Agentic Ad Lab: who writes it, who pays, what would reverse the call

AgenticAdvertising.org publishes AdCP, the Ad Context Protocol, as a versioned schema set. IAB Tech Lab publishes AAMP, the Agentic Advertising Management Protocols, as eight repositories of mostly reference code. If you hold a budget, that is a procurement question rather than a standards question, and it has an answer this quarter: AdCP is the only one of the two publishing a released contract you can write code against, so a pilot there is fundable now. AAMP’s one finished specification, ARTF, governs bid mutation inside a live auction, which is auction-side work an exchange or an SSP can fund today and no reason for a buyer or a publisher to move. The nearest thing to a second is the wire contract underneath the two reference agents, iab-agentic-primitives, and it is unfinished: at 2fc7029fb9eca801bc94586aec398906aa6af6c0 it is tagged below 1.0 while the OpenAPI document inside it declares info.version: 0.1.0. Everything else under the label is code without a contract, and I would keep it on a watch list instead of a roadmap.

Constantine Mirin wrote that paragraph, and Postindustria sells integration work on the stack it recommends. I am CEO of Postindustria and I build agentic systems that run in production. That is the qualification and the conflict, both. Price it in, then go and check the call.

What moves AAMP off the watch list is a versioned AAMP release with a schema behind it. Until a tag exists at the AAMP level, there is nothing to conform to. On 16 August 2026 the AAMP hub repository carried no tags at all. The exact tag that would do it is named below. If you are scoping Q1, that is one integration to fund and one to leave off the roadmap, and the side-by-side prices both.

Who paid for the other pages open in your tabs

Every source a buyer reads on this subject was funded by somebody with a position in it, and most of them say so if you look. Here is the map, this site included.

Those rows are disclosures, not accusations. A stated interest tells you which claim to test first, and it is the absence of one that should worry you. The comparison is where a commercial interest could most plausibly bend the writing, so it is the page that carries the file paths and the linked sources. Adgentek, which publishes its own AdCP-versus-AAMP piece, opens by disclosing that it is a founding member of the Ad Context Protocol, which is the right way round to do it. The disclosure runs both ways in the same passage, since Adgentek is an IAB Tech Lab member too and argues that interoperability outranks organisational allegiance.

So the test for any page in this category, this one first: does the writer name what they sell, and do they name the observation that would reverse the recommendation? The second half is the one people skip. Here it is: iab-agentic-primitives, the wire contract both AAMP reference agents import, tagged 1.0 or above with its OpenAPI info.version moved to match, which is the same trigger the comparison watches. Short of that tag an AAMP integration stays unbudgeted here; the week it lands, it gets scoped and costed alongside the AdCP one, and nothing already written against a pinned AdCP release is thrown away to do it.

Constantine Mirin, staff of one

Public trail: LinkedIn, GitHub, Crunchbase. The name appears as both Constantine and Konstantin depending on which profile you land on. Same person, and the transliterated spelling is what the LinkedIn and Crunchbase slugs use.

Postindustria is a general development shop with a Prebid and header-bidding practice and OpenX among its named clients. Its site, read on 16 August 2026, says nothing about AdCP, AAMP or agent readiness.

The protocol work is newer than the marketing, and it is described at /services/: integration and readiness reviews for companies deciding whether to expose a buying agent or a selling one. That is both sides of the same transaction. The direction matters more than the fact of the conflict, because if you are a publisher weighing advice about the sales agent you would expose on your own domain, that advice comes from someone who also advises the buying agents that will call it. This site is the showroom for that work.

When a page here says “I”, it means me, and it names the thing that would prove the judgement wrong.

Two claims here that nobody outside can check

This site takes no money from anyone it writes about, including for a place in a comparison. Nothing can confirm that. There is no register, no third party and no artifact behind it, so it rests on the author’s word in a way that no other sentence on the site does.

The second is non-affiliation. AdCP is maintained by AgenticAdvertising.org, which its FAQ describes as “an independent specification body — not a subsidiary or working group of IAB Tech Lab”, and AAMP is published by IAB Tech Lab. Neither standards body is a client, neither has read a word of this, and neither has approved it. A hostile reader would test that first and find they cannot: AgenticAdvertising.org’s member directory renders in JavaScript and returns nothing to a fetcher, and IAB Tech Lab publishes no open equivalent. Checked 16 August 2026. Both claims stand or fall on whether the rest of the page has earned them.

What you can check instead

Calling the two equally documented would be a lie. IAB Tech Lab announced AAMP 2.3 on 30 July 2026 and the string appears in no file in any of the eight AAMP repositories, which makes 2.3 a programme milestone, not an artefact you can pin; the version matrix shows what does exist instead. A GitHub code search scoped to the whole IABTechLab organisation, run on 16 August 2026, returns the string nowhere either.

agentic-audiences ships specs/v1.0/schema/agent_interface.schema.json at zero bytes, the file whose name promises to define the interface between buyer and seller agents. It defines nothing. It is not alone: at commit 106ebd645668677ff8be458877a5684a5d229961 that schema, specs/v1.0/examples/buyer_agent_request.json, specs/v1.0/examples/seller_agent_response.json and specs/roadmap.md are all empty files, which is a pattern rather than one slipped commit.

Now the concession a CTO reaches for, and it is a fair one. AdCP moves fast enough to be a moving target itself, at 17 releases in the 46 days from 1 July to 16 August 2026, across the 3.1.x line and the maintained 3.0.x line, both still cutting patches. What moves at that cadence is the number rather than the surface: diff the two most recent manifests and nothing an integration touches has changed, 64 tools in each, the same 64 names, 92 error codes and 21 specialisms, with the 3.1.13 side of that diff at a0fd0b5f97e9297acbd4b234f27fb440bf22d31e and the 3.1.14 side from the same release check. So the cost is not a rewrite every quarter, it is a re-pin and a re-run of the conformance suite every time the number moves, and that is a standing maintenance line rather than a one-off integration. A statement of work that says only “AdCP” has still bought you nothing. What survives the objection is narrow. A buyer can name an AdCP release number — this site pins every claim to 3.1.13, and 3.1.14 shipped on 15 August 2026 with a 3.0.23 for the maintained line the same day — and hold a supplier to it, and there is no AAMP number that does the same job.

What this site does not build

That evidence is the output here, not software. Both sides already run a standards body, a registry and a certification programme, and nobody needs a third of each. AgenticAdvertising.org publishes a brand.json builder and an adagents.json builder, both linked from AdCP’s own intro page, and they are good. Documenting the fields they emit is more useful than rebuilding them. That is why the estate is explorers and guides rather than tools.

The subject is those two advertising stacks and the transports underneath them. MCP and A2A are in scope, because AdCP defines its planning-time operations over both and AAMP’s agentic-direct is an A2A and MCP wrapper over OpenDirect 2.1. Not all of AAMP rides them: ARTF is a protobuf contract for serve-time bid mutation and agentic-audiences ships a Prebid RTD example, at prebid-module/ in the same 106ebd645668677ff8be458877a5684a5d229961 commit. Both transports version on their own schedule, which is a second cadence any integration inherits on top of the protocol’s. The commerce protocols are out of scope. AdCP’s glossary defines the Agentic Commerce Protocol (ACP), from OpenAI and Stripe, and the Universal Commerce Protocol (UCP), from Google with Shopify, Walmart and Target, as commerce standards for AI assistants; reading them as the layer next to advertising rather than inside it is this site’s judgement rather than the glossary’s, and nothing here covers them.

I run six sites on this subject, and this is the one that takes a side. Fund one integration and watch the other, and AdCP vs AAMP is where that call gets made, with the licence and governance risk priced in.

A publisher needs to know what to put on their own origin before a buying agent arrives. That turns out to be one static file, and agent discovery files is the whole of it, including the part of the schema that survives an agent domain being compromised. An implementer needs field names instead, and the reference sites carry no argument at all: AdCP Explorer and AAMP Explorer render one page per operation and per repository, holes included, the ad tech glossary takes one term per page, and Brief Gateway turns a campaign brief into a validated get_products request. What Is Agentic Buying answers that one question without assuming you care how it is specified.

How this page handles evidence

Claims about a schema, a repository or a file pin to a full 40-character commit sha in the sentence that makes them, so the artifact behind the sentence cannot move under it: the zero-byte files above sit at 106ebd645668677ff8be458877a5684a5d229961, and the hub README that names six children is read at 3b20e67b568235d86c4025bec7c1301edd7466a0. Claims about a live site, a member count or a release cadence carry the date they were read, because those do move. This page has seven of the second kind and every one of them reads 16 August 2026: the tag check, the code search, the AAO member count, the member directory that would not open, the release count, the Adgentek disclosure, and Postindustria’s own site.

The eight repositories are this site’s boundary rather than IAB Tech Lab’s: the hub README names six children at that commit, and the hub itself plus iab-agentic-primitives make eight.

Trademarks belong to their owners. “AdCP” is AgenticAdvertising.org’s; “AAMP” and “IAB” are IAB Tech Lab’s.

Point at the file or the URL that contradicts a sentence and mail hello@agenticadlab.com. The page gets edited.