Roger Haas

This week we announced more MongoDB product innovations that will help reshape what's possible in AI. Our new Voyage AI by MongoDB models deliver more accurate and cost-efficient results, while new partnerships for agentic AI workflows simplify scalability and reliability, making it easier than ever for businesses to build trustworthy AI solutions. Business and tech leaders across APAC are striving to unlock real-world impact from AI - whether it’s modernizing banking systems or building AI-first applications that scale globally. MongoDB’s ability to consolidate structured and unstructured data, power real-time decisions, and deliver integrated vector search has already demonstrated it's the ideal foundation for AI applications that demand accuracy, scalability, and adaptability in an ever-changing landscape. With these new advancements, I can’t wait to see how our customers push the boundaries of innovation. Read more here: https://lnkd.in/gADYwgyj #MongoDB #AI #VoyageAI #Innovation #AgenticAI

44

❓ Will 2026 be the year we stop talking about agentic AI and actually get serious about governing it? Over the last few weeks, #Databricks has quietly shipped a pretty wild combo: AI agents that don’t just write code, but build the data layer itself – with new data showing agents now create ~80% of enterprise databases and 97% of dev/test environments. Agentic Data Quality Monitoring in public preview, where AI looks at how data is produced and used to automatically flag and prioritize the most important issues – without a forest of brittle manual rules. MemAlign, a new way to build higher‑quality LLM judges that learn directly from human feedback, using scalable memory to keep improving over time instead of relying on one‑shot prompt engineering. All of this wired into MLflow and Unity Catalog, so evaluation, observability, and governance live in the same place as your agents and your data. Why this matters: When agents are building your databases and spinning up environments at machine speed, “check it manually later” is not a strategy. You need agents watching the agents: data‑quality agents, evaluators, and LLM judges that can explain why they scored a run a certain way, and logs that can stand up in a design review or audit. MemAlign‑style judges plus agentic monitoring turn evals into a continuous feedback loop, not a one‑time benchmark. The interesting shift here isn’t just more powerful agents. It’s that Databricks is treating governance, quality, judges, and explainability as first‑class citizens of the agent stack, not an afterthought bolted on once something goes wrong. 💡 If you’re experimenting with agents today, the real question isn’t “What can they automate?” It’s: “How fast can we wrap them in the right scaffolding (data quality, evals, and memory) so we can trust that autonomy at scale?” 🔗 References / inspiration: -> Databricks: “AI agents are no longer just writing code…” https://lnkd.in/eHyAKBk3 -> Matei Zaharia: Agentic Data Quality Monitoring on Databricks https://lnkd.in/eVU5um_b -> Matei Zaharia: MemAlign – building better LLM judges from human feedback https://lnkd.in/e9-sSKnq ->Databricks blog: MemAlign – Building Better LLM Judges From Human Feedback with Scalable Memory https://lnkd.in/e8PGcSpj

5

Friends and Colleagues... The story below resonates. In this era, the lesson is simple: show up with what you’ve built. Not a perfect deck. A working prototype. Here’s a hard truth I’ve learned — and it’s painful watching others miss it: The quality of any technology output — especially AI-driven decision intelligence — is a function of iteration speed. The teams that win are not the ones chasing every single iteration as a perfect cycle. They’re the ones compounding iterations. The faster you can build → test → refine → redeploy, the better the outcome. Quality emerges from velocity. It’s not about polishing the first version. It’s about accelerating the cycle. We’re in the middle of a once-in-a-generation building window. AI has collapsed the cost of prototyping and building. An imperfect, functioning system demonstrates more conviction than 50 slides ever will. I’m not dismissing communication — relationships and narrative still matter. But right now, builders win. The people who bring something real into the room — even if it’s rough — will outpace those who are still refining messaging. Seize this moment. Otherwise, while you’re optimizing presentations, while someone else is optimizing product. Let’s build — and show up with it.

17

4 Comments

The rise of LLMs and generative AI is unlocking new insight for enterprises but it’s also creating fragmentation, brittle pipelines, and unprecedented governance risk. In a new Forbes article, @Serge Lucio explains why the AI era requires a centralized automation control plane capable of hardening pipelines and enabling governed, ad hoc automation. Read the full article: https://lnkd.in/dR-Z8fia

2