The news
A new report covered by Marketing Tech News finds that agentic AI in marketing workflows is gaining real traction — with McKinsey estimating it could support up to two-thirds of current marketing activities, including synthetic audience testing. The report flags that the primary constraint isn't model capability — it's system interoperability, specifically whether your activation systems can expose reliable APIs for agents to act through.
Our take
This framing is right, and it plays out repeatedly. Teams get excited about agentic AI — the idea of an autonomous system that can research a prospect, draft a sequence, update a CRM record, and flag a deal risk without being babysat — and then hit a wall the moment they try to connect it to something real.
That wall is almost never the AI. It's that HubSpot's workflow triggers don't expose what the agent needs. Or that the campaign data lives in a spreadsheet that nothing can reliably read. Or that there's no documented process for the agent to follow because that process only exists in someone's head.
The McKinsey number — two-thirds of marketing activities — is a ceiling, not a floor. Most teams will capture a fraction of that because they're trying to automate chaos. Agentic AI doesn't impose order on a messy system. It amplifies whatever structure already exists.
Synthetic audience testing is the use case worth watching closely. The idea that you can simulate how a segment responds to messaging before spending budget is genuinely useful — but it requires clean audience definitions, documented messaging frameworks, and a content distribution layer that an agent can actually write to. That's three layers of operational readiness most marketing teams haven't built yet.
The report's call for "flexible model-serving infrastructure" is real but slightly misses the point for most GTM teams. The deeper requirement isn't infrastructure — it's documented processes that an agent can follow. That's the unglamorous work that determines whether your agentic investment returns anything.
So now what?
Before chasing agentic workflows, run a fast readiness check on three things:
- API coverage: Can your core stack (CRM, MAP, CMS) expose the actions an agent would need to take? If you're not sure, that's your answer.
- Process documentation: Pick one workflow you want to automate and write down every step — including the judgment calls a human currently makes. If you can't write it down, the agent can't run it.
- Data cleanliness: Synthetic audience testing and personalization agents are only as good as your segmentation data. Audit one segment before you build anything.
The teams who will capture that McKinsey ceiling aren't waiting on better models — they're getting their operational house in order right now.