Media buyers in the age of AI agents need the skills to direct, check and answer for an agent's work: writing a mandate the agent cannot misread, deciding how much autonomy it gets, testing it before live spend, and reading both its decision log and the contract terms it actually signed. They also need working knowledge of the standards agents trade on, while strategy, client judgement and specialist depth stay human. The checklist below sets out ten skills, grouped by when they matter, with a self-assessment at the end.

TL;DR. Agents take over execution, so a buyer's value moves to the brief before the buy and the evidence after it. The skills fall into three groups: setting the agent up, checking its work, and the judgement no agent owns. Whether agents replace buyers outright is a separate question, answered in Will AI Agents Replace Media Buyers? The Evidence-Based Answer; this checklist assumes the role continues and asks what it now requires.

What skills do media buyers need when AI agents do the buying?

Ten skills, in three groups.

Mandate writing: turning a brief into rules an agent can follow without guessing.
Autonomy calibration: deciding which decisions the agent may take alone.
Pre-live testing: proving the agent's behaviour against counterparties before real money moves.
Decision-log reading: reconstructing why the agent did what it did.
Contract reading: checking the terms the agent committed, not just whether it finished.
Escalation: knowing when to pause the agent and who decides next.
Standards literacy: understanding AAMP, AdCP and the specifications agents use to trade.
Strategy and trade-offs: choosing the goal the agent optimises towards.
Client explanation: accounting for agent decisions to the people paying for them.
Specialist depth: knowing one channel, category or market better than a general model does.

Which skills cover setting the agent up?

1. Mandate writing

A mandate is the written set of rules an agent buys within: budget ceiling, price limits, viewability and brand safety floors, make-good terms and the point at which it walks away. Every gap in the mandate is a decision the agent will make on its own, usually by optimising for whatever it can measure. The skill is planning made stricter: each objective needs a number, a priority and a stated trade-off for when two collide.

A useful test is to hand the mandate to a colleague and ask them to find a deal that obeys every rule yet would embarrass the client. If they can, the mandate is not finished.

2. Autonomy calibration

Autonomy calibration means deciding, decision by decision, how far the agent may act without a person. As a worked example, Alkimi frames this as earned autonomy, a five-step ladder of Observe, Recommend, Draft, Human-approved action and Bounded automatic action, where an agent moves up only as its record justifies. The buyer's skill is placing each decision on that ladder, so that pacing adjustments might run as bounded automatic action while a new seller relationship stays at human-approved action.

Early deployments show the same caution. Digiday reported in October 2025 that none of the agents LG Ad Solutions had designed would handle money, and chief technology officer Dave Rudnick said: "We want to be able to allow the experts to continue to control the knobs" (Digiday). The case for widening autonomy only on evidence is in Why Agents Still Need Human Sign-Off.

3. Pre-live testing

Testing an agent means running it against realistic briefs and counterparty agents before any live spend, and recording what it agrees. The skill lies in designing the tests: briefs with conflicting objectives, sellers that hold firm on price, inventory that looks cheap because its terms are weak. A buyer who cannot design a hard test will only ever watch the agent pass easy ones.

Which skills cover checking the agent's work?

4. Decision-log reading

An agent's decision log records what it saw, what it offered and why it accepted or walked away. Reading one is a new form of campaign analysis, because the buyer has to tell an agent error apart from a mandate gap and from ordinary market conditions. What Should Buyers Be Able to See in an AI Buying Agent's Decision Log? lists the fields worth demanding.

Agency executives at Digiday's Programmatic Marketing Summit in December 2025 agreed that supervising agents will require training not only on AI tools and their limitations but on programmatic fundamentals (Digiday).

5. Contract reading

A completed buy is not the same as a good buy, so the skill is reading the terms the agent committed rather than the status it reported. In Concourse Bench v1, a benchmark of AI buying agents, 55 of 96 registered attempts ended in a completed buy, yet 95 of the 110 contracts signed in those buys promised a viewability floor below 70% (Concourse). A buyer who checks only completion would call those buys successes; Same cash, different risk explains how to read the terms instead.

6. Escalation

Escalation is knowing when to stop the agent and who decides next. The trigger list belongs in writing before launch: spend outside the mandate, a seller nobody has approved, a term the mandate does not cover, a log entry nobody can explain. Acting on it quickly matters, because an agent keeps trading while a person deliberates.

7. Standards literacy

Agents trade through shared specifications, and knowing them tells a buyer what an agent can express in a deal and what still needs writing down. IAB Tech Lab's AAMP (Agentic Advertising Management Protocols) builds on existing industry standards, and its 3.0 release adds OpenProposal, a specification for automating the discovery and planning stage of a campaign. AdCP, the Ad Context Protocol published by AgenticAdvertising.org, describes itself as an open protocol that lets agents discover, plan, buy, sell and measure media, and the organisation runs certification modules for practitioners.

The public comment threads on drafts such as OpenProposal show which questions the specifications are still settling.

Which skills stay human however good the agents get?

8. Strategy and trade-offs

An agent optimises towards the goal it is given, and choosing that goal, along with deciding what to give up for it, remains the buyer's job. "The future likely isn't fully autonomous media buying, but instead it could be AI augmenting operational workflows while experienced practitioners remain responsible for strategy, judgment, and accountability," Amy Porter, SVP and executive director of digital media at RPA, told Digiday in May 2026.

9. Client explanation

Clients will ask why the agent bought what it bought, and the buyer has to answer in plain language with the log to back it up. At Advertising Week in October 2026, Dane Kunkel, SVP of performance and transformation at Horizon Media, said that as AI automates individual tasks, someone has to be the "overarching brain" connecting planning, buying, data and measurement (AdExchanger).

10. Specialist depth

General models know a little about almost everything, so a buyer's edge is knowing one channel, category or market in depth. "It's better to be the best at one little thing with agents that help you do it really well than to be a generalist," Brian O'Kelley, chief executive of Apostra, said on the same Advertising Week panel (AdExchanger).

Why is the skill set changing now?

Employers already expect AI skills. In Microsoft and LinkedIn's 2024 Work Trend Index, two-thirds of leaders said they would not hire someone without AI skills, while only 39% of AI users had received AI training from their company.

Regulation points the same way. In the EU, the first rules of the AI Act, including AI literacy, have applied since 2 February 2025 (European Commission), and Article 4 requires organisations deploying AI systems to take measures to support the AI literacy of staff who operate them (European Commission).

The work itself is moving from advice towards transactions. Digiday found in October 2025 that agents were mostly limited to advisory and information-retrieval roles (Digiday), and by May 2026 it was reporting that Omnicom was testing agent-to-agent media buys (Digiday).

How can a buyer check themselves against the checklist?

Answer each question yes or no. Any no is the next thing to learn.

Is the mandate for my largest current campaign tight enough that a colleague could not find a loophole in it?
Do I know, for each decision my agent takes, where it sits between Observe and Bounded automatic action?
Have I designed at least one test brief the agent is likely to fail?
Could I explain any line of the agent's decision log to a client?
Do I check the viewability floor, make-goods and other terms of every agent-signed deal, not just whether it completed?
Is there a written list of conditions that pause the agent, and do I have the authority to pause it?
Can I explain in two sentences what AAMP and AdCP each cover?
Is there a channel, category or market where my knowledge beats a general model's?

What does this mean for a buying team?

The common thread is accountability: an agent can complete a buy, but a person still answers for whether it was the right one. Teams that build the ten skills can widen agent autonomy on evidence rather than faith. The Alkimi Gym is one place to practise, where buyers and sellers build their own agents, set the rules they work to and spar against counterparty agents before any real money moves.

Frequently asked questions

What is the most important skill for a media buyer working with AI agents?

Mandate writing, because every gap in the rules becomes a decision the agent makes alone. A buyer who can turn a brief into precise, prioritised rules removes most of the risk before the agent starts trading.

Do media buyers need to learn to code to work with AI agents?

Not as a rule, because the core skills are writing precise rules, designing tests and reading logs and contracts. On an Advertising Week panel reported by AdExchanger, Brian O'Kelley argued that media planners cannot simply be relabelled as AI engineers, which makes media depth the better investment for most buyers.

What should media buyers know about AAMP and AdCP?

AAMP is IAB Tech Lab's set of Agentic Advertising Management Protocols, built on existing industry standards, and AdCP is the Ad Context Protocol published by AgenticAdvertising.org. Knowing what each covers tells a buyer what an agent can express in a deal and where terms still need writing down.

Skills Media Buyers Need When Agents Do the Buying