To train a media team to work with AI buying agents, teach three things in order: how to brief an agent through a written mandate, how to check what the agent agreed, and when to step in. A 30-day plan does this in four weekly blocks: shared foundations, mandate writing, practice in a sandbox with no real money, then supervised live work with a person approving every deal. The goal is not a team that operates a tool, but a team that can supervise one.

TL;DR: Week one gives everyone hands-on time and a shared vocabulary. Week two teaches the team to write mandates with minimum terms as well as maximum prices. Week three drills the team on catching an agent's mistakes before money moves, and week four moves to live buying where every deal still needs human sign-off.

This plan trains people. Testing the agent or platform itself is covered in How Do You Run a 30-Day Agentic Advertising Pilot?, and the two can run side by side.

What does a media team need to learn to work with AI buying agents?

A media team needs supervision skills more than operating skills. Digiday's report of a closed-door agency town hall at its Programmatic Marketing Summit in December 2025 concluded that even limited agent use would require training not just on AI tools and their limitations but also on the fundamentals of programmatic workflows, so that people could better supervise their AI counterparts. One participant put the risk plainly: "All it takes is for the AI to hallucinate one zero, and you're in big trouble."

Checking matters because an agent can finish a buy and still agree weak terms. In Concourse Bench v1, a benchmark of AI media-buying agents run by Alkimi, 55 of 96 runs completed a buy, and 95 of 110 signed contracts promised a viewability floor below 70%. A team trained only to confirm that the buy went through would have passed most of those contracts.

Who in the media team needs training, and how much?

Everyone who touches an agent-bought deal needs some training, but not the same amount. A suggested split:

Buyers and traders: the full 30 days, because they write mandates and approve deals.
Planners: weeks one and two, so briefs arrive in a form an agent can follow.
Account leads: week one plus the week four sign-off, so they can explain agent decisions to clients.
Finance: week one plus the record-checking exercise in week three.
Brand safety and legal: week one plus the mandate workshop in week two, where excluded categories and minimum terms are set.

Reshma Karnik, chief media officer at BarkleyOKRP, told the same Digiday summit that the agency was cross-training its search, programmatic and traditional media staff to build "all-rounders" who can "see the whole map".

Week 1: How does the team build a shared foundation? (days 1 to 7)

Week one gives every person hands-on time with an agent and a shared vocabulary. Monica Capelan, SVP of programmatic at Horizon Media, described the agency's starting point at the same summit, as reported by Digiday in December 2025: "the first thing that we did is we decided that everybody needed hands-on-keyboard experience."

Suggested sessions for the week:

Day 1: what an AI buying agent does, and the vocabulary of mandate, escalation, approval threshold and deal record.
Day 2: the five stages of earned autonomy (Observe, Recommend, Draft, Human-approved action, Bounded automatic action), so the team knows which stage it is working at.
Day 3: the standards. IAB Tech Lab's AAMP page links explainer videos and webinar replays, and AgenticAdvertising.org, which publishes AdCP, offers certification modules for practitioners.
Days 4 and 5: each buyer runs a practice brief through the agent with no spend attached and writes down how they would have done the same buy by hand.
Days 6 and 7: a data hygiene review of the naming conventions and reports the agent will read.

The hygiene step is easy to skip. Capelan said Horizon Media was fixing its naming conventions so that an agent reading the daily pacing report would not judge a CTV line item as if it were display.

Week 2: How does the team learn to brief and constrain an agent? (days 8 to 14)

Week two teaches the team to write mandates, because the mandate is the agent's only reliable instruction. A mandate states what the agent may buy, from whom, and which terms it must not go below. The method is set out in How to Write a Media Buying Mandate for an AI Agent.

The core exercise uses a real brief the team has already bought by hand. Each buyer writes a mandate for it, then swaps with a colleague who tries to find a deal that would satisfy the mandate but fail the client. Mandates that only cap price fail this test quickly, which is the lesson the Concourse numbers point to.

By the end of the week the team should hold a mandate template agreed with brand safety and legal, approval thresholds for each autonomy stage, and a list of decisions that always go to a person.

Week 3: How does the team practise catching an agent's mistakes? (days 15 to 21)

Week three is practice in a sandbox, where agents negotiate against test counterparties and no real money moves. The team reviews every result, not just the failures.

Suggested drills for the week:

Term review: check every agreed term against the mandate and flag any that fall short.
Planted errors: a trainer seeds practice briefs with traps, such as a seller offer with no make-good clause, and records how many each reviewer catches.
Escalation drill: time the team's answer to an agent's escalation against the turnaround agreed in week two.
Record check: finance compares the agent's account of each deal with the counterparty's and logs any mismatch.
Log reading: buyers trace one decision back through the agent's records, using the questions in What Should Buyers Be Able to See in an AI Buying Agent's Decision Log?

The planted-error score is the most useful number the training produces, because it measures the skill that protects the client.

Week 4: How does the team move to supervised live work? (days 22 to 30)

Week four puts the team on live buys at the Human-approved action stage, where the agent negotiates but a person signs off every deal before it commits. Each approval is checked by a second buyer for the first week of live work.

Two sessions close the plan. A tabletop incident exercise walks through what the team would do if an agent agreed a deal outside its mandate, and a sign-off review lets the budget owner decide who may approve what from day 31, based on the week three scores and the week four approval record. Moving the agent beyond human approval is a separate governance decision.

How do you know the training worked?

Training worked if the team catches errors, not if it uses the tool more. Suggested measures are the planted-error catch rate, escalation turnaround against target, the share of mandates that pass a colleague's review first time, and live deals where a reviewer changed the agent's proposal.

Adoption alone is a poor signal. A Digiday+ Research survey of 142 brand and agency professionals found 86% said their companies were investing in AI technology in 2025, up from 44% in 2022, while many marketers interviewed said training on how to use AI tools well lagged behind adoption.

Is staff training on AI a legal requirement?

For organisations that provide or deploy AI systems in the EU, some training is an obligation. Article 4 of the EU AI Act requires providers and deployers of AI systems to "take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf".

The European Commission's AI literacy questions and answers say Article 4 does not entail an obligation to measure employees' AI knowledge. The guidance suggests a programme should consider what AI is, how the organisation uses it, and the risks staff need to know about, which maps closely onto weeks one to three of this plan.

What mistakes should media team training avoid?

The most common mistake is teaching the tool instead of the job, a point Dan Gardner, co-founder of Code and Theory, made to Digiday: "Anybody can learn a new tool. Upskilling and reskilling is multiplying the value of your human ingenuity."

The second mistake is treating 30 days as the end. GSD&M, the Omnicom-owned agency, told Digiday that it set aside office hours each week for AI upskilling and recorded them for anyone who missed a session.

Where does a sandbox fit in team training?

A sandbox lets the team make its first mistakes where they cost nothing. One option for week three is the Alkimi Gym, where buyers and sellers build their own agents, set the rules they work to and spar against counterparty agents before any real money moves. Whatever environment a team uses, the test is the same: the team should be able to explain every deal the agent agreed, and say which ones it would have refused.

Frequently asked questions

How long does it take to train a media team to use AI buying agents?

A focused plan can build the core skills in about 30 days, from foundations to supervised live work. Skills then need regular refreshing, because agents and standards change.

Should training happen before or after an agentic pilot starts?

Weeks one and two work better before the agent touches a live budget, since they set the mandate the pilot depends on. Weeks three and four can run alongside a pilot.

What is the most important skill to test at the end of training?

The ability to catch what the agent got wrong, measured with planted errors in practice deals. A team that only confirms completed buys will miss weak terms.

Who should sign off that a team member is ready to approve agent deals?

The budget owner, using the week three error-catch results and a record of supervised approvals. Approval rights can then be widened step by step, in line with the earned autonomy stages.