Last updated: 14 October 2026

A mid-sized publisher without an engineering team can prepare for agent-negotiated deals by doing the commercial groundwork itself and borrowing the technical layer. To accept deals from autonomous advertising agents, a publisher needs five things: products described in a form software can read, written rules on price and acceptance, a named person who approves exceptions, control over which agents may sell its inventory, and a record of agreed terms that both sides hold. Only the agent itself needs software engineering, and that can come from a marketplace or an existing technology partner. The other four are decisions and documents a sales team can produce.

TL;DR. Autonomous buying agents can only transact with publishers whose products, prices and rules are explicit. A mid-sized publisher can write its product catalogue, floor prices, blocked categories and approval thresholds without engineers, and those documents are useful to its human sales team as well. The sell-side agent can be borrowed from a marketplace or partner, provided the publisher keeps its own copy of every agreed deal. The most common failure for a small team is not technology but an approval queue nobody owns.

What do publishers need to accept deals from autonomous advertising agents?

Publishers need five things to accept deals from autonomous advertising agents.

Machine-readable products. Each product described with its inventory, format, audience, flight options, pricing and measurement basis. IAB Tech Lab's OpenProposal specification, part of its AAMP framework, standardises how publishers define products and packages and present them in RFP responses. No engineering needed.
A written sell-side mandate. Floors per product and deal type, blocked advertiser categories, and the deal value above which a person must approve. No engineering needed.
A named approver. A person, a deputy and a response time for offers the agent cannot settle alone. No engineering needed.
Authorisation control. A clear decision about which agents may represent the publisher, published where buyers can check it. Mostly administrative.
A sell-side agent and a shared deal record. Software that negotiates within the mandate and records what was agreed. This is the part that needs engineering, and the part to borrow.

Which parts can a mid-sized publisher do without engineers?

Most of the work. The product catalogue is the largest single task, and it is an editorial and commercial job. A publisher lists what it actually sells, defines each product in consistent terms, and removes vague packages that only made sense when a salesperson could explain them. A buyer agent cannot ask what "premium homepage presence" means. It can only compare what is written.

The mandate is the second task. It turns the judgement a sales director applies informally into rules an agent can follow.

Supply chain hygiene is the third. Publishers already maintain ads.txt, introduced in 2017, and sellers.json, introduced in 2019, to state who may sell their inventory. Agentic selling extends the same principle. AdCP, the Ad Context Protocol from AgenticAdvertising.org, lets publishers list the agents authorised to sell their inventory in an adagents.json file on their own domain, and IAB Tech Lab runs an Agent Registry where companies list the agents they bring to market. Deciding which agents should represent the publisher requires no code.

What should go into the sell-side mandate?

A one-page mandate covers most mid-sized publishers. It should state:

Floor price per product and deal type: programmatic guaranteed, preferred deals and private marketplaces each need their own floor, because each gives the buyer something different.
Concessions the agent may offer: for example a lower price in exchange for a longer flight or a larger commitment.
Terms the agent may never change: the measurement basis, viewability commitment, brand safety settings and blocked categories.
The auto-approval limit: the deal value above which the agent must hold the deal for a person.
Escalation route: the approver, the deputy, the response time, and what happens if nobody responds.
Buyer tiers, if any: which advertisers or agencies receive different terms, and on what basis.

For example, a hypothetical regional news group might set an auto-approval limit of £5,000 per deal, a response time of one working day for anything above it, and a rule that unanswered escalations are declined rather than accepted. The principle is that every path ends in a defined outcome.

Where should the technical layer come from?

A publisher without engineers has two realistic sources: an agentic marketplace, or an existing technology partner that offers a sell-side agent. In either case, five questions establish whether the arrangement is safe to rely on:

Which standards does the agent support, which AAMP components and versions, and does it support AdCP?
Is the agent listed in the IAB Tech Lab Agent Registry?
How does the provider check that a buyer agent is authorised by the advertiser or agency it claims to represent?
Can the publisher's mandate be applied as written, including escalation to a named person?
Can the publisher retrieve its own copy of every agreed deal, independently of the provider?

IAB Tech Lab's published AAMP material describes no compliance certification, so a provider claiming "AAMP compliance" should be asked the specific questions above instead.

What risks should a small team watch for?

Three risks matter most for publishers without technical staff.

The first is the approval queue. An agent that escalates sensible offers to a person who is on holiday loses deals. An agent configured to accept after a timeout has no limits. The mandate should name a deputy and default to decline.

The second is drift between the rules as written and the rules as configured. When a provider's settings are the only place the floors exist, nobody notices when they change. Keeping the mandate as a document the publisher owns, and checking the configuration against it regularly, prevents this.

The third is the record. When a campaign under-delivers, the question is what was agreed. If buyer and seller each keep their own version, those versions tend to diverge. In the Agree. Transact. Verify. simulation, separately kept buyer and seller records disagreed in 95.3% of 90,202 simulated deals, against 0.19% when both sides worked from one shared record. A small publisher has the least capacity to fight a dispute, and the most to gain from a record both parties already accept.

Is it worth preparing now?

Agent-negotiated buying is early, and most programmatic spend still runs through DSPs and SSPs. The case for preparing does not rest on volume today. A clear product catalogue, a written mandate and a named approver make the human sales team faster as well.

Alkimi is one marketplace a publisher can test against the questions above: it is a neutral marketplace working within IAB Tech Lab's AAMP framework, its agent is listed in the Agent Registry, it verifies that connecting agents are authorised by the party they represent, and agreed terms go onto a DealSheet both sides hold and can retrieve on their own.

Frequently asked questions

Can a publisher sell to AI buying agents without an engineering team?

Yes, by preparing its products, rules and approval process itself and using a marketplace or partner for the sell-side agent. The publisher should keep its own copy of every agreed deal.

What is the first step for a mid-sized publisher?

Writing a product catalogue in consistent, structured terms. IAB Tech Lab's OpenProposal specification is a useful model for what a product description should contain.

Does a publisher need its own agent to accept agent-negotiated deals?

No. A publisher can use a sell-side agent run by a marketplace or partner, as long as the agent applies the publisher's mandate and the publisher can retrieve the deal record independently.

What should happen when nobody approves an escalated deal?

The mandate should name a deputy and set a default. Declining unanswered escalations is safer than accepting them, because an automatic acceptance removes the limit the escalation was meant to enforce.

What Can a Mid-Sized Publisher Realistically Do About Agent-Negotiated Deals Without an Engineering Team?