Ask Heather Richards for the plain-English version of Agent Factory and she corrects the question before answering it. The name suggests a factory – something that outputs a fixed thing. What Verint actually built, in her telling, is a workbench. Pick the best underlying model for a given bot. Orchestrate multiple agents deterministically or with agentic reasoning depending on where a conversation goes. Route through an MCP layer to pull whatever data the bot needs. “You can both augment and create new AI agents.”
That correction is small. It’s also the clearest evidence of how Richards’ whole portfolio is built.
Agent Factory: workbench, not factory
Agent Factory is Verint’s orchestration layer for hybrid workforces – AI agents and human agents managed under the same logic. The detail that matters most is the model-agnostic design: nothing locks a customer into one LLM. Pick the best model for the bot, not the model the vendor happens to be partnered with this quarter.
Verint’s own engineering teams are using Agent Factory internally to prototype bots before productizing them. A build tool a vendor’s engineers actually choose to use is different evidence than a roadmap slide.
The model-agnostic choice isn’t incidental. It’s the same instinct running underneath everything else in Richards’ portfolio, stated more plainly: nothing about the AI layer should require the customer to commit to a single vendor’s stack. Which is the argument Richards makes explicitly, and more aggressively, one level down.
“Infrastructure is just plumbing”
Richards’ portfolio – channels, desktop, knowledge management, IVA and copilots – isn’t new to Verint. Some of it predates the company’s public AI narrative by close to a decade. What’s changed is the framing: Verint no longer requires customers to consume the full stack to get value from any one piece of it. “Infrastructure is just plumbing,” she said. “You keep whatever you have.”
Richards contrasts this with CCaaS vendors, whose sales motion centers on migrating a customer’s full infrastructure to the cloud before layering automation on top. “We’re very happy to continue augmenting somebody’s existing CCaaS with our fantastic WEM portfolio… we’re absolutely happy to put our IVA on anybody’s voice channels.” Verint skips the infrastructure argument and sells automation of a single interaction point instead. One bot. One use case. Then expand.
That’s a mechanical difference, not just messaging. If the unit of sale is a cloud migration, the success metric is the migration – did the telephony move. If the unit of sale is a single automated workflow, the success metric is dollars saved on that workflow, which is a far easier number for a buyer to defend upward. Richards was direct about the intent: “If what we’re selling is CX automation and outcomes, as opposed to infrastructure with a bunch of applications tacked on, all of a sudden the goal of the project is very, very different.”
It’s also a philosophy that fits neatly with where Verint’s actual strength sits. Two decades of expertise in workforce engagement and knowledge management, not in owning anyone’s core cloud contact center platform – the Calabrio integration Richards mentioned in passing is that pattern in practice. The conviction and the business’s actual position point the same direction.
What’s live, and what’s still early
Asked directly where agentic AI for self-service is production-ready versus still pilot theater, Richards didn’t reach for the industry’s default confident answer. Customer-facing production use cases skew low-risk – a hotel chain automating an invoice request by dipping into CRM and booking data, real and operational, but not exactly a high-wire act.
Voice is where the story changes. The industry-wide push toward speech-to-speech IVAs – collapsing the old speech-to-text-to-LLM-to-text-to-speech pipeline into two models talking directly to each other – impresses as a demo, but Richards was candid that she’s seen very few actual production deployments, and only in small, low-stakes settings. The reason is structural: collapsing the pipeline also collapses the ability to control what the model decides mid-conversation, and Verint isn’t yet confident enough in that control to call speech-to-speech production-ready more broadly.
Co-opetition is a strategy, not just a description
On competitive positioning: CCaaS vendors, niche point solutions, hyperscalers, and increasingly the CRM vendors as Salesforce pushes into contact center territory. Several of those categories are also Verint partners, which makes for a landscape she describes as “very much co-opetition.” Because Verint has always competed this broadly, Richards argues, AI hasn’t changed the shape of the landscape – just added noise on top of it.
That’s one way to read it. The other read: a vendor that competes with, partners with, and gets built on top of by nearly every category of player in its market doesn’t have a clean differentiated lane. It has a durable position inside everyone else’s stack. Differentiation and durability aren’t the same claim – which one matters more depends on what you’re trying to predict.
The through-line is knowledge management
Richards came into Verint via acquisition – her prior company was around 75 people, Verint is roughly 4,000 globally. Her focus before the acquisition was narrowly AI-powered knowledge management, a discipline she says went through a lull. But generative AI made it foundational again. Grounding a model’s answers in a company’s actual knowledge – rather than letting it guess – depends entirely on a usable knowledge layer underneath it, and that infrastructure now sits under nearly everything in her portfolio – analytics, IVA, quality. “That’s been really fun to be able to take all the stuff that you know from past lives and all of a sudden it’s very core and foundational to all the new technologies that we’re delivering.”
The knowledge management story is about the structured, retrievable institutional knowledge every one of these AI applications actually depends on to be worth anything at all.
Photo by Samuel Sianipar on Unsplash
