Rob Strechay at the next place

Everyone is talking about AI models. But after spending time at @VAST_Data's VAST Forward, one thing became very clear: Agentic AI isn’t a model problem. It’s a data platform architecture problem. As organizations move beyond copilots into AI systems that take action, the real bottleneck isn’t the LLM. It’s the platform underneath: • governed data access • high-speed data platform services • persistent AI memory • GPU-accelerated data services • distributed infrastructure that can manage fleets of AI systems At VAST Forward, the conversation wasn’t about bigger models. It was about how the data platform becomes the runtime for AI agents. That shift is huge. The companies that win in the next wave of enterprise AI won’t just deploy better models; they’ll build data platforms designed for agentic systems. I break down the four architectural shifts emerging from VAST Forward and what they mean for platform teams, CTOs, and data engineers. 👇 Short video (6 min) and Full article below linkedin.com/pulse/agentic-…

Wrapping up the week with a writeup on @NetApp 's acquisition of DataPelago, with analysis from @RealStrech and @simonrob_esg about how it shakes up competitive dynamics in #AI data management vs @Dell, @EverpureData, @nutanix, @VastData and more: techtarget.com/searchstorage/…

The @databricks CustomerLake CDP makes a lot of sense, given it is probably the number one use case I have used Databricks for and many others. @tasso did a good job breaking it down in customer360, like 1st & 3rd party ID resolution, & Campaign Agents - really complete vision

AI costs are going up so much - @alighodsi @databricks kicks off day 2 with a recap of day 1 and reemphasizing the future is open with no lock in

Solving ML and Agentic partnership for @databricks customers - this is huge for a good number of customers - especially the ZeroOps for ML

From Apps to ML solving for the challenges

@databricks @kuhlenhuth @SpaceX Deep diving into Unity AI Gateway - agent registry, contextual policies, and the concept of budgets - smart routing, which is also coming to vLLM - agent tracing via MLflow | also adding Memory Service & Sandbox

AgentBricks now taking center stage for @databricks customers - explaining the how dev teams get stuck on 1/ frontier changes constantly 2/ competitive advantage is data in silos 3/ most privileged actors, sometimes too much from @kuhlenhuth - announcing Grok 4 from @SpaceX too

Giving the ability to share sessions and use multi-agent workflows with transparency extended to policies - such as the cost of each request to the agents.

This is very cool AI tech to help simplify and secure agents. It's not everything, but it is going in the right direction. Hope to see this in the @linuxfoundation AAIF ecosystem

As @matei_zaharia discusses the limiting factors in current way you use multiple agent harnesses

Next, @matei_zaharia dives deep into Omnigent meta harness that was open-sourced this past Saturday - this could be one of the most important announcements of the week in how context is gained at the harness layer

How do you make AI to accessible and cost efficient for everyone? Partner with folks like @RIL_Updates - open models and infrastructure is required.

Another great announcement is @databricks having true cross cloud disaster recovery for Lakebase - still some interesting networking that is not there yet, but they have gone to using managed private endpoints in each cloud, so you are still paying egress - @Mastercard customer

AI takes center stage with its life blood, data, at the @databricks Dais - with 20k customers at north of $5b in revenue equaling $250k acv.

My vote for best announcement today from @databricks is Genie ZeroOps

Global Chief Data & AI Officer Magesh Bagavathi from @PepsiCo shows off the new logo from the rebrand last year - talking about leveraging Genie - supply chain insights is the key use case, which is not surprising with 30k interactions out of the gate

@databricks Tid bit that is interesting and makes sense is Genie One MCP app for plugging it into other agents - good use of this

The how Genie Ontology works makes a lot of sense and is required for agents to build on accuracy for process intelligence, like that of @Celonis, as agents need more than a snapshot

Moving on to the "why use Genie One" section - interesting research @databricks did on coding agents showing they are less than 50% accuracy - missing context layer - which leads us to Genie Ontology

Next opinionated app CustomerLake is a CDP built in partnership with many organizations - this is an interesting and logical second place for @databricks to go as many CDPs are built on them, and puts them in a better position to compete with @Google BigQuery for example

Now buying Python SIEM Panther