Distyl

There are two ways to use AI in healthcare: optimize the workflows you have, or redesign them entirely. The first gets you some efficiency. The second gets you a different company. Our team is heading to Chicago for Reuters Digital Health 2026. Arjun Prakash, our CEO and Co-Founder, will be on the "Breaking the System to Fix It: Infrastructure for a New Era of Care" panel with Ashok Chennuru (Elevance Health), David Oakley (AdventHealth), and Dr. David Jaffray (MD Anderson). The infrastructure question is the right one. Today’s care delivery runs on systems that were not designed for what’s being asked of them now. That’s the work we do at Distyl, rebuilding them. If you're going to be in Chicago, reach out —  Arjun Prakash would love to connect!

Proud to share that T-Mobile's T-Life AI Assistant won the People's Voice Webby award for Best Use of AI Voice & Conversational Interface! In collaborating with T-Mobile, we’ve helped them innovate something entirely new: a real-time, creative, multimodal AI experience that changes how millions of users interact with the T-Life app. T-Mobile’s goal is to bring their industry leading in-store experience to customers at their fingertips, with visual, audio, and interactive elements to create an immersive experience. We are incredibly proud of our contribution, and the recognition T-Life AI Assistant has received at The Webby Awards.

What does it take to transform customer experience at Fortune 50 scale? This week, Julianne Roberson, Director of AI Engineering at T-Mobile, and Derek Ho, our COO and Co-Founder, were on stage at the AI Agent Conference in New York to answer exactly that. What did it take to build the world’s leading Agentic AI CX experience? 1. T-Mobile's obsession with delighting customers 2. Distyl's obsession with owning outcomes and building technologies that meet production demands This is what the most impactful AI CX offering in the market looks like, and we're bringing it to the world's most demanding enterprises.

One of the most difficult problems in enterprise AI isn’t the models – it’s the systems. After three years of building production-grade AI inside Fortune 500s, @Distyl developed an adaptive harness that autonomously designs, builds, and optimizes AI systems. Across 75 benchmarks, it outperformed the published state-of-the-art, including 15 system-first benchmarks previously dominated by expert-built solutions. No fine-tuning. No manual prompting. No expert bottleneck. Systems don’t just outperform models anymore. Now, systems are building systems.

The gap between "AI strategy" and "AI in production" is still the defining bottleneck for most enterprises. We're partnering with Google Cloud to close that gap — pairing our enterprise AI delivery platform and Forward Deployed Engineer model with Google Cloud infrastructure and Gemini Enterprise for CX to move Fortune 500 companies from scoping to production in weeks, not quarters. Not proofs of concept. Production-grade systems, deeply integrated, auditable, and built around the people who know the business best. Read More: https://lnkd.in/gg4cUYF8

Our team built PrefPO, a lightweight, pairwise preference prompt optimization framework, and today we're open-sourcing it! Below is the same task optimized by two methods. Both achieve 100% pass rate. One produced 4,000 characters of bloated instructions. PrefPO (ours) produced a prompt over 10x shorter, with significantly less repetition, and preferred by human evaluators. PrefPO matches or outperforms SOTA frameworks on the majority of tasks we evaluate on, while dramatically improving prompt quality. Getting started is simple: all you need to start optimizing is an initial prompt and a rubric for your task (no datasets required!). We're constantly pushing the frontier of what's possible with LLMs, which demands fast iteration. PrefPO is one of the tools that lets us and our partners move quickly. If you're interested in building the tools that define how some of the largest companies work with AI, we're hiring. We will be presenting this work at the Agents in the Wild workshop at ICLR 2026! Open Roles: https://lnkd.in/g_dvwzne Blog: https://lnkd.in/gk2ADv89 Paper: https://lnkd.in/gexsw9m5 Code: https://lnkd.in/gURgfggp

Distyl is excited to work with NVIDIA AI on deploying NemoClaw, an open-source stack for running always-on AI assistants safely in production. It installs the NVIDIA OpenShell runtime, part of the NVIDIA Agent Toolkit, to provide a secure environment for running autonomous agents, and open source models like NVIDIA Nemotron. Now integrated into Distillery, our enterprise AI platform. The next generation of enterprise AI isn't smarter chatbots. It's always-on assistants that persist across interactions, accumulate context, and operate continuously inside complex organizations. Most enterprise AI is missing two things: 1️⃣ Context. Without it, even the best models hallucinate — knowledge is siloed, locked, and inaccessible. 2️⃣ Enterprise-grade controls. Monitoring, governance, multi-agent coordination. Intelligence without the right harness never reaches production. Distillery's Context Mesh and NVIDIA NeMo Agent Toolkit, NIM microservices, and Nemotron 3 Super are the infrastructure stack always-on enterprise agents actually require. Read more ↓ https://lnkd.in/g8YbifG9

Tomorrow at #GTC26 ⚡ Catch more on Distyl's personalized assistant at session [S81570]: From Data to Decisions: Enabling AI Agents With Business Knowledge 📅 Wednesday, March 18 @ 10:00am PST Learn how Distillery gives AI agents the context they need to operate inside real enterprise systems and keeps them running safely in production. Powered by NVIDIA AI-Q with Nemotron 3 Super–the stack behind #1 Deep Research Bench I & II for report quality and factual recall. 📍 Session: https://lnkd.in/ggH8bWU7

Tomorrow at #GTC26 ⚡ Catch more on Distyl's personalized assistant at session [S81570]: From Data to Decisions: Enabling AI Agents With Business Knowledge 📅 Wednesday, March 18 @ 10:00am PST Learn how Distillery gives AI agents the context they need to operate inside real enterprise systems and keeps them running safely in production. Powered by NVIDIA AI-Q with Nemotron 3 Super–the stack behind #1 Deep Research Bench I & II for report quality and factual recall. 📍 Session: https://lnkd.in/ggH8bWU7

At Socium Ventures Inflection 2026, in partnership with Cox Enterprises in Atlanta, leaders from across the AI ecosystem gathered to examine what it truly takes to bring AI into the core of enterprise operations. Distyl co-founder Arjun Prakash joined Richard Socher (You.com) and Harish Abbott (Augment) for a panel moderated by Mitchell Gersten , CIO of Cox Automotive Inc., exploring a question many large organizations are now confronting: what does it take to move from AI experiments to systems that run the business? The next phase of enterprise AI won’t be defined by better models alone. It will be defined by how organizations redesign workflows, governance, and expertise around intelligence.