TL;DR: Agentic automation applies AI to execute predefined tasks faster and more consistently than humans can. Agentic negotiation applies AI to reach bilateral agreements with another party's agent through a structured exchange of proposals and counter-proposals. The first type of agent follows instructions; the second type reaches agreements. Both are called "AI agents," but the governance requirements, the accountability structures, and the deal infrastructure they need are different.
The word "agentic" is applied broadly in advertising technology to mean "involving AI that does something autonomously." Under this definition, a bid optimisation algorithm and a bilateral deal negotiator are both "agentic." The conflation is unhelpful: the two systems have different inputs, different outputs, different governance requirements, and different accountability records.
Understanding the distinction is practically important for buyers evaluating platforms that claim to offer agentic capability. The question is not whether the platform uses AI; it is what kind of work the AI is doing and what accountability infrastructure that work requires.
What agentic automation looks like
An agentic automation system applies AI to a defined task with specified rules. In advertising, common examples include: bid management systems that adjust bids in real-time based on performance data; pacing algorithms that distribute budget across dayparts according to predicted conversion rates; and brand safety systems that filter inventory based on content classification.
These systems have in common that they execute within a predefined rule set. The bid management system bids according to its optimisation model; it does not negotiate the price with the publisher. The pacing algorithm distributes budget according to its model; it does not agree delivery terms with a supply source. The brand safety system filters according to its classification; it does not agree safety standards with a publisher.
Agentic automation does not require a mandate in the bilateral deal record sense. It requires configuration: rules, parameters, objectives. The accountability record is the optimisation log: what the algorithm did, at what inputs, producing what outputs. This is the standard reporting that DSPs and ad serving platforms have provided for years.
What agentic negotiation looks like
An agentic negotiation system engages with a counterparty (another agent) to reach bilateral agreement on deal terms. The buy-side agent proposes; the sell-side agent evaluates and responds; the exchange continues until agreement or no-deal. The output is not an executed trade based on a pre-set algorithm; it is an agreed set of terms that both parties committed to through a negotiation process.
Agentic negotiation requires a mandate because the agent is making commitments, not just executing rules. The mandate defines the scope within which the agent can commit: what it can agree to, at what price, in what context, under what conditions. Without a mandate, the agent is committing on behalf of the buyer without a documented scope.
The accountability record for agentic negotiation is the deal record: the bilateral document that captures what was agreed. This is different from an optimisation log because it represents a commitment, not a decision: the seller also holds it, and it is the basis for a delivery obligation.
Why the distinction matters for buyers
Buyers who treat agentic automation and agentic negotiation as equivalent are applying the wrong governance framework to one of them.
Agentic automation governance is configuration governance: ensuring the algorithm is configured correctly, the rules are appropriate, and the outputs are monitored. This is an extension of the governance that buyers have applied to DSP campaigns for years.
Agentic negotiation governance is mandate governance: writing the mandate, versioning it, establishing approval thresholds, reviewing agent decisions against mandate, and maintaining bilateral deal records. This is a different set of disciplines, and it requires different processes, different documentation, and different oversight structures.
The practical error is deploying agentic negotiation infrastructure while applying only automation governance to it. An agent that is negotiating bilateral deals without a mandate is generating commitments that the buyer cannot audit. The platform may describe the product as "agentic," but the buyer's governance infrastructure does not match what the agent is doing.
How to identify which type a platform is offering
The test question is: does the agent commit the buyer to agreed terms with a specific counterparty, or does it execute transactions within a pre-set algorithm without counterparty agreement?
If the agent bids, paces, or optimises within a rule set: that is agentic automation. The deal structure is auction-based; the price is won, not agreed; the record is an impression log.
If the agent proposes terms, receives counter-proposals, and executes a deal when terms are agreed with a specific counterparty: that is agentic negotiation. The deal structure is bilateral; the price is agreed, not won; the record is a DealSheet.
Most major platforms now offer both. Buyers should confirm which mode a given buying activity uses and apply the appropriate governance model to each.