At a million conversations a month, a 99% eval score still leaves 10,000 customers with the wrong answer. Do you know which ones? Join Cyara CEO Sushil Kumar and VUX AI's Kane Simms to learn what proves an AI agent is doing its job and how to keep proving it after go-live.
Live conversation, 45 minutes, with audience Q&A. All registrants receive the recording.


Thursday 10 December 2026, 17:00 GMT. Live conversation, 45 minutes, with audience Q&A. All registrants receive the recording.
Thank you. A confirmation email will follow with the Zoom joining link. Add the session to your calendar now.
You'll get a confirmation email with the joining link, and the recording after the session.
Five yes or no questions, drawn from what this session covers. Your score stays on this page.
Why only 9% of teams trust automated evaluation, and what to show your board, your regulator and your customers instead.4
Measures that show whether the customer's problem was resolved and why containment misleads, especially for AI agents.
Sign-off at launch expires when models, prompts, knowledge and integrations change. How to keep validating after go-live.
What adversarial testing looks like when agents can issue refunds and change accounts.
How CX, IT and risk split responsibility, where human review protects customers and where it becomes box-ticking.
What the FCA, the EU AI Act and other frameworks ask for before an agent goes in front of customers.
AI agents now take action: changing account information, issuing refunds and even selling. Yet they're still probabilistic, which means the same request can get a different response each time.
Sure, you can test them. But a pass at launch only covers the agent you tested. How do you test something that changes whenever there's an update to a model, prompt, the knowledge base or an integration?
Then you have the unpredictability of customer behaviour. How well does your AI agent stick to its scope? In one study, 50 of 55 public customer service agents offered customers benefits they had no authority to give.2
And when it makes a mistake, who's responsible? The team that built it? Your vendor? Half of UK tech leaders say the CIO is held personally responsible.3
That means you've got to have proof that your AI agents are doing what they should, every day they're live.
Sushil and Kane look at what counts as proof for an AI agent that takes actions, and how CX, IT and risk teams build that proof together.
CIOs and IT leaders, contact centre technology leaders and AI leaders who are accountable for AI agents they don't fully control. Especially relevant in financial services, healthcare, telecoms and other regulated industries.

Sushil Kumar is Chief Executive Officer of Cyara, appointed in December 2025. Before joining Cyara he was co-founder and CEO of RelicX.ai, a generative AI test automation company acquired by Harness, and earlier led product and business organisations at Oracle, CA Technologies and Broadcom.

Kane has advised enterprises on conversational and agentic AI for close to a decade, with a focus on experience quality and testing.
If you're accountable for an AI agent in front of customers, this session will show you what to measure and how to prove it.
Register nowAI Breakfast: Keeping Control of Enterprise AI Agents. London. 26th November.
Meet VUX at CCW Europe, Amsterdam, October 5-7
Register now: Why your contact centre playbook is obsolete