Recap of the day
Chapters 6 chapters
About this session
Join Austin Kronz to close Context Conference with the day in one sitting: the shared company brain, how Verizon runs one context platform across its agent studios, what analytics agents need to get the numbers right, and the open ecosystem forming around context.
Transcript Click a time to jump there
Austin KronzFolks, we've done it, and we're only eleven minutes over, which I think is pretty great for a three-hour back-to-back conference. We're finally at the closing, and, and what better way to wrap this up than to just go over everything we talked about today. We started the day with Prukalpa telling us everything about why context is important. She introduced this concept of the enterprise brain and just a central nervous system across actors, artifacts, and activity. How do you create a closed loop of context that actually feeds back through traces, that updates the agent context, and starts to increase the value and accuracy over time? We then went right into an executive panel. Cindy moderated an amazing session with Leigh-Ann and Prakash.
Uh, some of the key takeaways that I've had from that were, first of all, everyone needs to be a builder. Uh, Leigh-Ann gave the example of she built an entire company, including a website, purchasing a domain, setting up areas on her commute to work. If that doesn't tell you you can be a builder, I don't know what does. She also introduced the concept of the four I's. She was talking about interlock, and you need to have intent, integrity, infrastructure, and ultimately business impact. Uh, Prakash was talking all about interoperability as well, and just the idea that lock-in and being able to pivot is so crucial when things are changing really every forty-eight hours, twenty-four hours at this point.
And then we went right into a practical session of what does an architecture for an organization look like? So Steven from Verizon shared everything about how they're already dealing with multiple what they call agent studios, so multiple platforms that are building agents. And having one enterprise context platform is crucial for them because if they ever decide to switch or move or they want to reuse context across any of these agent studios, they need to have that flexibility. We ultimately unveiled, uh, for the first time Atlan Trident. This is our answer to building that shared company brain. How can we bring our own concept of registry and observability and governance directly to the interfaces that your builders are already in?
So we started with a desktop app, desktop app. We bring it to all of the harnesses that you're already engaged with and allow you to share context artifacts like skills, ontologies, uh, and metrics across your entire organization, going from single-player to multiplayer. We followed it up with the analytics workload. You know, Ashwin and Rohan focused very much on here's how you can solve productivity workloads and, um, operational workloads, and there's different context artifacts that are needed to solve these different challenges. Uh, and ultimately, your use case will depend on which types of artifacts need to be assembled for your agent to be accurate and trustworthy. Joe led us in a panel, uh, with Piyush and Sowmya from Upwork and DocuSign, and we really just talked about what does it practically mean to be a data and AI leader, and how should you be, uh, kind of rolling out these talk to data agents in a, you know, meaningful way.
Um, we talked about kind of that fallback strategy of yes, certified metrics. Uh, if you don't have certified metrics, the agent should at least have AI-ready data to guess or browse through. But ultimately, you need to have that third level of control, which is just tell the user which mechanism you used. Is it we pursued a certified metric, uh, or did we, you know, use AI-ready data and here's our best guess of what the answer can be. Ultimately, we've been working with customers on these talk to data agents, and we shared a little bit about what the AI labs here at Atlan have learned about what context talk to data actually needs. And this idea of a metric is clearly the driving force behind accuracy.
Pick a capable model, and generally speaking, context is what will drive the differentiation and accuracy against evals. I should say performance against evals, not necessarily accuracy. We talked about in the panel having, you know, different levels of, uh, expectations for different use cases. Finance metrics might need to be a hundred percent accurate all the time, but Joe talked about the ish. You know, we need to be ninety-ish percent right for some of our other use cases. We also launched the forward deployed playbook. Uh, we talked about the open source, kind of what are our forward deployed engineers actually using to help customers get up to speed and get up and running with talk to data agents, and we open source that for you to leverage as well.
And we also introduced the accelerator. If you need some of that hands-on experience and you wanna, you know, uh, accelerate your organization's path to talk to data, feel free to sign up and, you know, reach out to us via that link. All this again will be in the chat and in the follow-ups. The announcements continued. Ben just talked about open source spec for AI registry. We wanted to meet the building community where they're at and kind of get influence, uh, and get, um, some input on how we can evolve registry together. Uh, and Ben talked about kind of our learnings from things like NPM from a registry perspective and what Iceberg did for separating storage and compute.
The panel, uh, the ecosystem again, I think it was such a cool group. You know, they mentioned that, that it probably wasn't as much spiciness or disagreement, which I think is good for everyone attending this session because the last thing we want is our ecosystems that everyone is leveraging to kind of disagree on the approach. And it seemed like openness and interoperability was the underlying thread there. There were a couple good nuggets about context is almost like a-- just like we have in the data world with this bronze, silver, gold, there's medallion architecture for data. There's almost a medallion architecture for context. Bob shared his Snowflake bible that he built when he was there. Chances are that has everything any agent might need, but maybe you don't need all ninety pages of the bible every single time.
And so you do have to curate little snippets and artifacts and metrics and pull things out of there to increase the performance and efficiency of your agents. And with that, that brings us to the end of Context Conference. If you're still hungry for more context, you, you're telling me, "Austin, this is great. I need more examples and more hands-on use cases," well, we are not here to disappoint you. Please follow the QR code. You can sign up for our Context Summit, uh, where we'll have real customers coming and sharing exactly how they're navigating everything from AI-ready data to context for their agents, uh, in production. So I will stop sharing my screen for the last time of Context Conference twenty twenty-six.
You have all been part of the first conference focused solely on context and making AI actually work for your business. So with that, I want to say thank you to our speakers. Thank you to everyone who participated in building content, and thank you to everyone who built the, uh, products that we are now releasing. Uh, but ultimately, thank you to everyone that attended. We had, you know, nearly fifteen hundred people live, thousands more registered that will watch this after the fact, and it's just been such an amazing opportunity to kind of go on this learning journey with you. So with that, I will officially wrap Context Conference twenty twenty-six, and I hope everyone enjoys the rest of their day.
Thank you.
