Mihai DARZAN’s Post

📢 𝗗𝗮𝘁𝗮 𝗦𝘁𝗼𝗿𝗲 𝗶𝘀 𝗹𝗶𝘃𝗲 𝗶𝗻 𝗣𝗥𝗢𝗖𝗘𝗦𝗜𝗢. Last week’s Airtable acquisition news was a good reminder: 𝘄𝗵𝗲𝗻 𝘆𝗼𝘂𝗿 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻 𝗱𝗲𝗽𝗲𝗻𝗱𝘀 𝗼𝗻 𝗮𝗻 𝗲𝘅𝘁𝗲𝗿𝗻𝗮𝗹 𝗱𝗮𝘁𝗮 𝘀𝘁𝗼𝗿𝗲, 𝘁𝗵𝗮𝘁 𝗱𝗲𝗽𝗲𝗻𝗱𝗲𝗻𝗰𝘆 𝗰𝗼𝗺𝗲𝘀 𝘄𝗶𝘁𝗵 𝗶𝘁. We wanted to remove that friction. With 𝗗𝗮𝘁𝗮 𝗦𝘁𝗼𝗿𝗲, you can now create and manage tables directly inside PROCESIO – no separate database, credentials, or connection to maintain. Build your schema in the UI, then read and write data directly from 𝗣𝗿𝗼𝗰𝗲𝘀𝘀 𝗗𝗲𝘀𝗶𝗴𝗻𝗲𝗿 and 𝗙𝗼𝗿𝗺𝘀. You also get: • Table-level permissions • CSV import & export • Built-in data viewer And one important distinction: 𝗔 𝗚𝗹𝗼𝗯𝗮𝗹 𝗩𝗮𝗿𝗶𝗮𝗯𝗹𝗲 𝗵𝗼𝗹𝗱𝘀 𝗮 𝘃𝗮𝗹𝘂𝗲. 𝗗𝗮𝘁𝗮 𝗦𝘁𝗼𝗿𝗲 𝗵𝗼𝗹𝗱𝘀 𝗮 𝘁𝗮𝗯𝗹𝗲. Fewer dependencies. Less setup. More of your automation in one place. ✅ 👉 𝗦𝗶𝗴𝗻 𝘂𝗽 𝗮𝘁 𝗵𝘁𝘁𝗽𝘀://𝗽𝗿𝗼𝗰𝗲𝘀𝗶𝗼.𝗮𝗽𝗽 𝗮𝗻𝗱 𝘀𝘁𝗮𝗿𝘁 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴. #PROCESIO #AI #AIAgents #AgenticAI #AIOrchestration #AutomationOrchestration #WorkflowAutomation #BusinessAutomation #LowCode #EnterpriseAutomation

Last week's Airtable news was a good nudge: the data your automations depend on is only as safe as whoever owns the tool it sits in. 🤔 So I'm genuinely happy about this one. We shipped Data Store, real tables, right inside PROCESIO. No separate database, no credentials, no connection to maintain. A Global Variable holds a value. Data Store holds a table. One less thing living outside your control. ✅ 👉 https://procesio.app

Your AI agents are smart, but they are choking on your legacy enterprise data. Most enterprise systems still spit out massive, deeply nested XML files. If you try to feed that raw XML directly into a Large Language Model, you run into a wall. You waste precious context window space, run up massive token bills, and get unpredictable agent behavior. AI agents need clean, predictable JSON. But rewriting your entire legacy backend to support modern data formats is a multi-million dollar headache you do not need. The practical fix is building a lightweight schema adapter layer. By placing a dedicated adapter between your legacy database and your AI agent pipeline, you can parse complex XML feeds on the fly. This layer maps the old schemas to structured JSON objects in real time. Your legacy systems stay untouched, and your AI agents get clean, optimized data they can actually act on. At Oracon Global, our senior in-house team builds these high-volume data pipelines to bridge the gap between legacy infrastructure and modern AI workflows. We design custom middleware that ensures your data is clean, token-efficient, and ready for agentic workflows. Best of all, you own 100% of the code and IP we write for you. If you want to see how we handle live data, you can test out our interactive AI demos right now on oraconglobal.com. How is your team handling legacy data formats as you build out your AI roadmap? #EnterpriseAI #SoftwareEngineering #AIAgents #DataPipelines #XMLtoJSON 👉 Read more: https://lnkd.in/dkDG2TeM

AI agents are becoming operational, not just conversational. With Snowflake CoCo Automations now in Public Preview, you can schedule CoCo to run recurring tasks without needing to be actively interacting with it. The interesting part for me is the shift from: “Ask an AI agent to do something” to “Have an AI agent do it for you, on a schedule.” Think daily data checks, pipeline monitoring, anomaly detection, business summaries, cost monitoring, or even workflows powered by MCP servers. And because each run can be inspected through its Cortex thread, you get visibility into what the agent actually did. This is a small feature with potentially a big impact on how we think about agentic data operations. Nathan Scharfe Amilee Alesna Elsa Mayer #Snowflake #SnowflakeCoCo #CortexAI #CortexCode #AgenticAI #AIAgents Snowflake Developers

The most expensive line in your agent isn't the agent. It's which model each node quietly calls. In 26C, Oracle made that the entire basis of pricing — and most delivered agents are running a premium model nobody chose on purpose. Here's the 26C model, simplified. The old world had separate token licenses and custom-agent SKUs — customizing an out-of-the-box agent made it "custom" and therefore chargeable, which pushed people to avoid the customization they needed. That's gone. Pricing is now driven purely by the LLM you use. The basic LLM is free — even if you customized the agent or built it from scratch. Premium models bill by token consumption, measured in AI Units. Your subscription includes 20,000 AI Units per month, where one unit equals one US cent. Now the part that costs real money silently: many out-of-the-box agents default to a premium model like GPT-4.1 Mini. Reported figures put a full premium chat interaction around $0.05, dropping to roughly $0.03 on your own key via BYOLLM. Multiply by your daily volume and the "we didn't change anything" agent becomes a budget line. So the architect's habit for 26C: check the model on every node before go-live. Classification and extraction rarely need premium — put those on basic, reserve premium for the reasoning that earns it. The 26C usage tab now shows AI Unit consumption per agent inside Studio, so there's no excuse not to look. Free where you can. Premium where it pays. Measured always. That's Week 4 — the platform shift. Tomorrow Week 5 opens with proof: the lead-upload CSV workflow, end to end. Oracle AI Agent Studio · Post 28 #OracleAIAgentStudio #OracleFusion #26C #FinOps #EnterpriseAI

Beyond the Chatbot: Why MCP is the Only Protocol That Matters in 2026 Chatting with your data was a 2024 parlor trick. In 2026, the real winners are using Model Context Protocol (MCP) to let AI execute across their entire Snowflake and Databricks stack. For the C-Suite, the distinction is critical: Retrieval-Augmented Generation (RAG) was about visibility. MCP is about agency. While RAG allowed your AI to read your data, MCP allows your AI to act on it. This is the transition from passive observation to true agentic execution. Why MCP is redefining the enterprise data stack: 1. From Insights to Action MCP turns your data lakes into active participants. Instead of simply surfacing a report, AI agents can now trigger workflows and modify records directly within your secure environment. 2. Unified Architectural Standard By providing a universal interface between LLMs and data sources, MCP eliminates the need for brittle, custom-coded integrations for every new use case. 3. Scalable Operational Velocity By connecting your models directly to the context of your business—your Snowflake and Databricks repositories—you reduce the latency between data discovery and business decision. We are moving beyond the era of Ask and Answer. We have entered the era of Task and Execute. Is your organization still treating AI as a search bar, or are you building a protocol-driven engine for execution? #ArtificialIntelligence #Technology #DataStrategy #Innovation #FutureOfTech #ModelContextProtocol #AgenticAI #EnterpriseAI #Snowflake #Databricks #DigitalTransformation #AIStrategy

If you are passing AI outputs to downstream systems as raw strings, you don't have a data pipeline you have a data leak. Here is why: LLMs and agentic workflows generate multi-turn, high-dimensional context.  When that data hits legacy APIs and databases (CRMs, EHRs, ERPs), fields are truncated, coerced, and silently discarded. Without a protocol, there is zero schema validation, zero version lineage, and zero loss tracking. CORD introduces the standard wire format: `application/cord+json` What happens during `$ cord validate`: 1️⃣ Canonical Envelope: Encapsulates state snapshots and delta version chains with monotonic versioning. 2️⃣ Dual-Format Output: Carries the full-fidelity AI payload alongside the legacy-coerced schema. 3️⃣ Envelope Fidelity Score (EFS): Calculates normalized semantic preservation (0.0 to 1.0) with per-field mapping status (`FULL`, `PARTIAL`, `NONE`). 4️⃣ Cryptographic Seal: Validates `x_cord_digest` using SHA-256 for tamper-proof audit trails. JSON standardized serialization. Protobuf standardized RPCs. CORD standardizes AI interaction fidelity. Explore the open specification, GitHub repos, and CLI tester: 👉 https://cordspec.org #DataEngineering #SoftwareArchitecture #DevOps #OpenSource #APIDesign #BackendEngineering #LLMOps

The upgrade gap is not another connector. In today’s AI-engineering discussion, one proposed build is an orchestrator that coordinates work across systems. Useful direction—but coordination does not answer the harder question: is the row current, consistent, and fit for the one workflow about to use it? For a PostgreSQL-backed engagement or support workflow, `public.customers.status` is a small example. `enabled` and `active` may describe the same business state—or they may not. An AI system can analyze that ambiguity. It should not become the write authority because the ambiguity was inconvenient. Refinery checks whether current operational data is trustworthy for one approved AI workflow. It is a Verified Data Gateway: purpose-bound trust between a system of record and one AI/workflow consumer. The proof loop is short: evidence → approval → repair → exact readback → receipt #PostgreSQL #OperationalData #SystemOfRecord

Qlik just landed a Leader spot in the 2026 IDC MarketScape for Data Intelligence Platform Software. This is validation of something we've believed for a while: AI doesn't fix bad data, it just moves the consequences faster.

You can't scale AI on infrastructure that wasn't built for it. Data management is where that foundation gets laid. I've spent the last few weeks on OSEMN, but the ML process is only half the picture. Around it sits the ecosystem that makes AI actually hold up in production. Data management and infrastructure is the load-bearing part. Here's the sequence that makes it work: ➼ Define business requirements. Always the foundation. Everything downstream depends on getting this right. ➼ Assess the data landscape. Map your current tech, your stakeholders, and where governance already touches the data. ➼ Ingest and store. Stand up ETL or ELT processes built to handle the volume without breaking. ➼ Process and transform. Run quality checks and build data catalogs. This is the step people cut, and the one they regret. ➼ Enable access. Put controls in place with tooling that documents in real time. ➼ Monitor and maintain. Every process, continuously. You can't improve what you're not watching. Get this right and scale stops being a risk. It becomes the plan. Miss it, and no model you build on top will save you. What's mattered most in the infrastructure you've built? Tell me below. #AIInfrastructure #DataEcosystem #DataManagement #Scalability #AI Citation: Marketoonist, 2023.