September 2026 Summaries

Sigma’s data platform team manages its semantic layer as code using dbt and deploys it through Dagster alongside warehouse-table changes, aiming to ensure that dashboards, analysts, and AI systems use consistent definitions for metrics such as ARR. The approach addresses governance, lineage, version control, reviewability, and AI reliability by storing business concepts, relationships, metrics, access restrictions, synonyms, and SQL guidance in domain-specific semantic models rather than broad, manually maintained interfaces. Sigma organizes these objects in a dedicated semantic dbt layer and uses open-source packages to represent Sigma data models, Snowflake Semantic Views, and Cortex Agents as nodes in the dbt dependency graph. Dagster then runs incremental dbt builds after merges, syncs changed warehouse connections and data-model definitions through Sigma’s API, and stores updated manifests, allowing tables and semantic models to be deployed together. These governed models support Sigma Agents, Sigma Assistant, Snowflake Cortex Agents, search services, and coding agents, enabling natural-language queries and automated workflows to trace answers back to shared business definitions.

Sigma announced the general availability of Sigma Agents, an enterprise AI workflow feature that operates on live warehouse data, can write results back to warehouses, and can trigger actions in systems such as Slack, Jira, and Salesforce. Organizations can build agents in plain language, connect them to governed data sources and warehouse agents including Snowflake Cortex and Databricks Genie, and enforce existing row-level, column-level, document, and tool-access permissions. Agents can run within Sigma workbooks and apps, on schedules, through webhooks, via the Sigma REST API, or from AI assistants such as Claude, ChatGPT, Cursor, and Codex through Sigma’s MCP server. The release adds chat history for users and centralized administration tools that show ownership, data sources, token use, execution logs, errors, and activity across all agents, addressing governance, auditing, and cost-management concerns. Sigma reports that more than 1,200 organizations used the beta to create nearly 6,400 production agents handling over 580,000 conversations, while noting that the feature is premium paid and that future updates will expand standalone agent creation and management controls.

AI agents for marketing analytics are designed to move beyond dashboards and chatbots by monitoring live data across advertising platforms, CRM systems, and data warehouses, then taking authorized actions such as reallocating budgets or pausing campaigns within predefined limits. Unlike channel-specific automation, these agents can assess cross-channel performance alongside budget pacing, historical benchmarks, and funnel context, while escalating higher-risk decisions to human approvers. Effective deployment depends on selecting low-risk, measurable workflows; using governed and standardized data; testing through historical replay and shadow mode; enforcing least-privilege access, spending thresholds, and audit logs; and expanding autonomy only after regular review. The text positions Sigma as a platform that supports such agents through governed warehouse data, shared semantic metrics, row-level security, configurable actions and alerts, approval routing, and traceable writeback capabilities.

AI agents for sales and revenue operations are configured systems that monitor live CRM, pipeline, quota, and warehouse data, interpret defined signals, and take limited workflow actions such as flagging risks, updating forecast scenarios, routing approvals, or notifying stakeholders. Their typical operating cycle is to read a trigger, evaluate it in context, execute a permitted action, and log the result, with potential applications in pipeline and forecast risk detection, deal desk approvals, commission disputes, and renewal or churn management. Effective implementation begins with a narrowly defined workflow, explicit triggers and success measures, governed access to relevant data, task-specific permissions, named ownership, human escalation paths, and detailed audit records for every decision and write. The text also presents Sigma as a warehouse-native platform that supports these agents through inherited warehouse permissions, workbook-based configuration, approval and notification actions, controlled writeback through Input Tables, and audit, lineage, and change-management capabilities.

Jev is a text-classification AI model from TypeSafe AI that returns structured labels, probabilities, or category selections rather than generating prose, positioning it as a lower-cost alternative to general-purpose LLMs for large-scale text analysis. Sigma’s public-beta AI columns can use Jev through MotherDuck’s `prompt_jev()` function to classify workbook data row by row, enabling users to filter and analyze results as standard data columns. In a reported test, tagging 6,578 customer reviews by referenced product features took 4.6 seconds and cost 5.3 cents, while projected comparisons suggest substantially lower costs than Claude Haiku for similar workloads. The approach is intended to let business users experiment with classifications without extensive data-engineering work, while Sigma provides caching, access controls, and monthly token caps. However, Jev cannot summarize or infer context absent from an individual record, its accuracy should be validated against trusted labels, and text submitted through MotherDuck is sent to TypeSafe for processing, requiring security review for sensitive data.

Operational reporting provides frontline teams and managers with timely, granular visibility into daily activities such as shipments, support tickets, inventory, financial transactions, and patient flow so they can identify exceptions and act within an appropriate operational timeframe. It differs from strategic reporting by focusing on current execution rather than long-term business direction, typically refreshing hourly to weekly instead of monthly or quarterly. Common forms include exception reports that flag threshold breaches, status reports that show workflow progress and capacity, and performance reports that measure results against targets or service-level agreements. Across retail, logistics, finance, and healthcare, these reports can support decisions such as replenishing stock, addressing shipment delays, investigating suspicious payments, and allocating hospital beds. Effective reports begin with a clearly defined decision, use documented and traceable data logic, match refresh schedules to required response times, and are monitored for adoption, freshness, and user trust. The text emphasizes that stale exports, changing source schemas, spreadsheet errors, inconsistent metric definitions, and unclear ownership can undermine reporting accuracy, advocating live warehouse queries, governed metrics, data lineage, assigned owners, automated alerts, and writeback capabilities. It presents Sigma as a warehouse-native analytics platform intended to support these practices by querying live cloud data, sending threshold-based notifications, and recording operational resolutions with audit trails.

Data apps are interactive software tools that connect directly to governed data sources, enabling users to analyze live information and take actions such as approving records, updating forecasts, or triggering workflows without relying on exported spreadsheets and email handoffs. Unlike traditional dashboards, which primarily support viewing and exploration, data apps include controlled writeback, permissions, and audit trails that keep decisions within the underlying system of record. They commonly consist of an interface for displaying data and collecting structured inputs, a data layer that reads from and writes to live governed sources, and a logic layer that validates actions, applies rules, and initiates subsequent processes. Common applications include approval workflows, financial and demand planning, inventory management, and account operations, where routine work can be automated and exceptions routed to people. The category is evolving toward AI Apps, in which AI can help build interfaces and workflows through natural-language instructions and operate defined tasks under scoped permissions, with emerging standards supporting agent access to enterprise tools. Sigma presents its platform as an example of this approach, offering warehouse-connected workbooks, writeback, workflow actions, audit logging, and AI-assisted building and operational agents.

Data reporting converts raw data into structured, repeatable records of business activity, primarily answering what happened, while data analysis investigates why it happened and what actions may follow. Organizations use operational, strategic, regulatory, and management reports with varying audiences, time periods, delivery formats, freshness requirements, and archival needs. Reports differ from dashboards, which continuously monitor current conditions, and from visualizations, which are individual charts or graphs used within both. At enterprise scale, reporting becomes more difficult because data is fragmented across systems, metric definitions conflict, teams create proliferating report versions, and manual spreadsheet consolidation increases error risk. Effective reporting programs therefore emphasize direct or live warehouse connections, self-service tools that reduce engineering request queues, and governance that standardizes metric definitions, security, and data lineage. The text presents Sigma as a warehouse-native analytics platform that supports live queries, spreadsheet-like report building, inherited security controls, and pixel-perfect exports for audit-oriented reporting.

Headcount planning software helps HR and finance teams forecast workforce needs, model hiring and attrition scenarios, calculate fully loaded compensation costs, and manage organizational approvals against budgets. The article argues that traditional systems often rely on periodic HR and finance data exports or spreadsheets, causing plans, forecasts, and approvals to become outdated as employee rosters and business conditions change. It describes AI agents as configurable workflows that use live data to continuously reforecast, identify budget or attrition risks, and route approvals based on defined thresholds. Effective use of these agents requires governed least-privilege access to sensitive compensation data, clear limits separating recommendations from actions, and complete audit logs. Sigma is presented as a warehouse-native platform that connects planning workbooks, agents, and governed HR and finance data in systems such as Snowflake, BigQuery, Databricks, and Redshift, allowing teams to maintain live plans while preserving security, approval controls, and traceability.

Cloud business intelligence solutions are hosted platforms that connect querying, dashboards, reporting, semantic metrics, and governance capabilities to cloud data warehouses such as Snowflake, Databricks, BigQuery, and Amazon Redshift. Compared with on-premises BI, they shift infrastructure maintenance, scaling, patching, and disaster recovery responsibilities to vendors, while organizations retain responsibility for data, metrics, access controls, and release testing. The guide argues that buyers should prioritize platforms that execute generated SQL directly in the warehouse, inherit existing row- and column-level permissions, maintain a single source of metric definitions, isolate workloads effectively, and provide traceability and access controls for AI-generated answers. It also recommends evaluating total cost, licensing, compliance, integrations, support, migration effort, and proof-of-concept measures such as query latency, concurrent load, permission-update speed, and the time required for nontechnical analysts to produce useful analyses. Sigma is presented as an example of a warehouse-native platform that compiles spreadsheet-like formulas and analysis tools into warehouse SQL, uses live governed data, and provides AI capabilities under existing warehouse security policies.

Demand forecasting informs inventory, procurement, production, and financial planning by combining internal sales and supply data, external market signals, and human commercial judgment, but traditional monthly cycles can become outdated quickly amid disruptions such as supplier delays, tariffs, promotions, and competitor pricing changes. AI agents are presented as a way to make forecasting a continuous, governed process by monitoring live warehouse data, detecting meaningful changes, running forecasting models, updating plans within predefined limits, and escalating larger exceptions for human approval. Effective implementation requires reliable and consistently defined data, restricted access and action scopes, transparent audit trails, planner override capabilities, and direct writeback to governed planning tables. The text distinguishes agents from forecasting models by emphasizing that agents manage the broader workflow, including triggers, interpretation, actions, and accountability. It also describes Sigma as a platform that connects such agents to live cloud warehouse data while preserving existing permissions, security controls, data lineage, and auditable writeback processes.

Sigma has upgraded its Sigma Assistant, a natural-language interface for exploring governed warehouse data, building dashboards and applications, and analyzing insights using organizational data models and certified metrics. The generally available update improves handling of open-ended questions by asking clarifying questions, disclosing assumptions and data caveats, and providing faster follow-up responses. Administrators can allow the Assistant to search all accessible data sources or a curated set while preserving existing row-level security, column masking, connection permissions, and audit controls. Each response exposes its sources, analytical steps, generated SQL, and underlying tables, and users can continue investigations through editable Sigma workbooks. Chat history, which requires administrator configuration, preserves past conversations, enables exports, and allows insights to be promoted into reusable workbooks. Sigma is also previewing an AI-first homepage in private beta, while future plans include helping users find Sigma content, learn navigation, and schedule data-driven workflows.

Sigma engineers used an agentic workflow to migrate more than 1,000 end-to-end tests from a legacy framework to Playwright in four weeks without a dedicated quality engineering team. The effort began by establishing reusable Playwright fixtures, page objects, authentication patterns, and parallelization practices, then encoding these standards into separate AI migration and review skills. For each test, agents translated the existing specification, ran it locally, submitted it to an independent adversarial review, and repeatedly executed it in CI with retries disabled to identify and correct root causes of instability before removing the original test. Sigma dashboards, Workbooks, Actions, and API connections tracked migration progress and enabled teams to trigger agent workflows through Cursor directly from the platform. The completed migration reportedly reduced test flakiness by 90%, saved 6,000 CI minutes weekly, improved debugging and feedback speed, and helped expose a previously overlooked intermittent production bug. Sigma plans to extend the approach toward automated acceptance-criteria generation, investigation, debugging, and self-healing test workflows.

Sigma has launched an updated plugin for ChatGPT Work and Codex, integrated with OpenAI’s Data agent, that lets enterprise users query governed company data and create, refine, filter, and publish interactive Sigma dashboards through natural-language prompts. Available to OpenAI enterprise customers with eligible Sigma licenses, the plugin retains Sigma’s existing permissions, metric definitions, live-data connections, and governance controls, avoiding separate extracts or security models. Users can ask business questions, receive answers based on trusted Sigma workbooks and metrics, customize dashboard layouts and visualizations conversationally, add repeat-use filters, and publish the resulting dashboard as a full Sigma workbook for continued editing and sharing. Query capabilities are available to all Sigma users with relevant access, while dashboard creation requires a Build license, and activation is available through the ChatGPT Work Data agent, ChatGPT’s plugin directory, or Codex.

Sigma has made its workbook-to-report converter generally available, allowing customers to turn existing live-data workbooks into pixel-perfect reports with a single click. Designed to reduce the manual effort of recreating analyses in formatted documents, the feature supports use cases such as recurring financial reports, retail PDFs, and invoices while preserving live warehouse data at the time of export. Users can select workbook pages, page size, and orientation before conversion, then edit, export, or schedule the resulting report, while an API endpoint enables the same process programmatically. Converted reports are separate documents with independent permissions and do not automatically remain synchronized with their source workbooks; users can re-convert as needed, though some layouts may require adjustment because of page-size or element-rendering differences.

Custom AI agents for data analytics are presented as a way to make AI-driven analysis more reliable, repeatable, and governed than general-purpose assistants, which can produce inconsistent results, struggle with complex enterprise warehouses, and require users to expose schema or sensitive business context outside established security controls. Effective analytics agents require carefully scoped access to governed data sources, semantic definitions and metric context, orchestration for multi-step querying and error correction, explicit limits on read and write actions, human approvals for consequential changes, output validation, and detailed audit logs. The article argues that agents should operate under existing warehouse permissions, use narrowly defined tools, apply retry limits and checkpoints, and validate SQL and results before business users rely on them. It describes Sigma as a platform intended to provide much of this infrastructure through warehouse-native governance, workbook-based workflow configuration, scoped actions, validation, and auditable writebacks to Input Tables, while supporting data warehouses such as Snowflake, Databricks, BigQuery, and Redshift and configurable AI model providers.

Sigma has launched a public beta feature that lets users export reports directly to formatted PowerPoint (.pptx) files in one click, preserving the report’s layout and using current data from the connected cloud data warehouse. Designed for sales, operations, and finance teams that regularly produce recurring presentation decks, the feature reduces manual tasks such as screenshotting charts and maintaining separate slide templates. Each export reflects live warehouse data at the time of download and retains the report’s existing permissions and row-level security, allowing organizations to distribute appropriately scoped versions to recipients. Sigma positions the capability as a replacement for custom export workarounds and separate deck-maintenance processes, with additional export formats and layout controls planned for the future.