March 2026 Summaries

Agentic retrieval extends traditional retrieval-augmented generation by enabling AI agents to decompose complex questions, conduct parallel searches across multiple sources, retain conversational context, validate results, and take actions through APIs and enterprise tools. Contextual data integration complements this approach by unifying live operational data, documents, chat histories, and other sources so agents can make more accurate, context-aware decisions. Effective enterprise architectures typically use planner-executor workflows, hybrid vector and metadata search, standardized connectivity through Model Context Protocol (MCP), tool calling, and agent-to-agent communication. The guide recommends cataloging data sources, prioritizing secure APIs, normalizing and filtering sensitive data, selecting frameworks such as LangChain, LlamaIndex, Haystack, CrewAI, or AutoGen, adding verification loops, and implementing monitoring, access controls, and audit trails. Examples in recruiting, analytics, and e-commerce illustrate how agentic systems can automate workflows and use live data, while the future points toward adaptive search, collaborative agent teams, and open interoperability standards, tempered by significant governance and security challenges.

Tableau Desktop Free Edition provides individuals and businesses with a no-cost desktop visualization tool that supports local saving, auto-recovery, unlimited live-query rows, and publishing to Tableau Public, although it cannot publish to Tableau Cloud or Server, share packaged workbooks, or use published Cloud/Server data sources. CData Tableau Connectors extend the edition’s connectivity to more than 350 applications, databases, and warehouses through certified Windows and Mac connectors that use SQL-92 to translate Tableau queries into live source queries, including filters, joins, groupings, and aggregations. By avoiding traditional ETL and local replication, the connectors aim to keep dashboards current while reducing data-staleness and security concerns, supporting sources such as GitHub, QuickBooks, Workday, HubSpot, Slack, and Microsoft Dynamics. Their shared SQL-based query engine is also positioned as compatible with future Tableau Cloud, AI-agent, MCP-enabled, and pipeline uses. Setup involves installing a connector, selecting it from Tableau’s “To a Server” connection options, authenticating with the source, and then using the resulting live data in Tableau dashboards.

ServiceNow data can be connected to Excel through CData Connect AI for Spreadsheets to reduce reliance on manual exports and provide live, bidirectional access for reporting, analysis, and updates. The guide distinguishes between a direct Excel add-in connection for analysts and ad-hoc use and an integration-platform approach for broader automated workflows across multiple systems, noting that ServiceNow requires a Connect Spreadsheets Pro plan. It recommends OAuth 2.0 authentication, token refresh management, role-based access, and least-privilege permissions to secure connections. Users can configure a ServiceNow connection in CData Connect AI, install the Excel add-in, import selected tables, refresh live data, and validate fields before updating records. Accurate synchronization also depends on field mapping, transformations, filtering, validation, and choosing either event-driven real-time updates for operational needs or scheduled batch synchronization for periodic reporting. Ongoing testing, logging, alerts, audit trails, permission reviews, and reconciliation are presented as important practices for maintaining reliable, secure data exchange.

Model Context Protocol migration is presented as a phased enterprise transformation that begins with a narrowly scoped, measurable pilot and expands only after organizations establish reliable access to legacy systems, security, identity integration, governance, performance, and observability. The recommended approach is to expose existing CRM, ERP, and support APIs through MCP adapters rather than rebuild core systems, while applying zero-trust controls such as input validation, least-privilege permissions, encryption, ephemeral credentials, RBAC, and audit logging from the outset. Connecting MCP tools to enterprise identity providers such as Okta or Azure AD helps ensure permissions follow organizational roles, while assigning tool owners, maintaining versioned schemas, and tracking changes can prevent schema drift and unauthorized behavior. Organizations are also advised to benchmark latency and error rates early, use caching and appropriate deployment models to improve performance, and monitor tool calls, failures, agent actions, and anomalous activity with telemetry and safety tooling. After validating the first deployment, teams should use its metrics, governance templates, and lessons learned to iteratively expand to additional domains without recreating integration debt; the text positions CData Connect AI as a managed platform supporting these activities through prebuilt enterprise connectors, passthrough authentication, caching, and centralized logging.

Connect AI’s Q1 2026 release introduces Custom MCP Tools, allowing organizations to centrally define, deploy, govern, and audit reusable AI agent actions across departments and applications. Alongside its existing Universal Tools for relational data discovery and actions across more than 350 enterprise sources, the platform now offers Custom Tools for scoped, pre-optimized workflow operations and Source Tools for tightly controlled system-specific actions. These tools are designed to reduce token use, limit unintended data exposure, improve reliability, and support transactional safety by using validated inputs, defined logic, and structured outputs. An example shows an AI assistant creating a Zuora subscription directly from confirmed deal details, replacing manual system navigation and data entry. Tools can be configured with validation rules and SQL logic, shared through workspace-based permissions, and monitored through detailed logs that record users, tool calls, executed queries, and accessed data. Connect AI also plans to introduce Toolkits later in 2026, bundling selected Universal, Source, and Custom Tools into governed MCP endpoints for particular teams or workflows.

SAP ODBC connectivity provides a standardized way for analytics, BI, and application tools to access enterprise data across SAP platforms such as HANA, S/4HANA, NetWeaver, SuccessFactors, and Datasphere. Establishing a connection involves identifying the target endpoint and its authentication requirements, selecting a compatible 32-bit or 64-bit driver, preparing network and firewall access, installing either native SAP or third-party drivers, and configuring a Data Source Name with server, port, credentials, and encryption settings. The process differs somewhat by operating system, with Windows using the ODBC Data Source Administrator and Linux or macOS commonly relying on configuration files such as .odbc.ini. Security practices include enabling SSL or TLS, validating certificates, and using secure credential and role-based access controls. Connections should be tested through ODBC tools or BI applications such as Excel, Power BI, and Tableau, then optimized with tracing, network checks, compatible SQL syntax, and performance testing for DirectQuery or import modes. The guidance also emphasizes centralized data governance, auditability, dataset management, and consistent management of SAP and third-party data sources, while presenting CData Drivers as a cross-platform option for unified connectivity across multiple SAP services.

SAP-to-Snowflake integration enables organizations to combine operational ERP data with external sources for analytics, dashboards, predictive modeling, and AI initiatives, but requires planning around SAP’s complex structures, data volumes, and module-specific extraction needs. A phased approach recommends prioritizing critical domains such as finance, sales, and materials management, validating pilot use cases through reconciliation and parallel reporting, and selecting ingestion methods—including batch loads, CDC, Snowpipe, streaming, or hybrid approaches—according to latency and cost requirements. The guide recommends a three-layer Snowflake architecture of immutable raw data, standardized staging data, and curated business models, while preserving SAP-specific semantics such as client fields, leading-zero keys, currencies, and reversal events. It favors ELT transformations within Snowflake using SQL, dbt, or Snowpark, supported by automation, version control, CI/CD, monitoring, and validation checks for data quality, schema drift, and anomalies. Security and governance practices include on-premises agents behind firewalls, role-based access controls, audit logging, dynamic masking, encryption, and zero-copy sharing, while gradual domain-by-domain expansion can support trusted enterprise reporting and AI use cases. CData Sync is presented as a commercial connector-based option offering SAP connectivity, CDC, scheduling, and secure on-premises deployment.

Launched in November 2024, the Model Context Protocol (MCP) is an open standard intended to let AI agents discover, authenticate with, query, and act on external enterprise data through a common interface, with its ecosystem reportedly expanding rapidly. The comparison centers on managed MCP platforms, which host infrastructure, maintain connectors, and provide centralized security and governance, versus self-hosted MCP servers, which give organizations direct control over infrastructure, networking, data residency, and custom configurations but require them to manage setup, authentication, patches, monitoring, and connector changes. CData positions its Connect AI product as a managed option offering a single endpoint for more than 350 data sources, semantic translation across differing schemas, and controls such as role-based access, passthrough authentication, field-level permissions, and audit logging. Managed platforms are presented as better suited to rapid prototyping and scaling across many sources because they can reduce time to a first live query and operational maintenance, while self-hosting may be preferable for air-gapped deployments, strict residency or compliance requirements, bespoke networking, and specialized performance needs. The text concludes that many organizations may use a hybrid approach, relying on managed connectivity for most systems while self-hosting exceptional workloads.

MongoDB’s document-based, often nested data model can be connected to Power BI for analytics through SQL-compatible connector layers, traditional ODBC/JDBC drivers, or custom ETL pipelines, with the appropriate method depending on deployment, governance, and technical resources. The approaching end-of-life of MongoDB’s BI Connector after September 2026 increases the urgency for Atlas, self-managed, and hybrid users to evaluate alternatives, including Atlas SQL Interface, managed platforms such as CData Connect AI, drivers, and Python-based transformations. Effective implementations require TLS encryption, restricted network access, read-only reporting accounts, role-based permissions, auditing, and monitoring. Because MongoDB schemas can evolve and contain nested objects and arrays, connectors or ETL processes must flatten documents into relational structures suitable for Power BI. Organizations must also choose between Import mode, which caches scheduled snapshots, and DirectQuery, which accesses live source data, then validate query accuracy, indexing, filter pushdown, and dashboard performance before production deployment. For Power BI Service refreshes, on-premises MongoDB commonly requires a gateway, while cloud connector layers may provide direct managed access.

Snowflake and Power BI can form a scalable analytics architecture when organizations prepare Snowflake for BI workloads with dedicated warehouses, auto-suspend policies, curated reporting schemas, and least-privilege role-based access controls. Dashboard performance depends on BI-ready data modeling, commonly using star schemas, aggregate tables, materialized views, or Dynamic Tables aligned to anticipated reporting queries. Teams can connect through native, ODBC, ADBC-based, or specialized third-party connectors, with the choice affecting query performance, authentication, and reliability across Power BI Desktop, Service, and gateway deployments. Import, DirectQuery, and composite models should be selected based on data freshness, dataset size, and Snowflake compute costs, while incremental refresh can reduce load by updating only changed records identified through timestamps. Additional optimization measures include query pushdown, caching, limiting report visuals, warehouse sizing, and query tagging. Secure deployments combine Snowflake RBAC with Power BI row- and column-level security, use OAuth or key-pair authentication, and rely on deployment pipelines and monitoring of both Snowflake queries and Power BI refreshes to maintain governance, reliability, and cost control.

Model Context Protocol (MCP) is presented as an open standard for connecting AI agents securely and consistently to enterprise tools, data sources, and workflows, reducing the need for costly, fragile custom integrations. As organizations move agentic AI from experimentation into production, MCP supports tool discovery, reusable skill composition, and event-driven orchestration, allowing agents to perform auditable, permissioned actions across systems such as CRMs, databases, and analytics platforms. The text identifies 2026 as a potential turning point for adoption, citing managed MCP offerings from major vendors and growing concerns about credential exposure, unintended agent actions, and the need for accountable automation. It emphasizes that successful deployments require governance measures including tool inventories, defined ownership, permission controls, versioning, regular auditing, secure authentication, credential vaults, API gateways, and approval workflows. MCP is framed as a core architectural layer for controlled AI autonomy, while CData Connect AI is promoted as a managed MCP platform that provides governed access to more than 350 enterprise data sources and integrates with major AI platforms without requiring custom code.

CData undertook a comprehensive rebrand ahead of 2026 to better reflect its evolution from a data connectivity tools provider into enterprise infrastructure for AI and live data, serving more than 10,000 customers. The company concluded that its early-2010s identity had become difficult to read, visually dated, and misaligned with its role in enabling governed, contextual, and reliable enterprise AI. Working with Harmonic Message on positioning and Focus Lab on identity, CData developed a strategy centered on “Connectivity, Context, and Control,” framing the core barriers to effective enterprise AI. The new system introduces a clearer uppercase wordmark, a black C icon on yellow, a navy, yellow, black, gray, and white palette, and typography combining Grafier, DM Sans, and DM Mono. Its messaging now emphasizes trust and accuracy through phrases such as “Built for data that has to be right” and “CData is the foundation for confident enterprise AI.” The public launch at the Gartner Data & Analytics conference includes major advertising in Orlando and will be evaluated through AI-related visibility, engagement, pipeline, recognition, event results, and whether target audiences quickly understand CData’s purpose.

CData announced a rebrand and major enhancements to its Connect AI platform, positioning itself as a governed data layer designed to help enterprises move AI agents from pilots into production. The company argues that inadequate data infrastructure is a primary barrier to AI adoption, citing its report that only 6% of organizations consider themselves fully prepared and that 71% of AI teams spend more than a quarter of implementation time on data integration. Connect AI now offers live read-write connectivity to more than 350 business systems, including on-premises sources through Connect Gateway, alongside layered universal, source-specific, and custom tools intended to provide relevant context without excessive access. The platform also adds identity and governance features such as per-user authentication, native permissions, audit trails, SCIM 2.0, and custom OAuth applications. CData says testing across 378 enterprise queries found its platform achieved 98.5% accuracy, attributing the result to a relational abstraction layer and source-level semantic intelligence rather than direct natural-language-to-API translation, and it will present its approach at the Gartner Data & Analytics Summit.

As AI agents become more common in enterprise systems, the passage argues that traditional iPaaS platforms remain valuable for deterministic, predefined integrations but may be insufficient for agents that need to explore and synthesize data across changing sources. It proposes “universal connectivity” as a complementary architecture that provides governed, live relational access to data through standardized interfaces such as MCP and SQL, allowing agents to discover schemas, use shared business semantics, query across systems, and perform reliable bidirectional write-backs. Unlike curated iPaaS endpoints and workflows, this model aims to let agents investigate unanticipated questions while maintaining platform-level controls such as permissions, rate limits, and audit trails. The passage concludes that organizations will likely need a mix of both approaches, using iPaaS for repeatable automation and universal connectivity for AI workloads requiring broad-context reasoning, flexible data exploration, and deterministic execution.

Unified APIs offer a fast way to build integrations through category-specific connectors and shared data models, making them potentially suitable for MVPs and standard use cases in areas such as CRM, HR, and applicant tracking. The text argues that their common models may expose only a subset of source-system fields, limit custom objects and flexible querying, rely on scheduled or read-oriented data synchronization, and introduce proprietary schemas, third-party data caching, and restricted connector coverage. It presents these limitations as especially consequential for enterprise applications and AI agents that require complete context, precise queries, real-time bidirectional write-back, compliance controls, and connections to specialized or legacy systems. In contrast, CData is positioned as providing direct, in-place access to full source data models through standard relational interfaces, event-driven bidirectional connectivity, broad connector coverage, and tenant-isolated security, while the recommended choice depends on whether a team prioritizes rapid standard integration delivery or long-term enterprise-scale flexibility.

Effective Salesforce-to-Snowflake pipelines require early definition of business objectives, required data, freshness targets, compliance obligations, and service-level expectations to avoid unnecessary cost and complexity. The recommended approach generally favors ELT, loading raw Salesforce data into Snowflake before transforming it with SQL-based tools such as dbt, while using batch loads for latency-tolerant use cases or incremental and change-data-capture methods for efficient, near-real-time replication. Reliable implementations use staging tables, source validation, stable primary keys, schema-evolution handling, standardized modeling practices, historical tracking where needed, and automated tests for row counts, null values, schema drift, and referential integrity. Monitoring latency, failures, throughput, and replay capability helps teams identify and recover from pipeline problems, while reverse ETL can return enriched lead scores, churn signals, and recommendations from Snowflake to Salesforce for operational use. The guidance also emphasizes encryption, role-based access, masking of sensitive data, compliance documentation, API-limit management, workload-specific Snowflake compute, incremental loading, and sandbox testing; it presents CData Sync as a low-code product for supporting these integration and reverse-ETL workflows.

Enterprise AI agents can search data, automate workflows, and act across business systems, but their autonomous behavior, machine-scale operation, and integrations with SaaS applications expand security and compliance risks. Effective governance treats agents as managed non-human identities with centralized registration, accountable ownership, secure authentication, short-lived credentials, least-privilege access, and dynamic authorization based on roles, attributes, policies, task context, data sensitivity, and delegation. The recommended framework also uses policy-as-code, enforcement gateways or AI firewalls to inspect tool calls, comprehensive audit logging and SIEM-integrated monitoring to identify anomalies, and human approval for high-risk or irreversible actions such as financial transfers, production changes, and bulk deletions. Implementation involves mapping agents, tools, data flows, and delegation relationships; defining and testing granular policies; centralizing enforcement; continuously monitoring activity; conducting red-team exercises for threats such as prompt injection and privilege escalation; and regularly updating credentials and permissions. While stronger controls can add complexity and latency, organizations can balance security and usability through risk-based agent tiers, distributed enforcement, centralized policy decisions, and zero-trust practices, with emerging approaches emphasizing Model Context Protocol governance, intent-based policies, and automated escalation workflows.