Cognigy's new release sells control over AI agents, and the speed claims come with no measurements
NiCE launched the next generation of its Cognigy enterprise AI agent platform on Nov. 12, 2026, according to CMSWire. The release adds an AI-assisted workspace called Agentic Experience Studio, agents built from reusable skills, tools and personas, and tracing that records each step an agent takes. It also adds native speech-to-speech models and bundled model pricing. Native model integrations cover Deepgram, ElevenLabs, OpenAI and Anthropic. Chief AI Officer Philipp Heltewig said enterprises can scale agents "with the visibility and control production demands." NiCE gave no figures for any speed or cost improvement.
What does decision tracing actually give a support team?
Tracing logs each step an agent takes, the data and tools it uses, and the decisions it makes. For a support team, that means a record of why an agent gave a particular answer.
This is the most useful item in the release. When a bot gives a customer a wrong refund answer, the usual question is whether the fault lies in the prompt, the knowledge source, the tool call or the model. A step-by-step record lets you answer that without guessing.
The coverage doesn't say how long traces are kept, whether they can be searched or exported, or how customer data inside them is handled. "Every decision" also describes what the platform can see. If an external model makes a choice inside its own black box, the trace may show the call and the result but not the reasoning.
Does bundled model pricing make agents cheaper?
Not necessarily. Default speech-to-text, text-to-speech and language models are included in the conversation price. Premium models from the named providers are optional and billed by usage.
Bundling makes the base bill easier to predict and removes separate vendor contracts for the default stack. The risk is that the defaults get you to a demo and the premium voices or models, which are metered, end up in production. No rates were given in the text we had. It breaks off partway through the pricing section, so check NiCE's own documentation for the full terms.
Will reusable skills shorten build times?
Probably for teams running several agents, less so for a single bot. Skills, tools and personas are built once, reused across agents and improved over time, so the saving comes from not rebuilding the same pieces.
NiCE says the new Studio gets teams "from idea to a working agent faster", which is the company's own claim. The coverage gives no baseline, no customer result and no definition of "working". The AI Assist features, covering guided setup, testing and improvement, will need to be tried on your own use case. Reuse also has a cost: a change to a shared skill reaches every agent that uses it. Ask how versioning and rollback work.
What should I ask NiCE before upgrading?
Ask what the trace retains and for how long, which premium models trigger usage billing, and how agent behaviour is tested after you edit a shared skill.
Existing customers get the new capabilities at no charge, and production deployments are said to stay unchanged. Ask NiCE to confirm that in writing for your contract. Also ask whether native speech-to-speech models, which work on audio directly, can be switched off per flow. NiCE says they give faster, more lifelike conversations, but a voice agent you can't read as text may be harder to audit.
Frequently asked questions
Is this available now?
NiCE launched it on Nov. 12, 2026. CMSWire reports that existing customers get the new capabilities at no cost.
Does it connect to other vendors' agents?
Yes, in principle. Agents can connect with external agents through open standards, including customers' personal AI agents. The coverage doesn't name the standards.
Which model providers are built in?
Deepgram, ElevenLabs, OpenAI and Anthropic are native integrations. Default models are included, and premium ones are billed by usage.