NiCE released a new generation of its Cognigy customer-service software on November 12, adding tools intended to make AI agents easier to build, reuse and supervise. The company says teams can create a skill once, share it across agents and improve it over time, rather than recreating the same instructions for each automated service channel.
The release focuses on the work behind an automated conversation. A business may run agents for different brands, regions or types of customer request. Those agents can need some of the same capabilities, such as looking up a service policy or passing a question to a staff member. Reuse gives administrators a way to keep those shared capabilities in one place, though each deployment still needs testing in its own context.
Building and sharing skills
NiCE calls its redesigned builder Agentic Experience Studio. The company says it includes AI assistance, ready-made resources and default models for speech recognition, speech synthesis and language processing. Speech recognition turns audio into text, and speech synthesis produces spoken output from text. NiCE also lists premium models from Deepgram, ElevenLabs and Anthropic.
Agents are assembled from skills, tools and personas, according to NiCE. The company says they can coordinate other agents for specialized work and communicate with external agents over open standards. That describes a design capability. It does not establish that a company can connect a particular outside service without checking compatibility, permissions and what customer information may be exchanged.
Voice and cost options
The release adds support for models that process speech end to end, rather than requiring separate components to handle each stage of a spoken conversation. NiCE says the approach can make calls faster and more natural. The announcement does not quantify a typical time saving or provide an independent comparison with its prior voice tools.
NiCE says default models are included in the conversation price. Customers can also choose premium third-party models billed by usage or continue with their own model contracts. Those are different cost structures, not a promise that every configuration costs the same. A service team comparing them would need to account for call volumes and the tasks it routes to premium models.
What teams still need to verify
NiCE also emphasizes visibility into agent decisions. Decision traces can help supervisors see what the agent did and investigate where a conversation went wrong. A trace alone does not make a result correct: teams still need someone responsible for reviewing failures, fixing faulty instructions and checking that problems are not repeated when a shared skill changes.
The practical question for contact centers is whether reuse and tracing reduce maintenance without making it harder to handle exceptions. That can be tested by reviewing how an agent responds when it lacks enough information or when a customer asks for a person. NiCE’s announcement describes the tools it provides, but the results will depend on how businesses configure, test and supervise them.



