Front page
InterviewCX AI2h · 15:01 BST · 3 min read
How To Manage an Ecosystem of AI Agents in Your CX Stack
Brian Donohue, VP Product at Fin, tells CX Today how businesses should structure their growing ecosystem of AI agents, from customer-facing bots to internal copilots, without creating new data silos. He also addresses where trust in AI agents is strongest and where it still falls short.
Most businesses aren’t deploying one AI agent anymore. They’re deploying several, a customer-facing bot, an internal copilot, maybe something handling operations behind the scenes, often without ever stepping back to ask how those pieces are meant to work together.
Brian Donohue, VP Product at Fin, spoke to CX Today about what it actually takes to manage that complexity without recreating the fragmented, siloed experiences AI was supposed to fix.
Where Should Businesses Take Risks, and Where Should They Play It Safe?
Donohue’s starting point is a distinction most companies haven’t drawn clearly enough: internal experimentation is fine, but customer-facing agents need a different standard entirely.
“You’re on the hook for the quality of that experience… I think it’s critical that companies are like, hey, we need rigor here. We need to treat this like a product, not like an experiment”
That might sound obvious, but Donohue says plenty of businesses have been more experimental with customer-facing agents than anyone would expect, treating live deployments with the same looseness they’d apply to an internal test.
How Do You Stop AI Agents From Creating New Silos?
Fin currently runs three distinct agents, one for customer conversations, one supporting human teammates in the inbox, and one managing the support operation itself. Donohue is candid that the long-term goal isn’t three separate tools, it’s one.
“In a future world… these will ultimately converge. And there will kind of be one agent doing that”
Until that happens, the priority is making sure all three draw from the same context, knowledge, and guidance, so a customer’s history doesn’t vanish the moment they move between a bot and a human.
It’s the same failure mode that plagued pre-AI customer service, just with new technology sitting on top of the same old fragmentation if businesses aren’t careful.
What Does the AI Colleague Mean for Human Training?
As AI agents absorb more routine volume, often 70-80% of inbound queries at the companies furthest along, the work left for humans skews toward the hardest, most frustrated cases. That creates an unusual problem: how do you train someone up on easy tickets they’ll rarely see anymore?
Donohue frames the fix less as a one-way handover and more as an ongoing back-and-forth between agent and human, closer to how a newer colleague might ask for help.
“…it’s like Fin acting exactly as a human would if they’re say six months into the job… Ping their tap their manager on the shoulder and say, hey, I actually need some help”
What’s His Advice for CX Leaders Deploying Agents Now?
Donohue’s advice splits into two parts. First, most companies are sitting on a “massive technology overhang,” using only a fraction of what today’s AI can already do, often avoiding entire channels like voice for no genuine capability reason.
Second, once a business commits to pushing further, it needs to start operating like a software team rather than a CX team running a pilot. That means regression testing, live monitoring, and treating every change as something that could just as easily degrade the experience as improve it.
Will Customers Ever Fully Trust AI Agents?
Fin’s own research shows trust isn’t uniform. Customers are notably more comfortable with AI on chat than on voice, and some still refuse to believe an agent even when it’s right.
“A big challenge they have is Fin gave the correct answer, but the customer didn’t trust it”
Looking ahead, Donohue sees two possible paths: trust climbs gradually as good AI experiences accumulate, or the entire question becomes irrelevant once agent-to-agent interactions, AI talking to AI on the customer’s behalf, become the norm.
Dive into expert insight on Salesforce’s recent acquisition of Fin here.
The Bigger Picture
Agent sprawl isn’t going away, and businesses can’t wait for it to settle into something tidy on its own.
The enterprises who manage that complexity deliberately, rather than letting it accumulate the way disconnected systems always have, are the ones likely to come out ahead.
rate this storyhelps rank stories across CX Today
Read nextordered by techtelligence · every pick explained
same beat · CX AI
Why AI Can Be Right and Still Get CX Wrong
23 Sept 2026matches this topicWhen AI Agents Ignore “No,” Security Gets Real24 Sept 2026matches this topicCRM’s Great Escape: Salesforce, HubSpot, Zeotap, and Zendesk Redraw the Data Battle21 Sept 2026
More from the CX AI desk
Discover how human oversight can strengthen AI service performance beyond basic resolution-rate measurement.
Francesca Roche · 23 Sept 2026Interview
Dreamforce 2026 Roundup: Salesforce Has an ‘Easy Button’ for AI. But CX Leaders Still Face the Hard Decisions.
Constellation Research’s Liz Miller explains Salesforce AIforce, Koa, headless CRM, and why CX leaders must prioritize customer outcomes over AI hype.
Rob Scott · 22 Sept 2026Interview
From Prebuilt Agents to Vibe Coding: Salesforce’s New CX Playbook
Salesforce Agentforce CPO John Kucera explains how AIforce, Koa, prebuilt agents, and custom interfaces could reshape CX—and what leaders must govern first.
