When one ad agent buys from another: What Magnite and LG mean for marketing in the Mittelstand

Until now, buying television advertising has worked like this: an agency receives a briefing, requests offers, negotiates, books, and later checks whether the spots ran as agreed. In between are many steps in which people carry data from one system to the next. On October 14, 2026, the technology company Magnite and LG Ad Solutions, the advertising arm of the manufacturer LG, published an announcement that is meant to shorten one of these paths: a software agent on the buyer side talks directly to a software agent on the seller side.

The news comes from the US and concerns the purchase of ad placements on the home screen of smart TVs. For a Mittelstand company in Germany, Austria or Switzerland, it is not directly usable. It is nevertheless a vivid example of something that is coming for businesses: agents that do not just complete tasks for people, but negotiate and close deals with the agents of other companies.

This article puts into context what the announcement documents, what it leaves open, and which questions you should already be asking yourself. A note on the sources: we have the press release itself in two copies, while the assessment of regions and exclusive rights comes from a trade report by the industry outlet PPC Land. We name each of them where relevant.

What was announced on October 14, 2026

According to the announcement, which was distributed from New York on October 14, 2026, Magnite and LG Ad Solutions are extending their partnership by three years. The key points:

• Magnite's ad server SpringServe remains the exclusive server for video advertising on connected TVs (CTV) from LG Ad Solutions.
• Magnite remains the preferred programmatic partner in North America, in the Asia-Pacific region and in the EMEA region. According to press coverage, the exclusivity applies only to CTV video ad serving, not to programmatic as a whole.
• The ad inventory on the home screen of LG televisions becomes available for "agentic buying" through Magnite's Seller Agent.
• According to the announcement, the independent agency H/L was the first agency to activate this inventory in this way, on behalf of the Northern California Toyota Dealers Association. The campaign is timed to the fall football season.
• Financial terms, budget, campaign duration, and reach or sales figures were not published.

One quote from the announcement: Mike Laband of Magnite says the company is replacing "manual execution with automated, high-precision workflows", that is, manual execution with automated processes. According to one copy of the announcement, the campaign runs from media planning through to execution via a single path. Jeremy Cobb of H/L describes his agency as the first to have activated this type of home screen advertising through Magnite and LG Ad Solutions.

Important for putting this in context: the statement refers to the process, not to performance. The announcement contains no figures on the campaign's results.

Buyer Agent and Seller Agent: How agents negotiate with each other

The core of the news is a pair of terms. A Buyer Agent sits on the buyer side and translates a campaign briefing into structured intents: what is to be achieved, which target audience, which time period. A Seller Agent sits on the seller side, here belongs to Magnite's SpringServe, and matches these intents against available inventory and audiences. The announcement does not say which Buyer Agent H/L used.

Magnite describes the principle in its own blog post from January 6, 2026, in connection with AdCP, the Advertising Context Protocol. According to that post, AdCP is an early, open protocol built on the Model Context Protocol (MCP). It is meant to standardize communication between ad platforms, publishers, data providers and AI agents. The post stresses that much of it is still conceptual and that first tests are only just getting under way. According to the post, a first test ran in December with the provider Scope3 as Buyer Agent.

What is new is not that advertising is bought automatically. That has been the case for years. What is new is that both sides of the deal are each represented by an agent that understands intents instead of merely placing bids.

What this means for you depends on how far you already use such systems today. If you buy traditional advertising through agencies, you will not notice any of this for the time being. If you manage campaigns yourself, you should be aware of the development, because it shifts the question: no longer "Which bid do I place?" but "Which intent do I specify, and which limits do I set for the agent?"

Availability: Live for one first customer, not confirmed for the DACH region

Availability can be stated clearly. H/L's campaign is the first use of this feature on LG's inventory. A broad rollout to any advertiser is not mentioned in the announcement. The only campaign named ran in Northern California. The partnership as a whole names North America, APAC and EMEA, but the companies do not say where the agentic home screen offering applies.

The PPC Land trade report adds a point that matters for Germany: in April 2026, LG and the provider Teads extended their exclusivity for CTV advertising on the home screen in 17 European markets, including Germany, as well as in nine markets in the Asia-Pacific region. According to that report, Teads is the sole point of access there. How this relates to Magnite's role in EMEA is left open by the announcement.

For the DACH region, this means: unknown, and probably not available through this route in the foreseeable future. We therefore do not describe it as something you can use now, but as something that is coming for businesses. The specific case is probably out of practical reach for you, but the pattern behind it is not.

Figures for context, and why caution is warranted

Anyone talking about agents in ad buying quickly runs into big numbers. One of them comes from the report: in July 2026, Magnite CEO Michael Barrett put spending on agentic advertising in 2027 at roughly 600 to 700 million US dollars. That is a forecast by the company, not an independently verified figure. According to the report, Barrett himself described it as modest compared with the programmatic market as a whole and said that protocol-based buying is in a "discovery phase", that is, an early exploratory stage.

This assessment is valuable because it tempers expectations. A provider that is convinced of its own technology is itself speaking of an early stage. For you as a decision-maker, this is a reminder not to confuse announcements of agent-based buying with market maturity.

The report also mentions an analysis that sees conventional buyers ahead of AI agents on one cost item. Because the figure is not clearly understandable in the version available to us, we deliberately do not reproduce it. If you see such comparisons in presentations, ask about the method and about what exactly was compared.

What this means for a company: Three consequences you should know about

First: Negotiations shift. When agents take over buying and selling, there are fewer phone calls about price and terms and more decisions about specifications. Anyone who formulates the specifications poorly gets poor results, only faster. The briefing becomes the most important piece of work.

Second: Control moves into evaluation. When a process runs through a single path from planning to execution, there are fewer intermediate steps at which a person happens to notice that something is wrong. This makes an evaluation that surfaces deviations early all the more important: what was booked, what ran, what did it cost, and does it match the briefing?

Third: Responsibility stays with you. The fact that two agents agree does not mean the outcome fits your business. The announcement says nothing about a human approval step. The statement that trading takes place within limits set in advance does not appear in the versions of the announcement available to us. Whether and where a human gives approval is therefore an open question that you have to ask yourself for every such system.

A small thought experiment makes the third consequence tangible. Suppose a dealer with several locations gives their agency a briefing for a fall campaign, and an agent derives bookings from it. If the briefing says only "as much reach as possible", it is unclear whether the budget will be spread across a few expensive placements or many inexpensive ones. A human would have asked. An agent that implements intents takes the specification literally. That is why follow-up questions and limits belong in the process before booking, not only in the evaluation afterward.

The assignment: Daniel and the Research team

On amazing-agents.com, we assign this topic to the Research team, more precisely to Daniel, the analytics specialist working under Nadja. The distinction matters: we do not offer an agent that buys television advertising, and we do not claim to. Daniel analyzes data.

According to his description on our website, Daniel collects performance data, builds reports and detects anomalies. He delivers a weekly performance report to Nadja and Natascha, and he reports irregularities not as alarms but as context with data. Natascha builds his recommendations into the editorial plan. That is exactly the kind of role that becomes more important in a world with agent-to-agent buying: someone who puts numbers in context before they flow into decisions.

The idea behind it can be transferred to your company, even without a single television spot: the more steps an agent completes on its own, the more important a second instance becomes, one that does not execute but observes and puts things in context.

How we approach the topic in an audit

Here we describe the approach, not existing projects. When an audit involves agents that buy, book or negotiate, we work through four points.

1. Which decisions may an agent make? We separate preparing, proposing and closing. Closing, that is, spending money or concluding contracts, belongs in our view as a matter of principle behind a human approval.
2. Which limits are set technically? A budget cap, a list of permitted channels or a maximum amount per booking are more effective than an instruction in text.
3. What is logged? Every booking should be traceable: who or what triggered it, based on which briefing, with what result.
4. Who evaluates? We determine which metrics are reviewed regularly and who decides when deviations occur.

Prepare instead of buy: Five steps for the coming weeks

Since the specific case is not confirmed for the DACH region, this is not a how-to guide but about preparation. The steps are useful to you even without this product.

1. Put briefings in writing. If agents will derive orders from specifications in the future, you need specifications that are complete: goal, target audience, time period, budget, exclusions. Practice this today with your agency.
2. Define metrics. Decide how you will measure a campaign's success before it starts, not afterward.
3. Define budget limits. Set maximum amounts per campaign and per month, and clarify who gives approval beyond that.
4. Ask your agency. Ask whether and where your agency uses or tests agents in buying, and how a human is involved there.
5. Clarify data and legal matters. When agents match audiences, questions about data and consent arise. Involve your data protection advisor early.

What the announcement does not say

Part of putting this in context is what is missing. There is no information on the campaign's cost, budget or duration. There are no results. There is no statement on when or whether the offering will become available in Europe, and there is no information on a human approval step. In addition, the source is a press release from the companies involved, that is, an account from the providers' point of view. The figures on the market forecast come from the company itself.

Anyone who turns this into "agents are now buying your advertising" overstretches the source. What is correct is this: a first customer in the US has run a campaign through an agent path, and the technology behind it is, according to its developer, still in an early phase.

The difference: An agent with limits and an approval step

The news shows where marketing processes are heading: away from many manual intermediate steps and toward agents that talk to each other. That can save time and avoid errors from retyping. But it also shifts responsibility onto the question of what an agent may do on its own.

An agent without limits executes what it understands, and nobody sees in time when that is not what was meant. An agent with limits and an approval step works within clear specifications, delivers traceable results, and a human approves whatever costs money or goes out to the outside world. On our team, Daniel stands for the part in between: putting numbers in context and making irregularities visible before they become a decision.

Sources

• GlobeNewswire press release "LG Ad Solutions and Magnite Expand Partnership to Advance Agentic Buying on the Home Screen", October 14, 2026 (copies at Yahoo Finance and StockTitan)
• PPC Land: "LG opens Home Screen ads to Magnite's AI agent in three-year deal", October 2026
• Magnite blog: "Why Magnite Built a Seller Agent — and What It Signals for AdCP", January 6, 2026
• Amazing Agents: description of Daniel, amazing-agents.com