Marketing software spent the last two years bolting a chat window onto every product and calling it AI. This week, two separate announcements point at something different: platforms and standards bodies are now building for AI agents that complete work directly, not assistants that wait to be asked.
Norway-based marketing platform Ibexa launched its Agentic Marketing Platform in beta on September 23, giving marketers a single place to run AI agents across the tools they already use rather than adding one more standalone product. On the same week, IAB Tech Lab published a new specification, OpenProposal, that lets buy-side and sell-side agents exchange campaign proposals in a shared machine-readable format. Different corners of the stack, same direction: agents that act inside a workflow, not agents that summarize one.
From copilot to operator
Ibexa’s pitch starts from a familiar complaint. “AI promised to make marketing easier,” the company said in its launch release. “Instead, two things broke at once.” Outside the organization, AI is reshaping how people discover and choose products, so the funnel marketing teams built their attribution around no longer matches the journey anyone actually takes. Inside the organization, the stack kept growing, with dozens of point tools each getting a bolted-on copilot that can see inside its own four walls and nowhere else.
The Ibexa Agentic Marketing Platform tries to close that gap by connecting to more than 200 marketing tools and letting agents pull data, draft and translate content, publish pages, and adjust live campaigns, rather than just answering questions about them. “For the first time, marketers can bring AI agents directly into the flow of their work and ask them to take a campaign from idea to execution,” said Gregory Becue, Chief Product Officer at Ibexa. A landing page that took two to five days to localize now takes 15 to 30 minutes, according to the company, with the marketer reviewing the finished output rather than each individual step along the way.
Ibexa is explicit that this is not full autonomy. “Control and governance belong in the hands of the people doing the work, not least when it comes to data activation and AI,” said Bertrand Maugain, CEO of Ibexa. Every agent runs inside guardrails a team sets, is tied to the identity and permissions of the person who launched it, and is accountable to a budget. Marketers pick the underlying model themselves, including private and European options, so the platform is not locked to a single AI vendor.
The same shift on the buy side
IAB Tech Lab’s OpenProposal spec addresses a narrower but related problem: the RFP process that decides which publishers get considered for a campaign at all. Today a media team might have time to evaluate proposals from ten or twenty publishers for a single brief while hundreds of others never see it, largely because reviewing and comparing proposals by hand does not scale. OpenProposal defines eleven standardized data categories, covering everything from audience reach and commercial terms to measurement and prohibited adjacency, so that buyer and seller agents can assess a proposal automatically instead of a person reading a deck.
“Agents can review far more options than a human team ever could, which is what makes them so powerful,” IAB Tech Lab wrote in describing the specification. “But scale alone is not enough.” The spec sits on top of existing IAB Tech Lab standards, including AdCOM and the Deals API, and is designed to work across programmatic guaranteed, preferred deals, open auction, and linear formats alike, which is what makes it a genuine industry layer rather than one vendor’s feature.
What it means for the marketing leader
Both moves are evidence of the same underlying shift: agentic AI in martech is becoming an execution layer that touches production and spend, not a drafting tool that sits beside it. That raises the bar for what a CMO needs to ask a vendor before adopting anything labeled “agentic.” Whose identity does the agent act under. What is the agent authorized to spend or publish without review. Which model is it running, and can that be swapped for compliance reasons. HubSpot’s own September platform update, covered here, showed the same pattern from a different vendor: point-solution AI features getting absorbed into one shared context layer rather than staying siloed per tool.
The RFP layer matters just as much as the execution layer, because it changes who gets seen. A platform that can evaluate hundreds of proposals instead of twenty is also a platform that decides, by whatever criteria its agent is tuned to, which publishers make the shortlist. That is worth tracking alongside the productivity story, not after it.
The skeptic’s case
None of this guarantees the agents work as advertised. Ibexa’s own numbers, a landing page in 15 to 30 minutes instead of two to five days, come from the company that built the platform, not an independent benchmark, and the beta has not yet reached general availability. OpenProposal, similarly, only standardizes the format proposals travel in. It does not guarantee that the buyer’s agent will fairly consider a small publisher’s response over a familiar large one, and IAB Tech Lab’s own framing acknowledges the specification is a foundation for agentic workflows to build on, not a finished marketplace. Vendors have oversold “agentic” before. The distinguishing detail worth checking in each case is whether the agent is actually authorized to take an action end to end, as both of these are, or whether it still hands the real work back to a person once the demo ends.
What to watch next
Ibexa’s Agentic Marketing Platform reaches general availability on January 12, 2027, after a beta period the company says has already run ahead of the places it set aside for early adopters. OpenProposal builds on the same trust and transparency work IAB Tech Lab has been doing through its Agent SDKs and registry, which points to a maturing rather than one-off effort. Demandbase’s Mojo launch is a third data point in the same direction: agents that carry out a defined job end to end, inside guardrails a team controls, are becoming the default shape of “agentic” in martech, whatever the product category.
Source: Ibexa
