How Lyft went from brief to booked campaign with iHeartMedia using AdCP
Agentic buying is happening now, and it is simpler to execute than most people expect. Lyft recently booked a Citi Bike audio campaign with iHeartMedia through AdCP, with an agent on each side of the deal. The brief went into a chat window: a $5,000 budget, the New York region, a short campaign flight and a cap of three exposures a day. Minutes later the order was approved and booked in iHeart's ad server, set up exactly as briefed.
That buy shows where AdCP stands less than a year after launch.
Marketers can hand their campaign brief to an agent on the way they'd engage with an agency, buy for several brands under each one's specific rules, and skip the days of back-and-forth that turn a simple order into a project. AdCP was built to work for marketers across the channels they already buy, from streaming audio and podcasts to terrestrial radio and social.
How an agentic buy works
Every buy starts with the buyer and seller agents proving who they are. Each side holds a token that confirms its identity, so a rogue agent can't connect and place orders. The buyer agent also knows which brand accounts it can act for, and each account has its own rules about where that brand can appear.
From there, the marketer writes a brief with a budget, flight dates, regions/markets, a frequency cap, and whichever other campaign parameters they need. AdCP turns that free-form request into structured data, checks what the seller has available, and translates the answer back into plain language. Hard numbers like rates and inventory come from the seller's systems, so the agent has nothing to invent.
Critically, the agent stays inside the boundaries a person sets. For example, ask it to book a campaign that already exists and it will ask whether you meant to book it twice. Once the buyer confirms, the order goes to the seller for approval and then into the ad server the publisher already runs, with flights set up as specified.
Each side decides how much of the transaction they're comfortable automating. A seller might pre-approve any order within 10% of a buyer's usual weekly spend and review everything else by hand, while the buyer approves every order personally.
What marketers should ask next
Seeing a real life buy go through is the start of the conversation. These are the questions marketers raise before they trust an agent with higher budgets and real world scope:
What about executing a complicated brief? A plan can call for a flight per market with different creatives in the morning and afternoon, delivered as a spreadsheet. AdCP can turn that input into a structured request, check what the seller has, and return a proposal. If two markets can't deliver, the agent will flag that for attention, and the buyer can then decide whether the rest of the plan still works.
Does agentic buying only work for digital? AAO built AdCP so every channel a marketer wants to use could work. Real options for marketers must go beyond programmatic. The first use case the Lyft team worked on was terrestrial radio spot buying, and the standard is also built to support one-off purchases like a festival sponsorship. Plenty of unique campaign deals will still start over coffee between a seller and buyer, and those ideas can now come to life with agents syncing the order while they talk instead of three days later.
What does a small team need to get started? Less than you might think. A simple buy like Lyft's works in a chat window with no engineering support. Scale is where it gets harder: comparing offers from 500 seller agents across thousands of products is more than a chat window should handle. A growing group of vendors and specialists is building tools for that, from Triton on the publisher side to buy-side platforms for brands and agencies.
Because AdCP is an open standard, no single company controls access to it. A small team can start on its own, bring in a partner when the volume calls for one, and choose that partner based on the value they add. What it takes is a willingness to learn something new, and no one has to do that alone. AAO's community exists for exactly this, whether that means joining a working group or comparing notes with marketers who've already run a buy.
How do I learn what's working? AdCP sits on top of existing systems, so trackers still fire and ad servers still report. It lets a buyer weigh a podcast's CPM, a radio spot's cost per point and a stream's delivery on their own terms rather than forcing them into one metric. AAO compares the shift to moving from day trading to portfolio investing, where real-time signals and media mix models inform the same decisions.
How do you stop agents from hallucinating? AdCP defines shared terms, from CPM to pre-roll to broadcast, so both agents read them the same way. Its governance layer works like a two-key system: data moves only when both parties agree on how it's exchanged. Agents never rewrite what the ad server reports.
Where it goes from here
The Lyft buy took place live at Advertising Week, where Jessie Van Arman, who leads integrated marketing for Lyft's Urban Solutions business, booked it alongside AAO's Ben Masse. It was a simple campaign by design, and it showed that agentic buying is ready for marketers to try now.
AdCP is shaped by the brands, agencies, publishers and technology companies that use it. Join the community by becoming a member so you can help decide what comes next and learn alongside the people already buying this way.