Level AI

How is the Cerebras IPO, voice agent quality and future of AI scaling linked? CX Voice Agents are the largest product category which depends on shaving every 10 milliseconds of latency. If a coding agent takes a few more seconds it does not really matter but the extra 2 seconds of latency on a phone call is difference between terrible and great. What is one key thing this turn by turn latency depend on? Inference speed of LLM, which in turn depends on the chip LLM is running on. At Level AI we tried to benchmark model latency on Cerebras systems vs. GPUs and we saw up to at-least a 50% (!) latency reduction on a single call. That is transformational impact when shaving every 10ms is at a premium. This is big for the space as work loads move to production and agent quality at runtime becomes key. Cerebras chip beats GPUs fundamentally on inference due to single chip memory and inference architecture. The world will see an explosion of inference time agents and thus this moment is critical to the future of AI Agents. Congrats to Steve Vassallo for getting to this breakthrough early. I remember first meeting him in a GSB class. His engineers brain and beginners, design oriented mindset has always kept him apart.

Our #customerdelivery team at Level AI exists to make our users' lives easier. Every day, we are enabling them with Workers to ensure: ✓ Coaches get development plans built from 100% of agent interactions. ✓ Analysts get answers to strategic questions in seconds. ✓ CX leaders understand the why behind the numbers, not just the numbers. It’s so fulfilling to hear the relief and confidence in our champions’ voices as they realize the opportunities for improvement that our Workers create.

After ~12 months of building in silence, Level AI is launching AI Workers today. The hardest product decision we made on AI Workers was the one we made first: scope each worker to a single CX function, and don't compromise on that. Every pressure in product development pushes toward generality. A broader scope looks like a bigger market, a flexible interface looks like easier adoption and a general-purpose worker looks like it serves more personas. It also produces shallow outputs. A QA leader's job looks nothing like an analyst's. A coaching workflow requires an agent's full interaction history, their current development plan, and your company's rubric. A root cause analysis requires transcript data joined to CRM records, VoC signals, and QA scores before a single conclusion is drawn. Scoping each worker tightly meant building the context layer that makes outputs operationally useful. That's the work the past 12 months produced, testing against real rubrics at Smartsheet, real QA workflows at Empyrean, real coaching processes at Ollie Pets. Every time an output missed the operational standard those teams held, we rebuilt the context layer, not the interface. The workers launching today are the product of that calibration. Nearly 100 enterprise customers. 38,000 tasks completed. 88% running across three or more CX functions. Explore the full suite of purpose built workers: https://lnkd.in/gMrsqrkd

Today, Level AI is launching AI Workers. Most AI in CX has focused on the front office: virtual agents, chat deflection, IVR routing, and self-service automation. But a large amount of operational work still happens behind the scenes: QA reviews, coaching prep, transcript research, trend analysis, product feedback synthesis, executive reporting, and compliance monitoring. This is where AI Workers come in. AI Workers are role-scoped agents for CX operations. Each worker owns a specific workflow and produces a concrete deliverable: a coaching plan, a conversation research report, a team performance analysis, product feedback themes, or executive insights. What makes this different from a generic copilot is the underlying context. These workers operate on Level AI’s customer intelligence layer: transcripts, QA rubrics, scorecards, coaching signals, sentiment, resolution outcomes, VoC themes, CRM context, and team hierarchy. They can combine semantic retrieval, structured analytics, multi-step investigation, and evidence attribution to produce outputs that are grounded in the same data layer CX teams already trust. The important shift is that AI is no longer just summarizing conversations or extracting labels. It is moving closer to workflow execution: from raw interactions → structured intelligence → role-specific outputs → operational action. That is the part I’m most excited about. AI Workers are a step toward making CX operations more scalable, measurable, and data-driven. https://lnkd.in/gRc4Z5pA

Today Level AI is launching AI Workers. AI Workers know your agents, your customers, and your policies from day one. They automate the routine work that drains contact center teams. Teams in beta keep telling us the same thing in different words: they're quantifying things that used to live in gut feel and bringing data-backed trends to client conversations. AI can save you time, but it will also let you do things you could not do before. AI Workers are the latter (and they also save you time and effort). https://lnkd.in/eYSUfBct