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.
In English, please
For years, the basic unit of marketing work was the campaign: plan it, launch it, measure it afterward. That is giving up its central role to something broader — an "action space" — the boundary of what an AI agent is allowed to do on its own, covering which audiences it can target, how much it can spend, and what checks and success measures it must clear before acting again. The marketer's job shifts from running campaigns by hand to designing that boundary and reviewing what happens inside it.
The proof is a three-day stretch in mid-June 2026: on Monday, a company called Hightouch published a post naming this new category of marketing software; on Tuesday, Databricks launched its own version, already piloted by HP and AB InBev; and on Wednesday, BlueConic acquired a company called Blueshift and rebranded the deal around the same label. Categories that used to take years to form took a single business week — a sign something real was moving, even though no one company invented it.
The software behind marketing campaigns has moved through four eras, each built around a different scarce resource: reach, then a single customer profile, then control over where the data lives, and now memory — what a business is trying to do with a customer, and what has already been tried with them. That era forked three ways: Databricks folded the software inside its own data platform, Hightouch bolted AI agents onto tools companies already use, and Treasure AI stripped down to bare plumbing that outside AI agents can plug into.
The more an agent's memory and ongoing decision-making run on one vendor's servers, the harder it becomes to leave — switching now means retraining that memory, not just rewriting a data connection. Fittingly, the executives named as the launch's flagship customers hold data and IT titles, not the chief marketing officer's — a sign the buying decision is drifting from marketing into the data team. And because the vendor claims here are running ahead of proof — no outside audit yet confirms any of it actually works better — every claim is worth checking before acting on it.
On this page
The Seventy-Two-Hour Category · How to Read the Numbers in This Series · Correct the Record Before Building On It · Generation Four: The Break Is in the Operator, Not the Plumbing · The Fork at Birth: Three Agentic CDPs, One Label · The Lock-In Swap, Round Two — and the Moved Buying Center · Monday Morning · Appendix: the receipts
A four-part series — Part 1: The CDP grew agents (you are here) · Part 2: The customer grew agents first · Part 3: The market grew agents on both sides · Part 4: The CMO owns the action space
| Term | Definition |
|---|---|
| Action space | The bounded set of goals, content, audiences, budget rules, policy checks, approvals, measurement gates, and execution surfaces an agent is allowed to act inside. |
| Agentic marketing | Marketing where software does not just recommend actions, but selects, sequences, negotiates, or executes them inside human-defined constraints. |
| CMO role | Not to approve every action, but to design the constraints and proof system that govern agent behavior. |
Start here if you only read one: Part 4. Then read Parts 1–3 for the evidence.
| If you are | Start with |
|---|---|
| CMO / CDO / CIO | Part 4 — the operating model and the procurement test |
| CDP / martech strategist | Part 1 — you are here |
| Search / content / SEO leader | Part 2 — the answer layer and demand interception |
| Adtech / programmatic leader | Part 3 — protocols, autonomy, and the fee stack |
The campaign used to be the unit of marketing work. In agentic marketing, the unit is the action space. And to be precise about the obituary in this part’s title: campaigns do not disappear; they stop being the primary operating unit.
Marketing is no longer organized around channels. It is organized around agents. The customer has an agent — it reads your email before the human does. The platform has an agent — it composes the ad, picks the audience, and increasingly places the buy. The seller has an agent — it negotiates inventory over a protocol. The CDP has an agent — as of last month, it has the name to prove it. Every surface marketing touches is becoming an agentic counterparty: something that reads, decides, remembers, and negotiates back.
One sentence has to sit at the top of this series, because the final part is built entirely on it: in a manual marketing world, measurement explains what happened. In an agentic marketing world, measurement decides what the agent is allowed to do next. Permission, not narration.
And the frame for everything in between: the question is not whether agents will run marketing workflows — they already do — but who defines the space they are allowed to act inside. That is the new CMO job: constraint design, not campaign execution.
Call it the Great Inversion: for a decade the data commuted to the channels; now agents commute to the data, and the channels send agents of their own to meet them.
The proof that the ground moved is a 72-hour sequence in June 2026.
The Seventy-Two-Hour Category
On Monday, June 15, 2026, Hightouch published “The next chapter: Agentic CDP” and named the category. On Tuesday, June 16, Databricks shipped one: CustomerLake, launched at the Data + AI Summit in Private Preview with HP, Circle K, AB InBev, and Getnet by Santander already inside. On Wednesday, June 17, BlueConic bought one, announcing its Blueshift acquisition as an “agentic CDP foundation” [Reported]. Categories used to take years to congeal; this one took a business week, and the trade press logged the whiplash in real time: “Last week, Hightouch published a blog post about its vision for an agentic CDP. One day later, Databricks announced CustomerLake” (MarTech, June 22 MarTech ↗). (The BlueConic customer count and the CEO’s moat quote are in the receipts.)
The compression is the evidence. When agents accelerate the work, they accelerate the market forming around the work — and the signal for brands is not any single vendor but that every vendor needed the same two words in the same week.
This series is the sequel to “The Composable CDP: How Customer Data Stopped Moving,” published June 19 — three days after the fourth generation announced itself. The prequel ended with the activation layer becoming the agent’s execution surface. This one asks what happens when every surface the data activates against grows an agent of its own: the buyer’s, the seller’s, and — first of all — the customer’s.
How to Read the Numbers in This Series
Three rules, stated once here and enforced in all four parts.
First, every load-bearing number carries a tag. [Confirmed] means a primary source or filing. [Vendor-claimed] means the company selling the thing supplied the number. [Reported] means credible secondary reporting with no primary document. [Author inference] means this series’ own arithmetic.
Second, every volatile ratio carries a source date. Crawl-to-refer ratios, ad run-rates, referral conversion, AI-search forecasts — these are fast-moving metrics, not evergreen truths, and a number without a date is a number pretending to be permanent.
Third, the finding that shapes every recommendation to come: across the public claims checked here, I found no independent public audit of agentic-marketing lift as of July 2026. The category is pre-audit. Remember that at every vendor meeting.
Correct the Record Before Building On It
The naive story — Databricks invented the agentic CDP — was this author’s working assumption going in. It does not survive the ledger. Hightouch’s AI Decisioning launched in August 2024 as a Snowflake Native App [Confirmed, PR Newswire]: goal-driven reinforcement-learning agents choosing content, timing, and channel on warehouse-native data. Roughly a half-dozen more vendors shipped agents-on-CDPs through 2025 — the full chronology, funding included, is in the receipts. The functionality shipped in 2024; the agents shipped all through 2025; the name shipped June 15–16, 2026.
The sentence that holds up: CustomerLake appears to be the first major agentic CDP shipped by the data-platform vendor itself — embedded inside the platform and governed in place under Unity Catalog, rather than layered on top of it.
Know your referees, too. The CDP.com glossary that formalized the category is bylined by Kazuki Ohta — CEO of Treasure AI, a combatant — and refuses to credit a coiner. Every taxonomy cited in this series has a position, including this one. (The full timeline table, the Aampe naming asterisk, and Databricks’ own no-first-claim footnote are in the receipts below.)
Generation Four: The Break Is in the Operator, Not the Plumbing
What actually changed? Put the four generations side by side and the answer is not the pipes.
| Era | Scarce asset | Operating model |
|---|---|---|
| DMP | Reach | Segment and syndicate |
| Packaged CDP | Profile | Copy and activate |
| Composable CDP | Control | Keep data in warehouse |
| Agentic CDP | State | Let agents decide within rails |
State means memory plus current intent plus what has already been tried — the operating context an agent uses before it acts. (One counting caveat: “fourth generation” works only under this series’ DMP-rooted lineage; CDP.com and MarTech count three because they omit the DMP.)
The asset shift is the story. Databricks’ founders’ manifesto Databricks ↗ — signed Argyros, Ghodsi, and Xin on June 17, under a section headed “Beyond the Composable CDP” — coins Golden Context: the Golden Record plus what the business is trying to accomplish right now and what has already been tried with this customer. In other words, state, productized: the strategic asset moves from identity to memory. Discrete campaigns dissolve into Infinity Campaigns Databricks ↗ — in the launch blog’s own words, “continuous, agent-driven engagement loops that analyze customer signals, decide on the next-best action, and act across channels” — the mechanism behind the obituary at the top of this part. Databricks does not soften the verdict either: presenting the launch, Tasso Argyros says of the weeks-to-months campaign model, “We do have to throw this model out of the window and come up with a new paradigm that is appropriate for this new era of buying agents.” Strip the branding from what replaces it — a goal, an agent-drafted brief as source of truth, guardrails (opt-outs, frequency caps, suppression lists), and per-customer decisioning inside them — and an Infinity Campaign is an action space wearing a campaign’s name badge [Vendor-claimed — launch video].
And the line that carries the argument: generation four does not remove the marketer. It moves the marketer from operator to governor.
The plumbing, meanwhile, is generation three verbatim — Databricks’ own launch blog describes Reverse ETL and bi-directional pipelines, the composable era’s last mile intact under the flagship of the fourth.
The lineage fits in one career: Tasso Argyros founded ActionIQ, the archetypal packaged CDP, sold data duplication for a decade, and now co-signs its abolition from inside a data-infrastructure company.
One honesty beat before building on any of this: “Humans still define the strategy, goals, and guardrails. Agents help scale execution” is the launch blog’s governance promise, and a promise is what it is — private-preview marketing until audited. The launch video’s demo does show real gates — the marketer approves the agent-drafted campaign brief before anything is built, campaigns save as drafts, and a pre-launch simulation runs the decisioning against a hundred random profiles, each with a readable log of the rules, signals, and guardrails behind its treatment [Vendor-claimed — launch video]. But demo gates are not audited gates, and the generational shift remains a direction confirmed across roughly ten vendors — not a verified property of one product.
The Fork at Birth: Three Agentic CDPs, One Label
Within a week the category forked into three incompatible architectures. Databricks re-bundles — the CDP pulled inside the data platform. Hightouch layers — agents atop the composable stack. Treasure AI goes headless — MCP infrastructure serving external agents.
“The CDP executes ideas, but it has never had one of its own.” — Hightouch, June 15, 2026 Hightouch ↗
Scott Brinker’s frame: for two years, application platforms became infrastructure — this is infrastructure marching on the application layer. What if the warehouse was the CDP? (Disclosure: Brinker published a Databricks-partnered research report in March 2026.)
The counterweight is coexistence, not conquest. MarTech’s verdict: “It’s unlikely either approach wins.” MarTech ↗ Hightouch sells to CMOs; Databricks sells to CIOs.
And notice who fled the noun: Salesforce, Adobe, Treasure Data, and MoEngage all shipped the agents and dodged the label — it took the vendor with no CDP legacy to reclaim the word. David Raab, who coined “CDP” in 2013, called it “a good move by Databricks, moving up the value chain.”
The Lock-In Swap, Round Two — and the Moved Buying Center
The prequel’s hardest-won finding gets promoted here: composable CDPs did not eliminate lock-in, they relocated it — from the CDP vendor to the warehouse. Now the weight moves again, upward, from custody to operation. Custody is unchanged or improved: “if you stop using CustomerLake, your data doesn’t go anywhere” (Datawhistl, June 20) [Reported]. What deepens is everything above the data: agent memory, always-on loops, governance semantics, the operating model itself — running on, and billing through, one vendor’s compute.
The composable era’s exit cost was rewriting SQL. The agentic era’s exit cost is retraining an organization’s institutional memory. Pricing is the tell: a “specialized, value-aligned consumption model,” no published meters — the campaign was a unit of billing as much as a unit of work, and Infinity Campaigns dissolve both.
“You cannot out-price a company that earns money whether or not you buy its product.” — Greg Kihlstrom, CMSWire, June 19, 2026
So the exit question to ask every vendor is about state, not data. Plant it here; Part 4 harvests it as a ten-question procurement test.
Close the ledger on this launch with the silences. Across all five primary launch sources — the launch video included: no mention of MCP, A2A, or AdCP. Three weeks in, no CustomerLake coverage found in the ad-trade press as of July 2026. The silence is not proof, but it is a signal: the launch was received more as data infrastructure than as a media-market event.
And close it with the org chart: both named launch-customer champions hold data titles, not CMO titles. The buying center has already moved to the data organization. Brinker’s name for the new job is Golden Context engineering — staff it inside marketing ops before the CDO staffs it for you. (The champions’ exact titles, the federation hedge, the Gartner secondhand prediction, the Ghodsi diplomatic fiction, and the services-bench roster: receipts below.)
Monday Morning
| From | Do this |
|---|---|
| Part 1 (this part) | Audit where agent memory, campaign state, and Golden Context will live. |
| Part 2 | Stop treating AI answer visibility as SEO. Treat it as demand interception. |
| Part 3 | Ask which surfaces are protocol-ready, copilot-only, or truly autonomous. |
| Part 4 | Add agent-state ownership, native holdouts, and decision logs to procurement. |
Appendix: the receipts
The full correct-the-record timeline. Aug 2024 — Hightouch AI Decisioning ships as a Snowflake Native App [Confirmed, PR Newswire]. Feb 2025 — Hightouch $80M Series C at a $1.2B valuation, raised explicitly “to build an agentic marketing platform” [Reported]. Mar 2025 — Adobe AEP Agent Orchestrator (agents GA Sept 2025); Zeta AI Agent Studio [Reported]. Apr 2025 — GrowthLoop’s agent-powered Compound Marketing Engine on BigQuery/Snowflake/Databricks, later self-branded “Agentic composable CDP” [Reported]. Jun 2025 — Salesforce Marketing Cloud Next (never says “agentic CDP”). Oct 2025 — Treasure Data AI Marketing Cloud [Reported]. Jan 2026 — Amperity Customer Data Agent, “the First Enterprise Customer Data Agent” [Reported]. Jun 15–16, 2026 — the name. Two honesty footnotes: Databricks never claims market-first anywhere in its launch materials — its only “first” is a capability claim (“true 1:1 experiences at an infinite scale”). And “first to name” carries a startup asterisk: Aampe’s blog uses “Agentic CDPs” with an on-page date of May 2024, but the page was updated June 24, 2026 and remains unverified pending a Wayback check; the G2 listing “Aampe Agentic CDP” is title-dated 2025. No naming-primacy sentence travels without that asterisk.
The BlueConic Wednesday, in full. The Blueshift acquisition was announced as an “agentic CDP foundation” for roughly 600+ combined customers, CEO Melissa Murray Bailey declaring that “real-time context is the new competitive moat” [Reported].
The continuous-loop quote. Ghodsi’s frame for Infinity Campaigns: “marketing stops being a series of campaigns and becomes a continuous loop” [Vendor-claimed, launch materials].
Referee dossier. The CDP.com glossary that formalized the generations (June 29) is bylined by Kazuki Ohta, CEO of Treasure AI; it says the term “entered industry vocabulary in mid-2026, with multiple vendors adopting it simultaneously” and credits no coiner. MarTech and CDP.com count three generations (“CDP 3.0”) because they omit the DMP.
ActionIQ afterlife. ActionIQ lives on as Uniphore’s “CDP Agent” after the ~$145M December 2024 acquisition [Reported].
The federation hedge in full. Lakehouse Federation is read-only foreign catalogs executed on Databricks compute under Unity Catalog governance [Confirmed, Databricks docs] — pulling Snowflake’s and BigQuery’s data through what Niederberger (Martech Therapy) called “Databricks’ control plane, its governance, its bill.” Kihlstrom’s CMSWire analysis agrees federation “significantly weakens the naive cross-cloud argument.” The compute-capture reading — the CDP as a compute-demand generator, a revenue coupling no composable-era CDP had — is Martech Therapy’s reading, not established fact [Reported].
Kihlstrom’s playbook move. Inventory your current CDP’s capabilities against what CustomerLake gives you inside a contract you already pay for; pilot before renewal as negotiation leverage.
The skeptic’s concession. Even perform.digital, the designated skeptic, concedes the real mechanism: the customer profile becomes something software acts on instead of something a person reads.
The silence roster in full. Three weeks after launch: no CustomerLake coverage found on AdExchanger, Digiday, Ad Age, ExchangeWire, or Marketecture as of July 2026 [Confirmed by site-restricted search]. No agency or holdco said a word. The services bench is consultancies — Accenture, Deloitte, Slalom, Stitch, Lovelytics — no media agencies. Retail media went unmentioned despite Circle K in private preview and Magnite, Unity, and Snap on the partner list. Only ContentGrip (June 19) [Reported] claims third-party agents can plug in via APIs or MCP.
The moved buying center, receipts. HP’s champion: Global Head of Marketing Technology and AI Enablement. Getnet’s: Chief Data and AI Officer. Raab’s political warning: “IT gains more control, which marketers may not appreciate.” The Gartner-attributed prediction — 80% of net-new enterprise CDP deployments embedded in or composable with data platforms by 2030 — is secondhand via CMSWire; no primary document located as of July 2026 [Reported]. Ghodsi told ITdaily “We are not a competitor to Salesforce… CustomerLake is about activation” (likely a translation), while ITdaily’s own verdict is that the real overlap is Data 360 and Agentforce.
Disclaimer ledger. “Humans set goals, review outcomes, and stay in control” is private-preview marketing; CustomerLake has no public docs; whether agent state, Unity Catalog policies, or campaign loops are exportable is unverifiable today [Author inference].
The marketer’s agent is a week old as a named category. The customer’s agent has 900 million weekly users and reads your campaign before any human does. Part 2: the audience deployed first — and the answer layer is already selling ads against it.