November 2026 Summaries

No summary generated yet.

Fivetran research based on a survey of 500 senior IT, data, and AI/ML decision-makers describes an “AI tax,” or the portion of AI spending lost to avoidable rework such as addressing data-quality problems, rebuilding integrations, reprocessing data, and retraining models. Thirty-nine percent of respondents said at least one-quarter of their AI/ML budgets went to such work, while only 16% kept it below 10%, despite most organizations investing at least $1 million annually in AI and relatively few operating multiple initiatives at scale. Data quality and integration were the most commonly cited sources of unexpected work, and poor data was associated with inaccurate outputs, model recalibration, and reduced stakeholder trust. Organizations with high AI taxes were substantially more likely to have AI projects delayed or stalled and reported weaker returns than lower-tax peers. The research associates standardized integration platforms, combined with limited custom development for specialized needs, with lower avoidable costs than primarily project-specific custom integrations. It recommends treating data integration as shared infrastructure, tracking all data-related AI costs centrally, and establishing clear ownership and cross-team alignment to direct more AI investment toward production and business value.

Fivetran’s founders argue that AI agents—systems combining intelligence with tools and business data—will transform knowledge work more profoundly than previous technology shifts, particularly in areas such as software development, analysis, and operational processes. They emphasize that agents require direct access to broad, reliable organizational data and clear definitions, logic, and evaluations to develop trustworthy models of how a business operates, rather than relying only on documents and conversations. The company describes a progression from AI augmenting human workers to autonomous agents managing end-to-end processes, such as proactively resolving customer support issues and initiating engineering fixes. It identifies “agent engineers” as professionals responsible for redesigning processes, connecting agents to systems and data, evaluating performance, and managing organizational change. Fivetran says it is building infrastructure for this agentic future around open data infrastructure principles, arguing that data ownership, governance, open standards, and portability are necessary to prevent dependence on a single vendor, while maintaining that human qualities such as empathy, innovation, and goal-setting will remain important.

Fivetran introduces a terminal-based CLI designed to let engineers and AI coding agents manage data pipelines through commands for connections, schemas, syncs, account resources, webhooks, logs, and custom connectors. The CLI is intended to work alongside dbt v2 for transformations and coding harnesses such as dbt Wizard, Claude Code, Codex, or Gemini CLI, while MCP servers supply context from task trackers and observability platforms. The proposed workflow enables an agent to read a ticket, configure a source such as Stripe, sync data to a destination like Snowflake, create and validate dbt models, open pull requests, update ticket status, and respond to operational failures. Fivetran emphasizes structured JSON output, command discoverability through help options, exit codes, configuration files stored in version control, and scoped API permissions to support automation, reviewability, and safer handling of destructive operations. Resource-management commands are currently in beta, while the CLI package and related Connector SDK support creating custom connectors when no prebuilt source integration is available.

Fivetran argues that the main challenge in enterprise AI is not model capability but giving models governed, accurate business context through a context layer that captures data semantics, relationships, rules, lineage, and metadata. It describes context engineering as a four-stage process of centralizing data from structured and unstructured sources, transforming it into trusted models, defining shared business meanings, and publishing that information through open schemas and Model Context Protocol tools for AI clients. This approach is intended to ground AI responses in current organizational data, reduce hallucinations, improve repeatability and governance, and support applications such as conversational and generative analytics, knowledge bases, and AI-assisted data engineering. The company contrasts this architecture with its earlier custom retrieval-augmented generation system, explaining that its expanded Fivetran AI platform now combines a context layer, warehouse querying, and cited responses; it reports that adoption of FivetranChat rose from roughly 10% to 52% of employees in one month during 2026, with especially strong usage in customer-facing teams.

Microsoft Fabric Community Event discussions in Barcelona emphasized building an open, AI-ready data foundation that combines unified analytics with portability and governance. Fivetran advocates storing data once in customer-controlled Azure Data Lake Storage using open Apache Iceberg or Delta formats, enabling Fabric and other compatible engines to access the same governed data without duplicate ingestion. dbt supports reusable SQL-based transformations, tests, metrics, and lineage across Fabric Warehouse and Lakehouse, while the generally available dbt Job in Fabric adds native scheduling, logging in OneLake, and deployment support. The dbt MCP Server extends governed business context to AI agents so they can use established definitions rather than interpret raw tables independently. French flooring manufacturer Gerflor illustrated the approach as it modernizes a fragmented on-premises platform serving 40 subsidiaries and over 400 applications, using Fabric, Fivetran, and dbt to automate data movement, standardize reporting, reduce manual consolidation work, and prepare for expanded analytics and AI use.

Fivetran conducted a blinded two-week workplace study in which 39 engineers compared three OpenAI coding models, Luna, Terra, and Sol, through Claude Code while model identities were masked. Although Sol, the most expensive model, was preferred in both head-to-head comparisons, the margins were small and statistically inconclusive, with nearly half of preferences described as weak. Costs differed sharply, with typical weekly spending rising from about $5 for Luna to $41 for Terra and $168 for Sol, while pull request activity remained relatively steady across tiers and did not establish a productivity ranking. The findings suggest that higher token prices do not reliably translate into better day-to-day coding experiences, as engineers valued factors such as responsiveness, planning quality, scope control, and tool compatibility differently depending on task complexity. Fivetran plans further blinded research that separates planning and execution roles across models and measures code quality, human effort, latency, completion, and cost under controlled conditions.

Fivetran announced that it was named a Leader in the 2026 Gartner Magic Quadrant for Data Integration Tools, marking its seventh consecutive appearance and reported improvement in both Ability to Execute and Completeness of Vision. Following its June 2026 merger with dbt Labs, the company presents a unified platform for data movement, transformation, activation, semantic definitions, lineage, and AI-oriented business context, supported by more than 750 managed connectors and a managed data lake offering. The company argues that reliable AI depends on trusted, current data, efficient reusable pipelines, and flexibility to adopt changing tools, and highlights capabilities such as dbt modeling and testing, the Context Layer, dbt State, AI-assisted model development, and automated connector creation. It also emphasizes investment in open-source projects and standards, including dbt v2, Great Expectations, Apache Iceberg, and Agents Schema, while maintaining partnerships across major cloud, warehouse, and application providers. Gartner’s designation reflects the research firm’s opinion and is not an endorsement of any vendor or product.