Agentic Advertising.

2026 · 35 essays

1. Oct 17
   I Gave Away Ten Billion Tokens

   Building four open-source systems took about 9.9 billion tokens of AI compute, most of it on a working signals agent for AdCP, an open standard for agentic advertising. The code is MIT-licensed, so the next team can fork it instead of spending that compute again.

2. Sep 25
   Sixteen Fridays, One Question

   Two seasons of the Friday Thought Experiment put 16 questions and 64 options to the same room and collected 786 ballots across all sixteen. Read as one ledger, they were one question asked sixteen ways: when an agent acts for you, who can say no, and who can be held to the answer? Season 1 found the institution that question needs. Season 2 found where the money settles once it exists. In between, the room changed its mind in public, and the reversals are the most useful data in the set.

3. Sep 25
   Where the Surplus Settles

   The Season 2 finale. Eight Fridays asked where advantage, money and power move once agents can act, and the closing question puts the season's own weeks on the ballot: objectives were Week 1, access Weeks 2 and 7, execution Weeks 3 and 4, verification Weeks 5 and 6. The room is ranking its own arguments without being told that is what it is doing. Two findings come out of the ledger rather than the news. The season's opening answer lost to itself — proprietary data took 78% of Week 1 when the question was what creates an edge, and 13% of Week 6 when it was what wins a bid — and two of the finale's four control points were already on the Week 1 ballot in different words, finishing at 4% and 7%, two of the three lowest scores of the season. The room prices what it imagines holding and discounts what it imagines going through. The week's term is Where the Surplus Settles, and the criterion is not refusal but binding refusal: surplus settles where the refused party has no substitute it can route to. That rules out the tidy version of this argument. Verification is not structurally excluded from capture, since credit rating agencies are pure examiners running operating margins around 64%; an examiner captures in proportion to how much its verdict moves the price, and in advertising it moves none — DoubleVerify measured 9.5 trillion transactions in FY2025 at seven cents per thousand, falling. Amazon told Perplexity no in writing, in code and in court and lost, not for want of authority but for want of a way to tell who was at the door, which makes refusal an identity problem before it is an access one. And binding refusal attracts the state: three federal courts found access rents and three declined to remove them, while the ad tech remedy left the asset in place and took away the refusal.

4. Sep 18
   The Record Is Not the Relationship

   Season 2's seventh poll asked who owns the customer relationship when an agent chooses the brand, and the four options turn out to be four different senses of the word own. The consumer owns the intent, the brand owns the fulfilment, the agent platform owns the interface, the data holder owns the record — and only one of those can be exercised without anyone's permission. The week's term is Interface Capture: the party that receives the intent accumulates the relationship whoever fulfils the transaction and whoever signs the contract, because it is the only party that can open the next conversation without asking. The test is not who was present at the sale, it is who can start the next one unilaterally. Shopify publishes the list of checkout blocks that do not load in an agentic direct checkout, and it is loyalty, the consent block and the client-side pixel — the merchant's entire apparatus for turning a buyer into a customer. Meanwhile the Agentic Commerce Protocol does hand the merchant the buyer's real name, email, phone and address. The record arrives; the permission does not. Booking Holdings now files that mechanism in its own risk factors, two federal appeals courts decided the compellability and the defensibility of the position a day apart in August 2026 and the interface won both, and AdCP's own data-protection analysis assigns the first-party role to the interface holder rather than the brand. My vote went to the brand, which is the party with the best claim in equity and the worst claim in mechanism, and I say why I am not changing it.

5. Sep 11
   The Layer That Didn't Ship

   In 2025 the industry agreed it needed a shared layer for exchanging user context between agents, LiveRamp proposed one, donated it to IAB Tech Lab in late 2025, and it was renamed Agentic Audiences by early 2026 and filed under AAMP. Read at the repository rather than the press release, that layer has eleven commits since its creation in October 2025, no releases and no tags, two pull requests still open, a spec still headed Draft v0.1, and a wire format string that still carries the retired name. Its one working integration — the OpenRTB community extension on Segment.ext.aa and the Prebid.js RTD module — shipped inside a four-week window between March 22 and April 14, 2026 and has had nothing added since. Meanwhile the job got done twice somewhere else: AdCP's Trusted Match carries contextual embeddings at serve time against a 50ms provider timeout inside the router's latency budget, and AAMP's in-bidstream runtime ARTF reached v1.0 on July 20, 2026 with gRPC bid mutations under tmax. Neither references the other's embedding vocabulary. The mechanism is not a governance failure, it is a latency budget: an embedding is only worth carrying if it arrives before the auction closes, so every decision about dimensions, quantisation and timeout behaviour is a property of a runtime, and whoever holds the clock ends up holding the vocabulary. A neutral layer donated to a standards body is not the same thing as a layer that ships. The general test, applied here to AdCP as well, is commits, tagged releases and whether the one integration that exists is still maintained — and it closes with a checkable prediction about what would prove Agentic Audiences is maintained rather than parked.

6. Sep 11
   The Evidence Premium

   Season 2's sixth poll asked what wins the bid when every seller can claim performance, and 37 ballots put Verifiable outcomes on top with 54%. Lowest price took 21%, Proprietary data 13%, Exclusive supply 10%. The finding isn't the winner, it's the comparison: seven days earlier the same room priced verifying results at 21% when it appeared as a job an intermediary does, and the identical idea took two and a half times the support once it was worded as a property of the offer. Sold as a service, proof polls like overhead. Built into the offer, it wins the bid. The week's term is the Evidence Premium: the price gap between a claim a buyer has to take on faith and the same claim carrying proof the buyer's own side would accept. It is not collected by whoever has the best number, it goes to whoever can produce a number the counterparty accepts without reopening the negotiation, and almost no seller holds that arrangement. The accreditation layer already built this machinery one layer down — research proves a method is valid, accreditation proves a number is transactable — and the MRC's annual audit clock says the premium is a subscription rather than a purchase. Nothing accredits an agent-generated claim yet, which is why the 21% who picked price are reading the next two years correctly. My vote went to Verifiable outcomes, with an interest to declare and a concession about who ends up funding the check.

7. Sep 6
   The Risk Residue

   Season 2's fifth poll asked which intermediary function stays valuable once agents transact directly, and 19 ballots put Assuming risk on top with 42%. Finding counterparties and Verifying results tied at 21%. Negotiating price finished last at 16% — one week after a third of the room called price the hardest thing agents will negotiate. That gap is the essay: hard and worth paying a middleman for are different properties. The week's term is the Risk Residue. Strip the information work out of intermediation — the finding, the matching, the price discovery — and software does all of it at zero marginal cost. What is left is balance-sheet work, because software cannot absorb a loss. Absorbing a loss needs capital, a legal person to hold it, reserves priced against the miss, and an incentive not to lie about all three. The intermediary of the agent era looks less like a marketplace with better matching and more like a small clearinghouse: escrow, guarantee, spread. Verification is the interesting near-miss — scarce only while it stays independent, because an agent employed by one side of the trade cannot check the other side's numbers. My vote went to Assuming risk, and I changed it mid-week from Negotiating price, which is the argument this essay has with itself.

8. Aug 28
   Negotiating the Nouns

   Season 2's fourth poll asked what buyer and seller agents will negotiate hardest, and 55 ballots produced the tightest finish of the season: Success criteria at 34%, Price at 32%, Audience definitions and Liability at 16% each. Read as a block, two-thirds of the room said the hardest negotiation is no longer the number — it's the nouns around the number. That's the week's term: Negotiating the Nouns. Automated negotiation is mostly semantic negotiation wearing a price tag. Price is the one term on the ballot machines already clear — fifteen years of real-time auctions settled it to the millisecond — and it stayed easy only because every other term was held still by convention: the audience was a list name on an insertion order, success was handled at the QBR, liability was boilerplate nobody read. An agent can't execute ambiguity. Every term has to parse before a dollar moves, so the negotiation migrates from the price line into the definitions — what counts, whose count settles it, over what window, against what baseline. My vote went to Success criteria, and it won by two points; the 32% who picked Price weren't wrong either, because every noun left unsettled eventually presents its invoice as a price dispute.

9. Aug 21
   Nobody Sells an Outcome

   Season 2's third poll asked what autonomous buying agents will ultimately buy, and 73 ballots produced the clearest majority of the season: Outcomes at 57%, with Impressions at 27%, Audiences at 8%, Attention at 6%. The room voted for the one unit on the list that nobody can currently sell, because selling an outcome means warranting it, and warranting it means pricing the failure case. That pricing function — Outcome Underwriting — is the institution the whole result quietly depends on: a definition both sides accept, a measurement both sides trust, a settlement window, and someone who pays when the outcome doesn't happen. The first three are measurement work. The fourth is capital. Outcome optimization exists inside the walled gardens; outcome insurance does not. Owned loops reprice faster and sometimes carry action-level variance, but no scaled platform guarantees the advertiser's business result. My vote went to Outcomes anyway — for where the market ends, not for what's tradable now — and the 27% who voted Impressions weren't nostalgic; they named the only unit with a working settlement stack, which is why it gets demoted to the receipt rather than killed.

10. Aug 14
    The Shelf You Can Restock

    Season 2's second poll asked what becomes the new shelf space once agents decide what gets considered, and 16 ballots produced a tie: Retrieval results and Paid recommendations at 31% each, Training data and Default integrations at 18%. Sixteen votes was the smallest turnout of the run to that point and settles nothing, but the grouping is worth a paragraph — the room picked the two routes into a consideration set that an operator can change inside a planning cycle without anyone else's approval. The Restock Test sorts all four by how long a change takes to land and whose approval it needs. My vote went to paid recommendations, for one reason: it's the only one of the four with a price, and a price drags an owner, an auction, and a contract along behind it.

11. Aug 14
    The Mandate Finished Last

    Season 2 opened by proposing that once every advertiser has a capable buying agent, advantage moves to the quality of the mandate. Forty-two operators voted, 78% chose proprietary data, and the mandate came last at 4%. The room is right about the market as it stands, and the reason has nothing to do with which asset actually decides the outcome.

12. Aug 3
    The Answerer of Record

    The Season 1 finale. Eight Friday polls were one question asked eight ways — who issues the credential that lets an agent act? — and the room's own ballots supply eight structural constraints on the answer. Run the three candidates against all eight and only one survives: protocols carry the credential, constitutions bound it, contracts create it. The issuing layer is an Answerer of Record — the agentic successor to adtech's seller-of-record — because trust is conferred, never computed.

13. Jul 31
    AdCP vs AAMP: The Retrofit and the Greenfield

    Two agentic advertising standards shipped within 48 hours of each other in late July — AdCP tagged v3.1.8, IAB Tech Lab announced AAMP 2.3. The trade press calls it a war and mislabels the battlefield. Read at the repository level, the real difference isn't altitude (both negotiate deals; both touch the impression path). It's ancestry: AAMP retrofits agents onto advertising's existing rails — OpenDirect, OpenRTB, the ad servers and DSPs already in production — while AdCP writes a new campaign-lifecycle protocol with trust in the wire format. Every claim verified against the repos and specs, both directions, with the press-release-vs-repository deltas printed for each side.

14. Jul 31
    The One Standard That Clears the Deal

    Season 1's last poll asks which standard matters most in five years: Identity, Measurement, Agent Protocols, or Trust & Accountability. This essay's answer: three of the four only reduce friction — a market can run without them, just more slowly and at more risk. Only one clears or blocks the transaction outright. That's the test that actually separates a standard from infrastructure, and it's independent of which option the room eventually picks.

15. Jul 24
    Business Outcomes Isn't a Number

    Friday's poll asked what agents will optimize for, and 54% of 51 ballots said Business Outcomes — more than Lowest Cost and Highest ROAS combined. This essay's answer to why: Business Outcomes was never a fourth number competing with the other three. Lowest Cost, Highest ROAS, and User Satisfaction are each a single scalar an optimizer can directly maximize — and a single scalar is exactly what gets gamed. Business Outcomes can't be scalarized the same way, which is the point: the room didn't vote for a better objective function. It voted against handing the agent one at all.

16. Jul 21
    The Loop Closed Inside the Wall

    Snowflake just shipped a governed agentic loop between the warehouse and Meta — conversion signals out, campaign context back, an agent that prepares and a human who approves. It is the closest thing yet to this site's thesis running in production. It is also the exact pattern this site warned about: the loop closed, but it closed inside one walled garden, on one vendor's proprietary surfaces.

17. Jul 20
    The Intelligence Economy

    At Cannes, OpenAI declared the shift from the attention economy to the intelligence economy — and put a rate card on the answer. Six months of verified record tell a stranger story: the product is a familiar sponsored card, the price fell as the pitch rose, the minimums collapsed from $200K to zero, and every trust metric is self-reported with no referee. Meanwhile the ecosystem split seven ways on the same question: OpenAI all-in, Microsoft selling quietly, Google hedges by surface, Meta monetizes around the assistant, Grok announces, Perplexity exited over trust, and Anthropic bought Super Bowl airtime to refuse. These aren't seven marketing strategies. They're seven objective functions — and the forecast gap between OpenAI's $2.5 billion and the category's sub-$1 billion is where the argument gets settled.

18. Jul 13
    Measurement for Agentic Commerce

    The agent doesn't click, doesn't see ads, and buys anyway. A breaks/survives/emerges triage of commerce measurement, and the instrumentation to build now.

19. Jul 12
    Fire the Dashboard, Hire the Agent

    A complete build recipe: GA4, Search Console, Claude, and GitHub wired into an editorial agent that reads a month of data, writes five ranked recommendations, grades its own previous advice, and pings you — on two cadences, for pennies. Five APIs, zero dashboards, one afternoon.

20. Jul 10
    Whoever Writes the Constraints Owns the Outcome

    Friday's poll asked who should be accountable for an agent's decision, and the room put half its ballots on The Agent Owner, with Shared Responsibility a distant second. This essay takes that answer seriously and then sharpens it: shared responsibility is what accountability looks like before a contract assigns it. In practice, accountability follows the action space — the vendor owes enforceable rails and honest measurement, the user owes informed approval, but ownership concentrates on whoever authors the constraints the agent acts inside. Which is why my vote went to The Agent Owner, and why the fix is infrastructure, not sentiment: versioned operating rules plus an append-only decision ledger turn 'shared' from a shrug into an audit trail.

21. Jul 10
    The Metric That Was Never About You

    Friday's poll asked which marketing metric breaks first once agents become the primary buyers. This essay's answer is Attention, and for a mechanical reason: it was never a measurement of the market — it was an inference about something no one could directly observe. Reach & Frequency and Attribution make the same category error one layer removed. ROAS survives by accident, not merit. And the record that replaces all three isn't a better guess — it's the first time anyone could just read the answer instead of inferring it.

22. Jul 5
    Adtech and Martech Were Never Two Industries. They Were Two Copies of the Customer.

    Adtech versus martech was never a strategy — it was a custody arrangement: two industries built on two copies of one customer, one reaching through rented access, the other through owned relationships. The agentic CDP did not end the split; it exposed how little of it was strategic. One governed customer memory — read, written, and learned against by agents on both sides — is absorbing the operating boundary between paid and owned. Sequel to 'The Composable CDP: How Customer Data Stopped Moving.'

23. Jul 5
    The CMO Owns the Action Space

    Measurement stops explaining the past and starts issuing permissions. The post-agentic CMO's deliverable is not a channel plan — it is guardrail architecture. Here is the ownership split, the operating loop that already works, and the ten-question procurement test.

24. Jul 5
    When the Buyer's Agent Meets the Seller's Agent

    A natural-language brief in Claude became live CTV spend through a protocol handshake. Ten surfaces are becoming agentic counterparties — but autonomy and interoperability are inversely correlated, so far, and the consolidation ledger shows exactly what agents are hungry for.

25. Jul 5
    The Customer's Agent Arrived First

    Before any CMO deployed an agent, a billion customers did. The click did not die of natural causes — the answer layer intercepted intent, then started selling media against it. And a whole measurement industry formed before anyone validated the metric.

26. Jul 5
    The Campaign Is Dying. The Action Space Is the New Marketing Operating System.

    A category got named on Monday, shipped on Tuesday, and acquired on Wednesday. But the fourth-generation CDP is the entry point, not the story: the CMO's job is moving from campaign execution to constraint design.

27. Jul 3
    What Agents Can't Manufacture

    Friday's poll asked what becomes scarce in a market run by agents. The obvious answer is the last decade's winner — the biggest data. Wrong layer. Every cycle relocates scarcity rather than abolishing it, and an agentic market makes data, compute, and distribution all self-suppliable — leaving one asset a machine is structurally incapable of minting for itself: trust.

28. Jun 26
    Recommendation or Decision? The Veto That Doesn't Scale

    Friday's poll asked when an AI recommendation becomes an AI decision. Most said 'when humans stop reviewing.' Right answer, wrong tense — in agentic ad tech that line is already behind us, and the surviving signature isn't a decision-maker, it's a name for the post-mortem.

29. Jun 20
    The First AI-Native Fraud Isn't a Fake — It's Laundered Authority

    Friday's poll asked what the first truly AI-native fraud is. The crowd answered with familiar villains. The trap was the question: the new scarcity isn't authenticity, it's accountability.

30. Jun 13
    Do We Still Need Standards When Machines Understand Context?

    The crowd voted 'more than ever' — but the real question isn't whether AI kills standards. It's which standards survive when meaning goes to zero and accountability becomes the scarce thing.

31. Jun 13
    Notes from the Implementer Seat

    What running a live AdCP signals agent teaches you that reading the spec never will. Revised September 2026: a privacy gap I filed against Trusted Match shipped in AdCP v3.1.22 — on stricter terms than I proposed, in a maintainer PR rather than mine — and two threads listed here as open proposals have since closed as completed.

32. Jun 7
    Signal Containerization: The Next Abstraction Layer for Agentic Advertising

    A framework for packaging buyer intent, semantic meaning, provenance, privacy policy, activation rules, and measurement logic into executable advertising signal containers.

33. Feb 23
    Stop reading about "The Agentic Future"...

    Start building it. A bi-monthly tutorial series on shipping real GTM agents in Rube, Claude, n8n.

34. Feb 16
    AI Agents in Ads need a "Common Language"

    Vector alignment as practical infrastructure — projectors, capability discovery, and the handshake between buyer and seller agents.

35. Feb 16
    The Missing Link in Ad Tech: How UCP & AdCP Are Finally Unifying the "Alphabet Soup"

    How UCP and AdCP retire ad-tech's "Tower of Babel" — Architect (intent) and Engineer (math) of the agentic stack.

2025 · 9 essays

1. Dec 1
   Part 4: Why Your Campaign Reports Are Already Obsolete

   From post-mortem PowerPoints to continuous learning systems. The 90-day learning lag is killing performance — and attribution theater has us making million-dollar calls on correlation dressed up as causation. Here's what changes when learning happens in milliseconds, not months.

2. Nov 24
   Part 3: The 100-Millisecond Battlefield Where AI Makes or Breaks Campaigns

   From rule-based bidding to intelligent real-time decisions. Why a luxury brand's AI refused to bid against layoff articles — and what changes when programmatic systems start to reason in 100 milliseconds instead of executing rules.

3. Nov 17
   Part 2: The Death of the Quarterly Campaign Planning Cycle

   How AI is transforming static strategies into living intelligence. A real CPG case found $3M in incremental revenue from a behavioral pattern no human would ever connect — and the campaign-planning era that made it possible is over.

4. Nov 12
   Four Article Series: The Agentic Transformation of Digital Advertising

   Part 1 of 4. The quiet revolution from tools that execute to systems that think — and the human role that survives it.

5. Nov 3
   AdCP: The Open Standard for Agentic Advertising

   Connect once. Operate anywhere.

6. Sep 1
   Embeddings: The Next Frontier in Advertising?

   Dense numerical representations as the post-ID connective tissue — lookalikes, contextual targeting, cross-media without identifiers.

7. Apr 25
   Blending Freytag’s Pyramid and Plutchik’s Wheel with MCP and Agentic AI

   Narrative structure × emotional resonance × autonomous agents. How story-aware ads replace impression-level metrics.

8. Apr 14
   MCP. A2A. OMG. The New ABCs of Advertising AI

   Model Context Protocol meets Agent-to-Agent. How AIs talk to tools and each other to ship DCO at machine speed.

9. Mar 31
   Contextual 3.0 - How Context Agents Are Transforming Media Planning

   From keyword matching to fluent reasoning. Five categories, twelve evaluation questions, one machine-readable JSON schema.