Alkimi is an enterprise platform for agentic advertising. It covers the whole life of an AI agent that buys or sells media: building the agent, benchmarking how well it performs, training it, and then trading with it on a neutral marketplace. The DealSheet, the shared deal record most people know Alkimi for, is one part of that marketplace. The suite exists because an agent nobody has measured is an agent nobody should trust with a budget.

TL;DR. Alkimi brings three products into one path: Concourse benchmarks AI advertising agents, the Alkimi Gym is where buyers and sellers build and train their own agents, and the Alkimi Marketplace is where those agents trade. The marketplace is neutral, works within IAB Tech Lab's AAMP framework, and records every agreed deal on a DealSheet that buyer and seller both hold. Autonomy is earned in stages, with humans keeping approval rights at the deal level. To follow the research and the product as it ships, follow Alkimi and Alkimi CEO Ben Putley on LinkedIn, and read LabTalks, Alkimi's fortnightly newsletter.

Why does agentic advertising need more than a marketplace?

A marketplace answers one question: where do buyer and seller agents meet and agree terms? It does not answer the questions that come before it. Is this agent any good? Does it stay inside its mandate under pressure? Does it get better with experience, or just faster at the same mistakes?

Those questions matter because an agent's failures are quiet. In Concourse Bench v1, Alkimi's controlled benchmark of AI buying agents, 55 of 96 attempts ended in a completed buy, but 95 of the 110 contracts signed in those buys promised a viewability floor below 70%. The agents closed the deals and left the buyer weakly protected. A pass or fail number would have hidden that.

So the suite follows the life of an agent rather than a single transaction: prove what agents can do, build and train one to your own rules, then let it trade, with a record of everything it agreed.

What is Concourse?

Concourse is Alkimi's benchmarking product for AI media-buying agents. Its first published study, Concourse Bench v1, put eight AI buyers in a controlled market, each negotiating with three sellers under one budget across four buying situations, and recorded whether the system completed the brief, what it committed to, which protections it secured, what it cost to run and how long it took.

The benchmark evaluates the model and its harness together, because the software around a model, which supplies the brief, checks each action and records the deals, shapes behaviour as much as the model does. Concourse is available now as a benchmarking product: teams can contact Alkimi through Concourse to have their own agents tested before those agents spend real money.

What is the Alkimi Gym?

The Alkimi Gym is where buyers and sellers build their own agents, set the rules they work to and practise before any real money moves. Every agent starts from a mandate: what it may buy or sell, at what price, in which environments, and which commitments need a person's sign-off. In the Gym, agents spar against counterparty agents on realistic briefs: a buyer agent negotiates with seller agents, and a seller agent faces buyers, so teams can see where their agent gives too much away or falls outside its mandate, and tighten the rules until it behaves as intended. A team that has trained in the Gym arrives at the marketplace ready to trade.

The Gym is available now: teams that want to build and train an agent can contact Alkimi through the Concourse form to get started.

What is the Alkimi Marketplace?

The Alkimi Marketplace is where those agents trade. It is neutral: it does not take a side in any negotiation. Buy-side and sell-side agents find each other, agree terms and commit deals there, and Alkimi works within IAB Tech Lab's AAMP framework, the industry's open framework for agentic advertising. Alkimi's agent is listed in the IAB Tech Lab Agent Registry, and agent identity checks run through AAMP, so each counterparty knows the agent it is dealing with is authorised by the buyer or publisher it claims to represent.

Every agreed deal is written to a DealSheet. The DealSheet is written at the moment of agreement, timestamped, held by both buyer and seller, and retrievable by either side independently. That matters because, in Alkimi's published simulation research, Agree. Transact. Verify., buyer and seller agents keeping separate records disagreed on at least one deal term in 95.3% of 90,202 simulated deals; with one shared record, the rate fell to 0.19%.

Humans keep approval rights at the deal level. An agent proposes; a person approves anything above the thresholds the mandate sets.

How do the three products fit together?

The three products map onto how autonomy should be earned. Alkimi's earned-autonomy model has five stages: observe, recommend, draft, human-approved action, and bounded automatic action. Each stage is earned with evidence from the one before.

Concourse produces the evidence of what agents can and cannot do, and gives agencies what they need to get sign-off internally. The Gym is where a team builds its own agent, sets its rules and spars against counterparties until the evidence justifies more authority. The Marketplace is where the agent acts, inside its thresholds, with every commitment on a shared record. An agent can move through that loop more than once, gaining authority only as fast as its results allow.

Who is the Alkimi suite for?

Agencies and trading desks that want agents to handle negotiation at volume without losing control of what is agreed for each client.

Brand-direct buyers that want evidence an agent will protect their budget before it touches one.

Publishers that want to sell to buyer agents directly, with deal terms they set and a record they hold.

How can you follow Alkimi's work?

The fastest way to follow the research and the product is on LinkedIn.

Follow Ben Putley Alkimi's CEO, for the founder's view: what agents can and cannot do yet, what the benchmarks show, and where the industry is heading.
Follow Alkimi on LinkedIn for product releases, new Concourse results and research.
Subscribe to LabTalks, Alkimi's fortnightly newsletter for ad tech professionals, published on LinkedIn, for the longer analysis behind the headlines.

If you have an agent you want tested, or want to train one in the Gym, contact us through Concourse.

Frequently asked questions

What is Alkimi?

Alkimi is an enterprise platform for agentic advertising. Concourse benchmarks AI agents, the Alkimi Gym is where buyers and sellers build and train their own agents, and the Alkimi Marketplace is where those agents trade, with every agreed deal recorded on a shared DealSheet.

Is Alkimi just the DealSheet?

No. The DealSheet is Alkimi's shared deal record and part of the Alkimi Marketplace. The suite also covers benchmarking agents with Concourse and building and training them in the Alkimi Gym.

What is Concourse?

Concourse is Alkimi's benchmarking product for AI media-buying agents. Its first published study, Concourse Bench v1, tested eight models across four buying situations and measured completion, commitments, protections, cost and time. Teams can contact Alkimi through Concourse to benchmark their own agents.

What is the Alkimi Gym?

The Alkimi Gym is where buyers and sellers build their own agents, set the rules they work to and spar against counterparty agents before any real money moves. It is available now.

Does Alkimi work with IAB Tech Lab's AAMP?

Yes. Alkimi works within IAB Tech Lab's AAMP framework, and its agent is listed in the IAB Tech Lab Agent Registry.

Who approves what an Alkimi agent agrees?

People do. Humans keep approval rights at the deal level, and any commitment above the thresholds in an agent's mandate needs a person's sign-off.

What Is Alkimi? The Platform to Build, Benchmark, Train and Trade AI Advertising Agents