@lindavivah

What do NYC and AI agents have in common? Neither one sleeps! …which means they need an intelligent data platform to run on & MongoDB is exactly that.

⏳ Here’s a 90 second crash course on what they just launched, the most significant expansion of their platform yet:
 🔹 𝗠𝗼𝗻𝗴𝗼𝗗𝗕 𝟵.𝟬: agents hit the database with reads and writes all at once, and a spike can take a cluster down. 9.0 delivers up to 2x higher throughput on large instances vs 8.0, and Intelligent Workload Management queues the spikes so operations keep moving. And Queryable Encryption, which lets you query sensitive data while it stays encrypted, now also covers prefix, suffix, and substring searches

🔹 𝗔𝘁𝗹𝗮𝘀 𝗜𝗻𝗳𝗶𝗻𝗶𝘁𝗲: compute and storage used to scale together, so you either sized for peak and paid for idle capacity or sized for normal and got taken down by a spike. Infinite separates compute from storage so each scales independently, you pay for the capacity you use, and there are no application code changes.

🔹 𝗔𝘁𝗹𝗮𝘀 𝗔𝗴𝗲𝗻𝘁 𝗘𝗻𝗴𝗶𝗻𝗲: without built-in memory, agents start every conversation from zero, and the guardrails, identity, and audit around them get stitched together from separate tools. Agent Engine builds memory and retrieval (Voyage AI) into the platform so agents remember and act on accurate context with fewer tokens, logs every action against a real identity, and runs on open standards (MCP, A2A) with any model or framework. 
 There was much more, from Voyage’s new reranking models running right inside the query pipeline to a Cognition partnership that puts Devin to work moving legacy apps onto Atlas. Comment BUILD & I’ll DM you the link to the keynote 💚 Thank you so much to the MongoDB team for having me! It was a blast & always a joy seeing the community!! 🎉

Imagine being the IT department for 7 million people! 🤯 That’s the Indiana Office of Technology. I got to walk with Brad Welsh, tech leader at IOT, as he breaks down how they keep state services reliable for millions of residents & how @splunk Observability helps them proactively find and fix issues fast Brad put it best: “I approach it like I’m a citizen, which I am.” Thank you for taking the time to chat at #splunkconf26 !

“Bring the AI to the data, not the data to the AI.”
 For air-gapped & on-prem systems like hospitals, banks, government - the data can’t leave. 

I got to walk with VP of Global Product for the @splunk Platform at Cisco - Seth Brickman - as he breaks this down!

… + watch till the end to see the coolest sneakers! 👟

TLDR on some of the launches built for air-gapped and on-prem systems:

🔹 Cisco AI POD for Splunk, their new launch with NVIDIA: Splunk’s AI running in your own data center instead of the cloud. On-prem, private cloud + fully air-gapped. 🔹 Desk-side computing, the “whole new class of computing” described at the keynote: your laptop is for you, and the desk-side computer next to it is for your AI agents, working 24/7. 🔗 Comment INFO & I’ll DM you the links to learn more about these launches Thank you Seth & the Splunk team for an amazing conversation at #splunkconf26 ! 💙 


⏳ 90 second crash course on @splunk ’s (Cisco) biggest launches at .conf26! It all came back to three questions: Can I trust it to do the job? Can I afford it? Can I secure it? Here’s how the launches address this⤵️ 🏢 Cisco AI POD for Splunk Runs Splunk AI on-premises, in private cloud, or air-gapped on Cisco Secure AI Factory with NVIDIA, so teams whose data can’t leave can finally use it. Available today. 🔍 Splunk Agent Observability Now in Splunk Observability Cloud and Cisco Cloud Control. It evaluates agent and model behavior and applies runtime guardrails, so hallucinations and sensitive data leaks get blocked at runtime instead of after. 🪙 Tokenomics Agents already use ~5x more tokens than humans, per Jeetu Patel’s keynote. Tokenomics attributes spend across AI agents and coding agents like Claude Code, Codex, and Cursor, and will forecast where the bill is headed before the billing period ends. 🛡️ Splunk Agentic SOC Workforce New agents across detection engineering, threat hunting, autonomous investigation, response, and policy governance, so analysts supervise the work instead of doing all of it by hand. Plus a multi-year Splunk and AWS agreement to co-develop security solutions. So much more was announced at .conf26! 🔗 Comment SPLUNK & I’ll DM you the full announcements & keynote replay #splunkconf26

Gartner just shared a forecast: 2026 is the first year cloud spending on AI inference passes AI training ($23.3B vs $19B). Training is how a model gets built. Inference is every time anyone actually uses it. And now that more companies are taking AI into production and building agents, there’s a lot more calling of models than training them. Here’s a quick breakdown 👇 🔹 Inference is memory-bound. To generate every single token, the GPU reads basically the whole model out of memory. For a 70B model at 16-bit, that’s 140GB of reading. Per token. 🤯 🔹 Since the weights are frozen at inference, you can copy a model anywhere. So instead of thousands of GPUs packed together like training, you can serve replicas from data centers all over the world, closer to your users. 🔹 But distributed compute still needs to connect back to the storage, data, clouds and apps the model depends on. And that’s where the network ends up being the interesting part. ^ The industry is moving in this direction & I’m excited to partner with @megaport on this video because they combine all these pieces: a vendor-neutral private network that’s already inside 1,100+ data centers across 30+ countries, and with Latitude.sh now part of Megaport, bare metal CPUs and GPUs you can add right into that same architecture. Love how the Latitude team put it: “Training rewards scale in a single location. Inference rewards being everywhere your users are.” 🔗Comment INFERENCE & i’ll DM you the link to try out Megaport #MegaportPartner