<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by piyush gupta Pythonista | Data Engineer on Medium]]></title>
        <description><![CDATA[Stories by piyush gupta Pythonista | Data Engineer on Medium]]></description>
        <link>https://medium.com/@piyush9194?source=rss-3489e9c7ed41------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*h_ovKuL4N3F9C7vdRGrNMw.jpeg</url>
            <title>Stories by piyush gupta Pythonista | Data Engineer on Medium</title>
            <link>https://medium.com/@piyush9194?source=rss-3489e9c7ed41------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 09:57:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@piyush9194/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Snowflake Cortex Agents: The Enterprise AI Framework That Brings Agentic Intelligence to Your Data]]></title>
            <link>https://medium.com/@piyush9194/snowflake-cortex-agents-the-enterprise-ai-framework-that-brings-agentic-intelligence-to-your-data-0c9e2be7a283?source=rss-3489e9c7ed41------2</link>
            <guid isPermaLink="false">https://medium.com/p/0c9e2be7a283</guid>
            <dc:creator><![CDATA[piyush gupta Pythonista | Data Engineer]]></dc:creator>
            <pubDate>Mon, 09 Mar 2026 01:06:10 GMT</pubDate>
            <atom:updated>2026-03-09T01:06:10.341Z</atom:updated>
            <content:encoded><![CDATA[<p><em>How Snowflake is turning your data warehouse into an intelligent, autonomous reasoning engine — and why it matters for every data team in 2026.</em></p><p>📖 12 min read · Feb 2026 · <strong>Deep Dive — Agentic AI</strong></p><p>If you’ve been tracking the agentic AI wave sweeping through enterprise tech, you’ve likely noticed a pattern: every major platform is racing to make AI agents a first-class citizen. Snowflake is no exception — but their approach is notably different. Rather than building yet another general-purpose agent framework, Snowflake has gone deep on what they know best: <em>data</em>.</p><p>The result is <strong>Cortex Agents</strong>, a fully managed service that became generally available in November 2025, and it’s quickly becoming one of the most compelling reasons to build your AI applications inside the Snowflake ecosystem.</p><h3>What Exactly Are Cortex Agents?</h3><p>At their core, Cortex Agents are autonomous AI orchestrators that sit on top of your Snowflake data — both structured and unstructured — and reason over it to answer complex business questions. Unlike a simple chatbot wrapper on an LLM, a Cortex Agent can <strong>plan</strong> multi-step tasks, <strong>select</strong> the right tools for each subtask, <strong>reflect</strong> on intermediate results, and <strong>iterate</strong> until it arrives at a high-quality answer.</p><blockquote><strong><em>Key Distinction:</em></strong><em> A traditional RAG pipeline retrieves context and generates a response in a single pass. A Cortex Agent operates in a plan → execute → reflect cycle that can span multiple rounds of tool use before producing a final answer.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/908/1*PAqgoTorgP0SdTkC5T4nuA.png" /></figure><p><em>[Fig 1. High-level architecture — the agent orchestrates across structured, unstructured, and custom tools within Snowflake’s security boundary.]</em></p><p>The agent sits at the center of everything, receiving a user query and routing it across three tool types — Cortex Analyst (structured data), Cortex Search (unstructured data), and Custom Tools (business logic) — all within Snowflake’s governance perimeter where data never leaves the boundary.</p><h3>The Agentic Loop: Four Phases of Intelligence</h3><p>Cortex Agents follow a well-defined four-phase workflow. This loop is what separates them from traditional retrieve-and-respond pipelines — the agent can self-correct, try alternative approaches, and combine multiple data sources before committing to an answer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/903/1*6Z5SKDrpDRVgNqgDSaCXYw.png" /></figure><p><em>[Fig 2. The four-phase agentic loop — Plan, Execute, Reflect, Respond — with iterative feedback when results aren’t sufficient.]</em></p><p>This loop means a Cortex Agent can handle questions that would stump a single-pass RAG system. For example, <em>“What are the differences between contract terms for Acme Supplies and Acme Stationery?”</em> requires the agent to split the task, search two contracts, extract the relevant terms from each, and then compare them — all orchestrated automatically.</p><h3>The Three Tool Types</h3><p>The agent’s power comes from the tools it can invoke. Each tool type is purpose-built for a different data modality, and the agent dynamically selects the right one based on the query.</p><h3>📊 Cortex Analyst — Structured Data</h3><ul><li>Natural language → SQL generation</li><li>Semantic views defined in YAML</li><li>Auto-generated charts and visualizations</li><li>Query timeout controls</li><li>Works with: Tables, Views, Warehouses</li></ul><h3>🔎 Cortex Search — Unstructured Data</h3><ul><li>Hybrid vector + keyword retrieval</li><li>Semantic reranking using Arctic Embed M model</li><li>Auto-refreshing indexes</li><li>Citations included in results</li><li>Works with: PDFs, Contracts, Documents</li></ul><h3>🔧 Custom Tools — Business Logic</h3><ul><li>Stored procedures and user-defined functions</li><li>External API calls</li><li>ML model inference</li><li>Supports: Python, Java, SQL UDFs</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/916/1*T3B1ZjXIkmkRdY4WWnKFyA.png" /></figure><p><em>[Fig 3. The three tool types — each handles a different data modality, and the agent selects automatically.]</em></p><h3>What Makes Cortex Agents Different?</h3><p>The enterprise AI agent space is crowded. Google, Salesforce, Microsoft, and countless startups all offer agent-building capabilities. Here’s how Snowflake differentiates:</p><p><strong>🔒 Data Never Leaves Snowflake</strong> Queries execute within your virtual warehouse. All RBAC, masking, and compliance policies apply automatically. For regulated industries, this is a non-trivial advantage.</p><p><strong>🔄 Unified Structured + Unstructured</strong> Query a revenue table and search a contract PDF in the same conversation. Most frameworks force you to choose — Cortex Agents do both seamlessly.</p><p><strong>☁️ Fully Managed, Zero Infra</strong> No vector databases to provision, no orchestration servers to maintain. Define your agent and Snowflake handles scaling, availability, and compute.</p><p><strong>🤖 LLM Flexibility</strong> Optimized for Anthropic’s Claude by default, with support for OpenAI GPT-5.2 and other models. Choose the right model for your use case.</p><h3>How RAG Compares to Agentic</h3><p>If you’re coming from a RAG-based architecture, here’s how Cortex Agents extend the pattern:</p><p>Capability Traditional RAG Cortex Agent <strong>Data sources</strong> Unstructured only (documents) Structured + unstructured + custom <strong>Reasoning</strong> Single-pass retrieve → generate Multi-step plan → execute → reflect loop <strong>Tool selection</strong> Fixed retrieval pipeline Dynamic tool routing per subtask <strong>Error handling</strong> Returns whatever it gets Reflects, retries, asks for clarification <strong>Governance</strong> Typically external to data platform Inherits Snowflake RBAC natively <strong>Infrastructure</strong> Vector DB + orchestrator + LLM Fully managed, single API</p><h3>Enterprise Integrations</h3><p>Snowflake isn’t keeping Cortex Agents locked inside Snowsight. The platform has expanded the agent’s reach through key integrations:</p><ul><li><strong>📱 Snowsight UI</strong> — No-code agent builder, native to the platform</li><li><strong>💬 Microsoft Teams</strong> — Chat-based data queries directly in Teams</li><li><strong>🤝 Microsoft 365 Copilot</strong> — Snowflake insights surfaced inside Office apps</li><li><strong>🔌 MCP Server</strong> — Open standard protocol for interoperability</li><li><strong>🧩 Dataiku / Partners</strong> — Agent evaluation and composition</li><li><strong>🏗️ Custom Apps</strong> — Streamlit, React, or any REST-based interface</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/922/1*y67aJr0bs1EH8l8CEZ248w.png" /></figure><p><em>[Fig 4. The growing integration ecosystem — from Snowsight to Teams to the open MCP protocol.]</em></p><p>The <strong>MCP (Model Context Protocol) server</strong> is perhaps the most strategically important integration. It positions Snowflake not just as a place to <em>build</em> agents, but as a data backend that <em>any</em> agent in your ecosystem can tap into — from Salesforce Agentforce to your own custom framework.</p><blockquote><em>The Snowflake MCP Server enables any MCP-compatible agent to use Cortex tools — no extra infrastructure needed.</em></blockquote><h3>Building Your First Agent</h3><p>Getting started is more straightforward than you might expect. Here’s a practical path:</p><p><strong>Step 1 — Define a Semantic View</strong> Create a YAML-based definition that maps business concepts to your database tables. This is what powers Cortex Analyst’s text-to-SQL capabilities.</p><p><strong>Step 2 — Create a Cortex Search Service</strong> Point it at a manageable corpus — internal policies, contracts, or product docs. Snowflake handles chunking, embedding, and index management.</p><p><strong>Step 3 — Wire Up the Agent</strong> Use the Snowsight UI (AI &amp; ML → Agents) or the REST API. Add both tools, configure system prompts, and set orchestration instructions.</p><p><strong>Step 4 — Tune Orchestration Instructions</strong> Natural language hints dramatically affect routing quality. Tell the agent when to prefer Analyst vs. Search, and how to handle edge cases.</p><p><strong>Step 5 — Add Custom Tools &amp; Iterate</strong> Register stored procedures or UDFs for domain-specific logic. Monitor performance, collect user feedback, and refine the agent over time.</p><p>Here’s a simplified API example of how an agent definition looks:</p><pre>{<br>  &quot;model&quot;: &quot;claude-4-sonnet&quot;,<br>  &quot;tools&quot;: [<br>    {<br>      &quot;type&quot;: &quot;cortex_analyst_text_to_sql&quot;,<br>      &quot;name&quot;: &quot;revenue_analyst&quot;,<br>      &quot;semantic_view&quot;: &quot;db.schema.revenue_view&quot;<br>    },<br>    {<br>      &quot;type&quot;: &quot;cortex_search&quot;,<br>      &quot;name&quot;: &quot;contract_search&quot;,<br>      &quot;search_service&quot;: &quot;db.schema.contract_svc&quot;<br>    }<br>  ],<br>  &quot;instructions&quot;: {<br>    &quot;system&quot;: &quot;You are a sales analytics assistant.&quot;,<br>    &quot;orchestration&quot;: &quot;Use Analyst for revenue queries. Use Search for contracts.&quot;<br>  }<br>}</pre><h3>Practical Use Cases</h3><p><strong>📈 Sales Intelligence</strong> Combine CRM data (Cortex Analyst) with contract documents (Cortex Search) to answer “What’s our renewal risk for Q2?” in a single conversation. The agent queries revenue tables and cross-references contract terms automatically.</p><p><strong>🏛️ Data Governance</strong> Agents that explore, document, and govern database objects by combining metadata queries with AI-powered description generation via custom tools. Automate your data catalog maintenance.</p><p><strong>💰 Financial Analysis</strong> Pull earnings data, search through call transcripts, and cross-reference market reports — all within Snowflake’s compliance boundary. Ideal for investment research and risk assessment.</p><p><strong>🎧 Customer Support</strong> Search knowledge bases and query customer account data simultaneously. Give support teams instant, contextual answers about any customer’s situation without switching tools.</p><h3>Limitations to Be Aware Of</h3><blockquote><em>⚠️ </em><strong><em>Be clear-eyed about the trade-offs.</em></strong><em> No technology is perfect, and understanding the boundaries helps you make better architectural decisions.</em></blockquote><p><strong>LLM selection is currently limited.</strong> The supported model list is smaller than what you’d get with a bring-your-own-model framework. The optimization for Claude is a strength if you’re using Claude, but constrains flexibility if you need a specific model.</p><p><strong>No visual workflow builder.</strong> Agent creation requires the Snowsight configuration UI or API/SQL definitions. If you need a drag-and-drop visual builder, that’s not part of the current offering.</p><p><strong>Accuracy isn’t guaranteed.</strong> Snowflake is transparent about this: LLM-generated responses and citations should be reviewed before serving to end users. Human-in-the-loop is still essential for high-stakes decisions.</p><p><strong>Ecosystem is still maturing.</strong> While GA, the custom tools ecosystem and third-party integrations are evolving. The MCP server is promising but production patterns are still being established.</p><h3>Final Thoughts</h3><p>Cortex Agents represent Snowflake’s bet that the future of enterprise AI isn’t about building the best model — it’s about building the best <strong>bridge between models and data</strong>. By making agents a native, governed, fully managed capability within the data platform, Snowflake is removing the infrastructure tax that has slowed enterprise AI adoption.</p><p>The agentic AI space is still young, and the patterns for building production-grade agents are still being written. But if your data already lives in Snowflake, Cortex Agents offer a remarkably low-friction path from “we want to build AI agents” to actually having them running against real enterprise data with real governance in place.</p><p>That’s a meaningful head start.</p><p><em>🎉 If you found this article helpful, give it a clap and follow for more deep dives on data engineering and AI infrastructure. Have questions or experience building with Cortex Agents? Drop a comment — I’d love to hear about your use cases.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0c9e2be7a283" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>