Atlan

Tredence Inc. has built AI agents inside some of the world's largest enterprises. On Aug 12, they show exactly how, live. Anugraha Sinha, Director of GenerativeAI at Tredence, joins Austin Kronz, AI and Data Strategy at Atlan. Come learn: → The mental model experienced teams use before the first line of code → What Tredence has learned building this at scale, including what breaks first → The reference architecture: where the context layer sits, how definitions and access get set Aug 12. 11 AM ET.

Super proud to bring something into the world today. 🎉 I'd love to introduce you to what has been a genuine labour of love: Rex, a simulation engine that tests our AI products before we ever launch them to the world. For five years, our research team studied the humans of data like anthropologists: 800+ interviews and 20,000 real queries, watching exactly where people trust an answer, and where they quietly give up. Then we turned all of it into personas that stress-test our AI before it ever meets a real user. This is the story of Rex: five years of anthropological research, encoded into personas that find the walls before your users have to. Link in the comments :) Atlan

Shared some learning & stories Atlan's AI Native journey last week at the UnBoxed Summit. Here's the tl;dr 👇 1. Execution is not the bottleneck anymore. Taste, judgement, ambition will be be the differentiators 2. Optimize for outcomes & not tasks. Small functional pods can own outcomes end to end 3. Build the shared context layer / source of truth for humans & agents. 4. Don't forget to have a lot of while you are at it! Shoutout to Sujit for moderating the discussion (and grilling beforehand 😂) and Aadil & Advait for sharing your journey & lessons as well! Also to the BambooBox team for putting everything together. On a personal note, finally met the legend Parminder Singh (aka Parry) in person!

25 years building ontologies. Zero mentions of "context layer," until the last two. Jessica Talisman founded The Ontology Pipeline before the AI industry started using that phrase. On July 30, she joins Atlan's Austin Kronz to work out whether it's a genuine evolution or the same discipline wearing a new name. Thursday, July 30. 11 AM ET. Live AMA, no script. Register → https://lnkd.in/eEvhZ5TS

Alex Karp called it AI sovereignty. David P. Mariani from AtScale calls it competitive corpus. The point is the same: your business semantics are intellectual property. If every company runs the same frontier model, intelligence is commoditized. The differentiator becomes how you codify your business, your metric definitions, your fiscal calendars, your rules for calculating revenue. In the agentic era, that knowledge gets encoded into semantic layers, skills, and context infrastructure. Whoever owns that layer owns the advantage. Whoever hands it to a single vendor gives away the thing that makes them different. Dave joined us this week for an episode of WTF is the Context Layer to break down where semantic layers fit in the stack and why they've become load-bearing for AI. Watch the full conversation → https://lnkd.in/eB5nMXX6

At Databricks Summit, Mastercard's SVP of Data & Analytics Vivek Radhakrishnan shared what a decade of governance investment looks like when AI arrives. They'd already done the hard work for BI. Catalog built, lineage mapped across systems, steward agreements in place. A full marketplace of governed data products. So when they deployed context agents against that catalog, the agents could draw on years of accumulated work. Ownership already assigned. Definitions already agreed on. Lineage already traced. 30,000+ assets enriched. 6,000+ hours saved. And each new agent builds on what the last one learned, because the governance work compounds. A lot of teams have BI-era governance sitting there. Turns out the best AI foundation is the governance work most teams already started years ago. Read the full story → https://lnkd.in/eXPCtCyr

We had an agent loop running in early 2026, and it still failed. Our reliability team built Sherlock, an investigation agent for production incidents. The first version investigated every alert, 11,000 a month, with the same intensity. It burned through our API budget on noise and re-investigated the same flapping alert from scratch every time it fired. So we built the scaffolding around it. A non-LLM filter that suppresses 85% of alerts before a model ever runs. Memory that includes past wrong hypotheses so the agent stops walking into the same mistakes. A hybrid mode that starts agentic but falls back to deterministic when confidence stays low.                                                                                                              Investigations that took 10+ minutes now finish in about 2 minutes, at less than a third of the cost. The full engineering story from Arnab Saha, Alok Sharma, and Nikhila Jain is on the engineering blog.        https://lnkd.in/eUDsttfN

Thrilled to see my first AI Engineer talk featured in Latent Space’s roundup of the five trends defining AI engineering—alongside work from Romain Huet, Thariq Shihipar, Pauline Brunet, Philipp Schmid, and Garry Tan. AI Engineer has long been one of my favorite sources for substantive thinking about AI, so this feels both very cool and slightly surreal. My talk, “WTF Is the Context Layer?”, explores a question I believe every company building agents will soon need to answer: How do you give agents the unique, situated knowledge of your business? I break down what a context layer is, why it matters, the different approaches emerging today, and which ones I believe will win. The talk is now live. I’ll share the link in the comments. Latent Space | AI Engineering News, Essays and Interviews