The news
Presidio published a piece arguing that the real AI value problem isn't a lack of agents — it's a lack of orchestration. Citing an MIT study finding that 95% of generative AI pilots fail to deliver measurable financial return, they make the case that isolated AI investments don't compound — they create coordination overhead. The fix, they argue, is an orchestration layer that routes work, maintains context, enforces policy, and captures outcomes across agents.
Our take
This maps exactly to what's happening in GTM and marketing orgs right now. Teams have a ChatGPT tab, a HubSpot AI feature, a Clay workflow, a Notion AI page, and a Zapier automation their ops manager built six months ago — and none of it talks to each other. Every "AI initiative" is its own silo. The human in the middle is still doing the context-switching and the handoffs. That's not AI leverage. That's AI theater.
The Presidio framing of "agent sprawl" is the right diagnosis. But the fix they're describing — a purpose-built orchestration layer with shared memory and centralized governance — sounds like an enterprise IT project, which means most GTM teams will read it as "not for us."
It is for you. The orchestration problem shows up at every scale. Even a three-person marketing ops team running five disconnected automations is experiencing agent sprawl. They just don't have a name for why their AI investments feel like they're treading water.
The difference between teams that are getting compounding value from AI and teams that aren't isn't the number of tools. It's whether those tools share context. A lead enrichment workflow that feeds into a scoring model that triggers a personalized sequence is orchestration. Three separate tools that each do one of those things, manually connected by a human, is not.
This pattern breaks in the same way every time: the automation works in isolation, but the moment it needs to hand off to the next step — or pull context from a previous one — someone has to jump in. That's the swivel chair IT problem Presidio is describing, and it's alive and well in B2B marketing stacks.
The so-what
The teams winning with AI aren't running more pilots. They're connecting the ones they already have. Before you add another AI tool to your stack, ask one question: does it share context with anything else you're running, or does a human have to carry that information between systems? If it's the latter, you don't have an AI problem — you have a workflow design problem that AI can't fix on its own. An orchestrated system that does three things reliably will outperform five disconnected agents every time.
---