Distributional

"As AI systems become more complex and autonomous, humans will have influence in how they construct reward functions, how they set the goals, and where they point this incredible technology. That means having the wisdom to know where to go, but also having the observability to know how far you are from that." —Scott Clark at @SAIRfoundation youtube.com/watch?v=7qDx8P…

98% of companies say they're "AI native" but 70% of them report having no reliable metrics or ways to measure how their AI products are performing. DBNL can help provide a peek into the black box, removing guesswork and providing actionable insights. Read the State of AI Transformation 2026 report from @OpCo_VC: dbnl.io/fV2MOo

It can be hard to understand user intent in relation to how an agent behaves in production. With DBNL, we’ll show how to use analysis of production agent traces to understand user intent, interesting usage patterns, agent performance, and cost, quality, and speed metrics in the context of downstream performance KPIs. See how it works at booth #7011 at the @nvidiagtc outdoor pavilion today.

There are plenty of opportunities to tweak agents in production to boost performance, but which changes should you prioritize? And how do you know those changes are effective? DBNL enables teams to discover, analyze, and track improvements with their production AI agents. Stop by booth #7011 at @nvidiagtc for a demo.

Automate finding issues with your AI agent in production—with greater confidence. DBNL surfaces issues from production AI logs, suggests the fix, enables A/B testing, and tracks relevant metrics over time to confirm performance gains. See how it works at booth #7011 at the @nvidiagtc outdoor pavilion this week.

Production AI is a black box, which leaves AI teams struggling to improve, fix, and scale their products. With DBNL, we’ll show you how to use open software to analyze over 10K daily production OpenTelemetry traces. We’ll also show you how to find issues automatically with daily analysis of production logs that involves topic modeling, LLM as judge scorers and classifiers, assessment of downstream KPIs with typical cost, quality and speed metrics, high dimensional clustering, and correlations across these attributes. Stop by booth #7011 at @nvidiagtc for a demo.

Once you understand your AI agent's baseline behavior, you can start to measure where and how it's changed in production. See how DBNL works with our fully open, free demo SaaS: dbnl.io/zOXlrw

Have you planned out your @NVIDIAGTC schedule yet? Make sure to add us to your calendar. Stop by booth #7011 in the outdoor pavilion for several demos showing how DBNL can help you better understand the behavior of your AI agents in production—and how to improve behavior over time: dbnl.io/WT4Qh2

Improve your AI agents in production. DBNL provides a peek behind the curtain and serves insights into how your agents are performing, so you can support, improve, fix, and scale in production. See yourself with our fully open, free demo SaaS here: dbnl.io/zOXlrw

I think where agents really start to shine are applications where they take on more automated planning and multi-step research Roami can not only do deep research on options, but the more interesting "agentic" capability is reasoning about a "sequence" or "plan" of spots to hit for an outing request I actually went on this sequence that roami suggested and it was excellent! To make a system like this work, you have to invest it understanding what is and isn't working for your users. This is where DBNL comes in: @dbnlAI is analytics for production agents 📊 Understand agent cost, quality, and usage ⚠️ Discover issues in agent logs 🔎 Investigate issues for rapid root cause 📈 Track and confirm improvements #moptheslop

The right analysis can make the next step obvious 📊😎 One of the analysis signals that DBNL computes automatically for teams is topic usage over all requests. After publishing logs for 7 days, teams can see the topic breakdown of all user requests of a production agent. In the case of our outing agent example, some clear usage patterns / topics emerged. An obvious improvement was to add some pre-baked prompts to the agents homepage. These make it more clear for users getting started and reducing keystrokes for common queries @dbnlAI is analytics for production agents Understand what to improve and fix next 🚀 deploy 📊 analyze 📈 improve 🔁 repeat Accelerate your AI product flywheel #moptheslop

Some of the most popular chatbots and AI coding tools are also the ones driving the most churn, according to the State of AI Transformation 2026—often due to low-quality output or being cost-prohibitive. Where does your AI product fit in this list? Get more insights from the @OpCo_VC report: dbnl.io/fV2MOo

"AI is going to start bridging gaps between different scientific fields in a way that was impossible before. An individual can only retain so much knowledge, and AI can help connect those dots." Scott Clark on the relationship between science and AI, in a conversation with Chuck Ng, Co-Founder of @SAIRfoundation. youtube.com/watch?v=7qDx8P…

One week until @NVIDIAGTC! Stop by booth #7011 in the outdoor pavilion to meet our team and get a live demo of DBNL. We’ll be showcasing how to: ➡️ Fix agent issues in production with tab complete analytics ➡️ A/B test AI agents in production at scale ➡️ Optimize agents in production with adaptive analytics ➡️ Understand agent intent to outcomes with analysis of traces, using the NeMo Agent Toolkit

What kinds of insights can you expect when you deploy DBNL? At a high level: behavioral patterns that emerge in production. Things like redundant tool usage within a single session, failures that appear when systems are rolled out to new regions with different terminology, non-linear failures caused by adding more tools or MCP servers, rare edge cases like intermittent 403 errors, and agents getting stuck in tool-calling loops. Each signal is coupled with concrete evidence and contextual guidance to help you address the issue. Learn more in our Q&A recap of our session with @jxnlco : dbnl.io/tB4qqF

"The problem is confidence, not performance." ICYMI, Scott Clark shared what's broken in AI—and what teams can do to overcome these challenges. Check out the highlights here: dbnl.io/iNqtYT

GLM-5 from z.ai is awesome, but I wonder if some distillation from Anthropic's Claude took place 🕵️‍♂️🤔 Check out this response from my GLM-5 powered agent ⬇️ I found this digging through my production logs with @dbnlAI #moptheslop

Find the needle you didn't even know you were looking for in your AI haystack. Distributional provides a guided or “tab-complete” analytics experience, where insights are presented without requiring weeks of manual data science work. Developers can quickly triage whether an issue matters, inspect specific evidence from traces, explore potential fixes ranging from prompt changes to architectural improvements, and track whether changes improve behavior over time. Learn more in our Q&A recap of our session with @jxnlco : dbnl.io/tB4qqF

Manually going through production logs to address LLM hallucinations can only get you so far. Automation can help you scale when it’s time to productionize your RAG app, so you can quickly understand behavior and resolve issues. Distributional AI Engineer Tobias Andreasen explains: dbnl.io/9zairI

When it comes to RAG applications, understanding behavior doesn’t need to be a black box. Distributional’s Tobias Andreasen explains how statistical distributions can help teams understand and test behavior changes continuously: dbnl.io/9zairI