September 2026 Summaries

Agentic AI is expected to reduce the labor required for many consulting deliverables, challenging traditional effort-based pricing while creating an estimated $200 billion market for services focused on AI outcomes, with many enterprises likely to rely on external providers to operate their priority use cases. The central constraint is presented as business context rather than model intelligence, since unreliable data, conflicting definitions, inadequate permissions, and weak governance have contributed to failed AI deployments and remain major obstacles to adoption. The proposed solution is Open Data Infrastructure, built on open table formats, shared metadata, lineage, semantic layers, and governance, which would allow organizations to use multiple data platforms and AI models without locking business logic into one vendor’s stack. Consulting firms are encouraged to shift from selling staff time toward designing governed AI workflows, managing AI costs, assessing client maturity, and creating reusable intellectual property such as reference architectures, agent patterns, and portable operational knowledge.

A benchmark comparing five dbt-based interfaces for AI agents found that governed access to data can affect analytical accuracy, trust, and cost as much as model selection. Using retail datasets, Claude Haiku and Opus answered join and causal “why” questions through raw SQL, optional or required Agents Schema documentation, manifest tools, and the dbt Semantic Layer. Required documentation eliminated Haiku’s repeated grain errors by ensuring it read column definitions, while the Semantic Layer resolved Opus’s inconsistent interpretation of whether canceled orders count toward revenue by enforcing a single metric definition; it was the only interface where both models achieved zero wrong answers on join questions and also produced the lowest cost per trusted answer. However, the Semantic Layer did not correct Haiku’s reasoning errors on causal questions, although it substantially reduced query turns and expense. The results also indicated that unrestricted documentation can cause stronger models to search irrelevant sources or overinterpret questions, while scoped access and governed metrics improve reliability. The author concludes that dbt projects can provide a foundation for agentic analytics through documentation, semantic models, testing, runtime-accessible metadata, and question routing that uses lower-cost models for governed metrics and stronger models for more complex investigative tasks.

Fivetran has announced Context Layer, a managed service in Limited Public Preview designed to help enterprises build governed AI-agent context directly in their data warehouses or Fivetran Managed Data Lake Service. Addressing the challenge that many AI initiatives lack unified, accessible data, the offering integrates data and metadata already ingested by Fivetran and modeled with dbt, allowing agents to use consistent business definitions, governed datasets, fiscal calendars, and unstructured organizational knowledge. Its features include metadata connectors for tools such as dbt, Looker, Sigma, and Power BI; discovery of lineage and business ontologies; indexing of sources such as Confluence, Google Drive, Jira, and Zendesk; open SQL-based context tables; an MCP interface for serving context to agents; and warehouse-based traces and evaluations for reviewing agent activity. Fivetran says its internal use of the technology has reduced agent tool calls, response time, and input-token consumption, and it is seeking design partners among customers using Snowflake, BigQuery, or its managed data lake service.

APAC retail and consumer packaged goods teams increasingly rely on automated systems and AI for inventory allocation, reordering, and promotions, making trustworthy data essential as fragmented sources, legacy ERPs, spreadsheets, and regional systems often produce inconsistent information. While centralization brings data together, dbt is presented as the layer that makes it reliable through automated tests that detect missing, duplicate, outdated, or disconnected data; governed semantic definitions that ensure metrics such as days of supply and promotional lift remain consistent across dashboards, analysts, and AI agents; and end-to-end lineage that traces decisions back to original sources. The piece argues that ungoverned data can turn automation into costly overstock, stockouts, missed promotions, and public operational failures, whereas combining Fivetran for centralization with dbt for testing, governance, and traceability enables teams to activate data more confidently across the business.

Ecommerce analytics becomes more difficult as businesses grow because transactional, marketing, web, finance, CRM, and operational data are often separated across platforms, making attribution, customer lifecycle analysis, and consistent reporting reliant on manual spreadsheets or fragile pipelines. The proposed approach is to automatically centralize data from sources such as Shopify, Google Analytics, advertising platforms, finance systems, and CRMs in a cloud data platform, then analyze it through business intelligence tools. Combining these datasets can support revenue-based attribution, a shared source of truth for metrics, analysis of the full customer journey and retention, and customer segmentation methods such as RFM analysis. Examples involving Koh, Sleeping Duck, Papier, Pet Circle, Ritual, Carwow, and Westwing illustrate reported benefits including improved targeting, clearer revenue and margin reporting, stronger retention analysis, and reduced engineering maintenance. The piece recommends evaluating integration tools based on source coverage, maintenance requirements, data completeness and freshness, and scalability, positioning Fivetran as a platform designed to automate these data connections.

Open Data Infrastructure (ODI) is presented as an open, standards-based data architecture designed to reduce friction in analytics and AI workflows by giving organizations greater control, interoperability, and access to AI-ready data. It aims to counter vendor restrictions on data exports and APIs, allowing companies to access and reuse their own data without repeatedly seeking permission or rebuilding pipelines. By storing data once in low-cost object storage using open table formats such as Apache Iceberg or Delta Lake, ODI separates storage from compute, enabling teams to select appropriate tools for business intelligence, data science, operations, and AI without duplicating data. Its shared data layer can also reduce redundant pipelines, reconciliation efforts, and inconsistent business definitions by providing a common source of governed data, metadata, lineage, and metric definitions. The approach emphasizes that openness does not mean weaker security, as standardized access policies, classifications, and lineage can travel with data across systems, helping organizations maintain controls, simplify audits, and support reliable autonomous AI agents.