VUX World ads

Manual testing can miss up to 17 out of 18 real-world failures in enterprise conversational AI. Most teams don’t find out until it’s too late. On the 4th of June, we're running a live session on what structured testing actually catches, and how to find failures before your customers do. Register now to secure your spot.

Most enterprise AI assistants go live without teams being confident their testing covered what actually matters. Traditional QA was built for deterministic systems. GenAI testing isn’t. Internal test scripts are written around what you expect users to say. Real users don't behave like test scripts. That gap is where most failures live, and where your customers find them first. Join us on the 4th of June to see what GenAI testing actually looks like.

Manual testing captures a fraction of real-world conversational behaviour. In enterprise deployments, we’ve found it misses 17 out of 18 failures before go-live. Not because teams are not trying. Because traditional QA was designed for deterministic systems. Gen AI agents are not deterministic. See what your team is missing in our webinar.

Most enterprises AI development skip one critical step. Your assistant went live. But traditional QA wasn’t built for how conversational AI actually behaves in production. Multi-turn failures, edge cases, and escalations breakdowns don’t show up in internal test scripts. They show up with customers. Join our webinar where we uncover the challenges and solutions.

We’re running a live session on enterprise conversational AI on the 4th of June. In 30 minutes we’ll cover what testing built for GenAI actually looks like, walk through catching real failures before they reach production, and take questions. Built for AI Product Owners, GenAI Engineers, and Heads of Conversational AI in financial services, telecoms, and energy industries. Register free. Recording available to all who sign up.

🤯 AI moves fast! A week after I got heat for saying AI agents will build and optimise themselves, Kore.ai releases Artemis to do that! Arch is an AI agent that builds AI agents within Artemis. Some things you prompt it to do (like design your initial agent). Other things, it does on its own (like recommending improvements - more or less one click later and they’re built, tested and implemented). 𝗔𝗜 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗮𝗻𝗱 𝗺𝗮𝗻𝗮𝗴𝗶𝗻𝗴 𝗔𝗜. If this is the direction, it won’t be long before this process is autonomous. This is interesting because the two biggest gaps in AI deployments today are: 1️⃣ Deploying production systems. 2️⃣ Governing AI agents at scale. If you look at the modern AI platforms (LLM wrappers), what do they miss? 𝗧𝗵𝗲𝘆’𝗿𝗲 𝗯𝗮𝘀𝗶𝗰𝗮𝗹𝗹𝘆 𝗽𝗿𝗼𝗺𝗽𝘁 𝗮𝗻𝗱 𝗽𝗿𝗲𝘆 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺𝘀. 👀 How do you govern the experience? 𝗣𝗿𝗼𝗺𝗽𝘁. 🎛️ How do you control what’s happening? 𝗣𝗿𝗼𝗺𝗽𝘁. 🧠 How do you manage context and memory? 𝗣𝗿𝗼𝗺𝗽𝘁. ⚙️ How do you manage workflows? 𝗣𝗿𝗼𝗺𝗽𝘁. This creates the 𝘪𝘭𝘭𝘶𝘴𝘪𝘰𝘯 that 𝘢𝘯𝘺𝘰𝘯𝘦 can design and build these things and a prompt is a good enough way to govern it! 𝘛𝘩𝘪𝘴 𝘸𝘩𝘰𝘭𝘦 𝘱𝘳𝘦𝘮𝘪𝘴𝘦 𝘳𝘦𝘭𝘪𝘦𝘴 𝘰𝘯 𝘵𝘩𝘦 𝘢𝘴𝘴𝘶𝘮𝘱𝘵𝘪𝘰𝘯 𝘵𝘩𝘢𝘵 𝘓𝘓𝘔𝘴 𝘢𝘳𝘦 𝘳𝘦𝘮𝘰𝘵𝘦𝘭𝘺 𝘳𝘦𝘭𝘪𝘢𝘣𝘭𝘦. 𝗧𝗵𝗮𝘁'𝘀 𝘄𝗵𝘆 𝘁𝗵𝗲𝗿𝗲 𝗮𝗿𝗲 𝘀𝗼𝗺𝗲 𝗳𝘂𝗻𝗱𝗮𝗺𝗲𝗻𝘁𝗮𝗹 𝗿𝗲𝗾𝘂𝗶𝗿𝗲𝗺𝗲𝗻𝘁𝘀 𝘆𝗼𝘂 𝙣𝙚𝙚𝙙 𝗶𝗻 𝗮𝗻 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗔𝗜 𝘀𝘁𝗮𝗰𝗸. Requirements that these ‘AI-native’ companies are only just beginning to learn as they scale. Requirements that Kore.ai has preserved from its previous platform because you can’t scale AI reliably without them! Things like: 1. The option to have rule-based determinism. 2. Memory (state) management sitting outside the LLM layer. 3. Deep orchestration between AI, processes, data and integrations. 4. Monitoring, analysis and observability being a core product (not a bolt-on or an external tool). When you make things like this part of an AI-native platform, you’re in a space where you’re leveraging AI, in all its forms, for the task that each is best suited for. This is what kore describes as the Dual-Brain Architecture. 𝗢𝗻 𝘁𝗵𝗲 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁 𝘀𝗶𝗱𝗲, few have the skills to go beyond a prototype and into full scale production. This is why OpenAI bought Tomoro (a services company). The tech only solves half of the problem. Artemis looks like the closest I’ve seen to the potential of actually democratising AI agent builds. I admittedly do have a soft spot for kore ai - Raj Koneru was an early supporter of VUX World. But I'm being sincere when I say that this looks to me to be the culmination of a decade’s work. It could, if my initial outside assessment is along the right lines, go some way towards solving those two 𝗕𝗜𝗚 challenges. Congratulations Raj, Prasanna Arikala, Cobus Greyling and team on what looks like a huge effort!