- Tension: Every platform is racing to compress execution time from days to minutes, but speed is becoming table stakes while the thing actually worth owning, the intelligence deciding what to launch, is quietly consolidating somewhere else.
- Noise: Vendor pitches compete on interface and velocity, “campaigns in minutes, not days,” which distracts from the fact that most of these agentic tools are running on the same handful of foundation models underneath.
- The Direct Message: Whoever owns the proprietary context, the data, the decision logic, the memory of what actually worked, owns the value. The software wrapper around it is becoming a commodity.
To learn more about our editorial approach, explore The Direct Message methodology.
Ninety percent of marketing organizations now use AI agents somewhere in their stack, according to Scott Brinker’s Martech for 2026 research, cited by EMARKETER. Not experimenting with. Using. The first half of 2026 was the period when that shifted from a pilot-program talking point to an operational default, and the pitch every platform converged on was speed: describe a campaign in plain language, and the software assembles, drafts, and launches it, turning what used to take days of manual configuration into a few minutes of review.
That compression is real. It’s also, on its own, not much of a moat.
The days-to-minutes pitch every vendor is making
Salesforce has Agentforce 360. Adobe has agentic capabilities layered across Experience Cloud. HubSpot, Braze, and a wave of AI-native challengers like Tofu all sell some version of the same promise: set the goal, and the system builds the campaign. Vendors report execution speed gains ranging from several times faster to, in some marketing materials, close to an order of magnitude. Gartner has forecast that 40% of enterprise applications will embed AI agents by the end of 2026, up from under 5% the year before.
The result is that “our AI agent launches campaigns fast” stopped being a differentiator sometime around the first quarter. It became the baseline expectation, the same way free shipping did for ecommerce two decades ago. Once every vendor promises it, the promise stops being the thing buyers are actually choosing between.
Same foundation models, different logos
Most coverage of this shift underplays one detail: a lot of these competing platforms are drawing on the same small set of underlying models.
As EMARKETER’s own analysis put it, many providers now rely on the same OpenAI or Anthropic models, making their agentic offerings “almost indistinguishable” from one another at the model layer. The interface is different. The workflow branding is different. The reasoning engine underneath is frequently the same one your competitor is also renting.
That reframes the whole conversation about what these platforms are actually selling. If the intelligence itself, the raw model capability, is increasingly a shared utility, then the fight over “who has the best AI” is a fight over something that isn’t really proprietary to any single vendor. What’s left to compete on is what sits around the model: the data it’s trained and grounded on, the workflow it’s embedded in, the institutional memory of what a specific brand’s audience has actually responded to.
What actually differentiates now
Content production agents are the most widely adopted category, used by 68.9% of organizations, according to Brinker’s research, followed by audience discovery agents at 40.8%. Those are the visible, easy-to-demo capabilities. But the deeper layer, the one that determines whether an agent’s output is actually good rather than merely fast, is the quality and structure of the data it has access to.
This is where composable martech architecture comes back into the conversation. Customer data platforms have been shrinking as the center of the stack, dropping from 26.9% to 17.4% of B2C martech architecture, according to chiefmartec’s 2025 marketing technology landscape research, with the underlying capability migrating either to a cloud data warehouse or directly into engagement platforms. That migration matters because it’s really a migration of leverage. Whoever sits closest to the clean, governed, first-party data, rather than whoever sits closest to the campaign-builder interface, increasingly controls what the AI layer is actually capable of producing.
The buyer-side problem nobody priced in
All of this vendor-side speed race is happening while a second, arguably more disruptive shift gets less airtime: buyer-side AI agents. ChatGPT, Perplexity, Claude, and Gemini are increasingly where prospects research and evaluate brands before a marketing team’s own campaign ever reaches them, bypassing search, social, and owned-website discovery entirely. McKinsey has estimated that 20% to 50% of traffic from traditional search channels is now at risk from this shift.
Sixty-three percent of B2B marketing leaders say they recognize this change in buyer behavior. Only 14% say they’ve actually adapted their content and discovery strategy to account for it, according to Scott Brinker’s own preview of chiefmartec’s Martech for 2026 research. That gap, between recognizing a structural shift and doing something about it, is a wider vulnerability than any campaign-launch speed metric captures. A brand can compress its insight-to-launch cycle to nine minutes and still be functionally invisible to the AI agent a buyer consulted before ever encountering that campaign.
Where the intelligence layer is actually consolidating
Put those two shifts together and the real fight comes into focus. It isn’t happening in the feature-comparison chart between marketing automation vendors, most of whom are converging on similar speed and similar underlying models. It’s happening one level up, in who controls the context those models reason over: the first-party data, the memory of prior campaign performance, the structured content that buyer-side agents choose to cite or ignore.
Software that launches campaigns fast is becoming infrastructure, necessary, table stakes, and increasingly interchangeable. The intelligence layer, the part that actually decides what to say, to whom, and whether an outside AI agent considers a brand worth recommending in the first place, is where the value is quietly relocating. Marketing teams still benchmarking vendors on launch speed are measuring the part of the stack that’s already commoditizing, while the part that will determine whether their brand shows up at all sits mostly unmeasured, one layer up, un-owned, and up for grabs.
