NiCE upgrades Cognigy with AI agent testing, tracing and voice support

NiCE’s November 12 launch of its next-generation Cognigy platform promises faster ways to build and deploy customer-service AI agents. But the upgrade is not a push to let AI handle every decision: it preserves fixed workflows alongside autonomous agents. NiCE’s product chief says most customers still use that hybrid approach, rather than fully autonomous deployments.

From building an agent to checking its work

The redesigned Agentic Experience Studio combines an AI-assisted workspace with ready-made resources. AI Assist guides setup, building, testing and later improvements. Reusable skills let teams package instructions and tools once, then apply them across agents instead of rebuilding the same capability for each one. The release also lets agents coordinate specialized sub-agents for more complex work.

There is still room for expert configuration. In a pre-launch briefing described by analyst Zeus Kerravala in SiliconANGLE, NiCE product management director Clement Boutier said advanced sections retain full access to an agent’s system prompt—the instructions that guide its behavior. The simpler interface targets smaller businesses, contact-center administrators and team leads, as well as engineers.

The testing tools go beyond trying a single conversation. Cognigy’s Simulator supports comparisons of up to 10 variants, using simulated customers and a language model to judge the results. During the briefing, Boutier compared the same agent running on two models; one ran faster and used fewer tokens, the units models process and bill for.

Tracing addresses a different job: reconstructing an agent’s actions. The turn-by-turn record shows which skill loaded, which tool the agent called, the arguments passed to that tool and when control moved to a predefined flow. Those records expose the steps taken during an interaction, rather than leaving operators with only the final response to inspect.

Capable and dependable do not always mean the same thing,

Nhu Ho, senior product marketing manager at NiCE Cognigy, in the briefing described by SiliconANGLE

Where fixed processes stay in charge

Boutier described fully autonomous deployments as being at the beginning of the adoption curve. The platform therefore supports movement in both directions between AI agents and structured processes. An agent can call an existing flow as a tool, while a flow placed before an agent can remove credit-card numbers or other personal information before it reaches a model.

That flexibility does not transfer all responsibility to NiCE. Boutier said customers must build in requirements such as regional AI disclosure rules and escalation to a human. He also described a practical cost to adding safeguards: each guardrail consumes tokens and adds delay. More checks are therefore not free, either financially or in the time a customer waits.

Voice options and the model bill

Native voice-to-voice support adds models that process audio end to end, rather than relying on separate speech transcription and speech generation stages. NiCE says this enables faster, more lifelike conversations. In the briefing demonstration, a voice agent kept talking while delegating tool calls to a separate model; that was a demonstration, not a measured production result.

The release also changes how teams access and pay for models. NiCE’s model-billing options separate the bundled defaults from premium choices and existing supplier arrangements:

• Included models: Default speech-to-text, text-to-speech and language models come within the conversation price, without a separate model contract or additional configuration.
• Premium models: Teams can select third-party models from providers including Deepgram, ElevenLabs and Anthropic, with usage-based billing.
• Existing contracts: Customers can keep using their own model contracts instead of moving every model relationship into Cognigy’s billing.

For existing Cognigy customers, NiCE says the next generation is included as part of the platform’s continued development. Production deployments keep running unchanged, and teams can adopt the new capabilities when ready. The rollout does not require them to replace working customer-service flows before trying the new tools.