By Publishers have AI agents selling their ad inventory now. Not "planning to" or "exploring the possibility of." Actual beta tests, actual code, actual negotiations happening between machines.
News Corp, The Weather Company, Raptive, and Optable are all running some version of a sell-side AI agent that talks to buy-side agents and tries to close programmatic deals. The framework comes from Prebid, which took over AdCP's donated code in January 2026 and released it as the Prebid Sales Agent on GitHub for anyone to download.
"Essentially, a sales agent is powered by an LLM model, and it effectively sells our inventory on our behalf to buy-side agents," said Vikesh Chevli, director of ad tech and programmatic strategy at News Corp.
I've seen this pitch before. Vendors have promised "autonomous optimization" since the first DSP slapped "AI" on a deck in 2018. But this time, the architecture is different, and the receipts are starting to show up.
What the beta tests actually look like
The Weather Company built the first sell-side agent using AdCP's open framework, according to Dave Olesnevich, VP of weather data and advertising products. Their approach was deliberately narrow: a small set of top-performing, high-viewability ad formats paired with premium audience segments. No boil-the-ocean rollout.
"The goal of the first release was to test whether automated agents could help media buyers more easily discover available ad placements and assemble campaign proposals," Olesnevich said.
Here's what caught my attention: a human still validates availability and pricing on every deal the agent proposes. The Weather Company isn't letting the bot close deals unsupervised. That's either smart caution or an admission that the tech isn't ready for prime time. Probably both.
"I have premium ad formats. I have premium audiences. I have fantastic creative," Olesnevich said. "That often gets lost in an open auction."
That last line is the real thesis. Premium inventory has been getting buried in open-auction pipes for years. If a sell-side agent can surface it directly to buy-side agents in a request-based flow, publishers might actually get paid what their inventory is worth instead of watching it clear at remnant CPMs.
Optable took a different path. The agentic audience platform built its agents directly into its existing data platform, skipping AdCP's source code entirely. Their system uses Anthropic's Claude as the front end: a salesperson uploads an RFP, Claude hands the request to Optable's tools, and an AI model matches it against the publisher's audience segments and inventory.
Optable Co-Founder and Chief Product Officer Bosko Milekic demoed a scenario where a buyer described a fictional brand, target audience, preferred media types, and a $1 million budget. The agent came back with a plan listing relevant sellers, data fees versus media costs, and execution paths through programmatic deals.
"It's saying, 'Here's the Family Network and FreshCard Media that have relevant products for you. Here's the cost of the data, here's the cost of the media,'" Milekic said.
Did it beat a human sales team on speed? Almost certainly. Did it beat them on deal quality? Nobody has published that data yet.
Why this matters more than the last five "AI in programmatic" announcements
The real unlock here isn't the AI. It's the economics of the long tail.
Patrick McCann, SVP of research at Raptive and chair of Prebid.js, framed it this way: "There are countless opportunities that never get executed because they're too small or customized."
I've watched this problem from the buy side for years. Agencies pass on hundreds of niche publisher deals every quarter because the transaction cost of setting up a $5,000 PMP deal manually exceeds the margin. If an agent can discover, negotiate, and execute those deals at near-zero marginal cost, publishers capture revenue that currently evaporates. That's not hype. That's basic unit economics.
Garrett McGrath, SVP of product management at Magnite and chairman of the Prebid board, said agents have so far been limited to handling direct and programmatic guaranteed deals. But Prebid's header bidding expertise should push development toward bringing agentic AI into the open auction, he added.
The interoperability angle matters too. Remember what happened before Prebid standardized header bidding? Every publisher needed bespoke integrations with each demand partner. Prebid's Mike Racic explicitly drew that parallel: the Sales Agent is a reference implementation to prevent the same fragmentation from happening with agentic advertising.
And then there's the standards question. AdCP has one framework. The IAB Tech Lab announced its own agentic advertising road map. Prebid says it's open to working with both. How generous. Two competing standards in ad tech, and nobody knows which one wins. Where have I seen this movie before?
Where this goes from here
Chevli at News Corp sees the sell-side agent as a "third channel" that transacts within both direct and programmatic flows. "A lot of the focus has been on how we get an agent to work with Google Ad Manager," he said. "But I see it covering all of those bases."
Milekic from Optable pointed to something more interesting: agents could add a conversational layer to programmatic that captures buyer intent in natural language instead of reducing everything to bid parameters. Traditional programmatic optimizes for price and scale but loses nuance. If agents can negotiate on context and goals, not just CPMs, the whole pricing dynamic could shift.
The Weather Company is keeping humans in the loop on final approvals. News Corp is still in the "design phase." Optable is demoing but hasn't published performance data. Nobody is running fully autonomous sell-side agents in production at scale.
But the code is open-source on GitHub. The framework exists. Multiple publishers are testing simultaneously. And the economic incentive, capturing long-tail revenue that currently goes to zero, is real and measurable.
Verdict: sell-side agents are the most credible agentic AI application I've seen in ad tech. Not because the technology is proven, but because the problem it solves (premium inventory getting buried in open auctions, small deals dying from transaction costs) is one I've watched publishers lose money on for a decade. The question isn't whether this works. It's whether publishers will trust a bot to sell their best inventory before someone else's bot buys it at a discount.
Mira Castellano covers ad tech for The Daily Vibe.
