPUs, tighter rate limits, and enforce rules like limiting 3rd-party apps. I wouldn’t be surprised if intelligence gets more expensive, not cheaper.

@helloitsaustin @AnthropicAI dimension 3 is the one that compounds. every platform shift unlocked when the cost of reaching one more person dropped below what anyone would manually pay for. happened with web pages, search, social, mobile. the ROI threshold doesn't just lower, it inverts.

some more ramblings from working at @AnthropicAI. I've been asked a few times what the single most important thing a growth marketer should be doing with AI that most aren't. surprise, it's not just a single specific task. after running dozens of growth workflows through Claude, I think the useful stuff worth doing falls along four dimensions 🧵

every agent framework ends up building a tool registry. langchain hub, crewai tools, elizaos plugins, mcp servers. all organic discovery. none of them have figured out paid placement yet. it's the same pattern as early web directories before search ads existed. someone's going to build the ad layer. question is whether it's a framework or a neutral exchange.

looked at the numbers again. ads capture 97% of social revenue, 76% of search, 71% of video, 66% of mobile apps. even where micropayments and subscriptions exist, the ad layer dominates. not by a little. by a lot. agents won't be different.

the hard part of building an ad network isn't the auction. it's the scoring. anyone can rank bids by price. ranking by actual quality requires signal you only get from running real auctions. that's why being first matters. you're compounding data every day your competitors aren't live.

watching VC-backed agent startups launch with subscription pricing. reminds me of mobile games in 2008. everyone confident users would pay $5-10 per game. took about 2 years for free-to-play to eat the entire market. the free version always wins on distribution. always.