Snowflake’s Post

I just spoke to a large regional bank today about their AI and data strategy. This was one of my key points: In today’s crowded world, organizations are faced with having to make an informed decision about where and how to build agents. There are too many options and too much noise. You must cut through that hype and noise by maintaining high standards for ease of adoption during the selection process. In other words, in 2026, you should demand more than to simply be able to say that you can build agents on [insert solution platform here]. That is no longer unique. Don’t go after the new, shiny thing without considering whether your team can actually see a clear path to scalability past the MVP. Of course, speed also matters. But you should not think that means compromising on security and completeness of vision. With Cortex Code, you can build and ship faster while still maintaining a high quality bar. If you’re trying to race against the clock to build AI agents today that will deliver real value tomorrow (instead of in a theoretical future), why would you settle for less? Don’t take my word for it. Look at what our customers are saying.

First it was just data, then it was ML, and now it’s Agentic. It’s cool to see how Snowflake is making these tools accessible for everyone, not just the hardcore coders. Definitely worth a look if you're trying to move faster with your data. #AI #DataProduct #Snowflake

Spent the morning at Snowflake Data for Breakfast in downtown Tampa. This mini session focused on AI Strategy & Implementation with Snowflake. My personal takeaways: The reality of AI maturity (low‑ROI pilots, hallucinations, fragmentation) and three must‑have pillars: embed AI into workflows, unify business context/semantics, and build a solid data foundation.

Context engineering is becoming its own discipline — the way data engineering did a decade ago. If you're trying to get up to speed, this is where I'd start. 44+ guides on context layers, context graphs, and making AI actually work in enterprise environments.

Klika Tech recently launched VELOCITY to help companies go from AI pilots to something that actually works. I sat down with Michael Klimaszewski (surrounded by an impressive amount of Batman gear) to talk about: *why most companies are still AI curious *the role of data foundations  *what the 5 phases of VELOCITY actually look like *how long it really takes to implement AI properly (not overnight) *what success looks like at the end of an AI rollout Watch the full interview and learn more about VELOCITY: https://lnkd.in/eAh_ZTPX

This is exactly the kind of AI thinking that matters right now — sharing this one for the whole feed. https://lnkd.in/gaE-frz5

Every once and a while in this life, topics of great happiness merge. In the case of the interview below with Kate Strachnyi of DATAcated, it was friendship and innovation! It was an absolute pleasure getting to sit down with Kate and discuss how Klika Tech is helping Middle Market and Enterprise customers implement and scale AI through the #VELOCITY program. This 5-phase, 16 month, fixed price consulting and development program puts the focus on #AIOps and data/cloud infrastructure development so that customers have the foundational elements in place to implement successful AI tools at scale, regardless of the model. By shifting left and putting #ROI driving use case identification and #governance ahead of model implementation, VELOCITY ensures a successful AI strategy investment and keeps companies focused on growth, not wasting cycles on trial and error. This was a wonderful conversation and Kate was a truly lovely partner for a Wednesday discussion. Enjoy!! Interested in #VELOCITY? Learn more here: KlikaVELOCITYLaunch.com

As more teams start building with agents, one thing is becoming very clear: It is not enough for an agent to occasionally get the right answer. It has to be reliable, efficient, and production-ready. That is why I’m excited for Josh Reini’s upcoming talk at AI Camp Atlanta: “Optimize your Agent’s GPA with Coding Agents.” The session will focus on a topic that feels more relevant than ever right now — how we think about evaluating and improving agents in a practical way as they move closer to real-world use. If you are working with GenAI, LLMs, or agents, this should be a valuable one to attend. March 26 | 5:30 PM | Improving, Alpharetta AICamp, in collaboration with Snowflake and Improving 👉 Event: https://lnkd.in/eRs8FNJt

Some use RAG without knowing what's inside it. I built it twice so I couldn't ignore it. Once with Vertex AI's managed RAG Engine. Once with ChromaDB: chunking, embedding, vector store, retrieval, and generation all written explicitly, step by step. Same documents. Same queries. Same answers. That's the point. Vertex AI gives you a single Tool.from_retrieval() call and the entire pipeline disappears. ChromaDB makes you write every step yourself. Running both back to back made something clear that I'd missed before: The RAG pattern doesn't change. Only the infrastructure layer does. That distinction matters when things break. The managed version surfaces region restrictions, deprecated SDKs, and cross-region import errors as cryptic failures, because the internals are hidden. The self-built version makes the same class of problems completely visible. Neither is wrong. But you can't debug what you can't see. For those evaluating managed RAG vs self-built: what's driving the decision, operational simplicity or observability?

I keep hearing the same question from founders: "Should I use Claude or Codex? What about Antigravity?" So I ran an experiment. Same prompt, same task, three platforms at once. "Build a landing page for my AI workshop." They all did the same thing. Read the workspace. Made a plan. Wrote the code. Spun up a local server. Three working landing pages. The outputs converge because every AI agent runs the same loop: observe, think, act. The platform is just the car. The driving skill transfers. Learn context files. Learn memory. Learn tool connections. Those work everywhere. Stop comparing dashboards. Learn to drive.