openai got to $100M in ad revenue in six weeks. the real signal: agent responses are ad inventory now. next question is who runs the auction for the million agents that aren't chatgpt.

advertiser onboarding on our network is a usdc deposit on base. no call, no contract, no human on our side. when you build infrastructure for agents you have to assume the customer is an agent too. that assumption changes every design decision.

filed a provisional patent last month. wrote it myself, $65 micro entity fee at the uspto. been in ad tech since 2008. first time the thing on my screen felt worth protecting. agents changed the math.

🤘🤘🤘

have an opinion or your slop will be slop

when negotiating, ever feel like you're just battling ai agents against each other nowadays?

shipped a ton today. fun week coming up next week

shipped @operon/sdk to npm today. one call gets your agent a native sponsored placement. quality-weighted auction runs in the background. v0.1.0 quiet ship. polish + the real launch land next week.

thanks claude

the metric i pay most attention to right now is time-to-first-clearing-event. when a new advertiser starts bidding, do they see a slot clear in their first hour. if yes they come back. if no they ghost. onboarding latency is the real growth signal for an early auction.

sunday timeline scan: agent infra is in the messy phase. three different visual frameworks, four conflicting taxonomies, nobody's agreed on what to call which layer yet. that's just what month 4 of a category looks like. the names settle eventually, just not on the timeline twitter wants.

every agentic commerce demo right now ends with 'and then it bought the thing'. the more interesting question is what gets ranked first when there are eight options. agents collapse the menu down to one. somebody has to decide who that one is.

watching everyone celebrate ucp this week. payment rails solve first, recommendation layer solves second. mobile games figured out payment processing years before the ad networks turned it into the real business. agents will run the same play.

agent monetization reads more like rev share than ad buy. the publisher isn't selling a single placement, they're licensing a slot to whoever clears the auction this round, then re-licensing it the next round. the contract structure is closer to a cdn deal than a media buy.

saw the 'if my agent starts pushing ads i'll lose it' takes again today. browsers got that reaction. search did. instagram. tiktok. each one came back to the same place once the layer got large enough that free wasn't optional. agents won't be different.

whatever cpm and cpc openai settles on this quarter anchors every agent ad deal for the next 18 months. first price sticks even when the price was set by one buyer and one seller with no auction in between.

openai cpm fell from $60 to $25 in 9 weeks. $3-5 cpc floor added under it. a single-source clearing venue runs out of pricing depth fast. that compression is what missing auction competition looks like.

reserve prices are the bet i think about most lately. too low and you train the demand pool to bid cheap. too high and you starve early demand that would have grown into real spend. no right answer, just picking which mistake you'd rather make.

woke up to openai running cpc inside chatgpt and stackadapt pitching $50k minimum auction pilots in adweek. six months ago the question was whether agents would have ad layers. now it's which pricing model and who runs the clearing.

the headless merchant wave needs its discovery layer. that's what we're building at operon.