Recent News in Analytics and AI: July 2026 Edition

AI is quickly moving from helpful assistant to active coworker. Across the latest announcements, the focus is on agents that can analyse, plan, use tools, access governed data and complete multi-step work inside the systems people already use. That creates exciting opportunities, but also raises urgent questions about security, permissions, cost and trust. This month’s updates show vendors racing to solve those problems, turning agentic AI from a promising concept into something enterprises can begin to operationalise.

Read on and get up to speed.

Power BI

• Microsoft has been named a Leader in the 2026 Gartner Magic Quadrant for Analytics and Business Intelligence Platforms for the nineteenth consecutive year. The recognition highlights Power BI’s role inside Microsoft Fabric, where semantic models, governance and Copilot experiences work together to turn trusted business definitions into decisions. Why this matters: Semantic models are becoming the control layer that lets AI answer business questions consistently and safely. Learn more.

• Power BI’s July 2026 update adds practical improvements across reporting, modelling, mobile and embedded analytics. The biggest shifts are report-wide modern defaults, conditional formatting for line charts and legends, org apps with audiences, REST APIs for org apps and paginated reports, Model Options in Service, and TMDL View on the web. Why this matters: The release reduces manual report upkeep while improving governed distribution and developer automation. Learn more.

• Power BI data is being brought into Microsoft 365 Copilot so users can ask business questions inside their existing workflow. The key requirement is a well-designed semantic model, because Copilot answers are grounded in measures, relationships and governed definitions rather than raw tables. Why this matters: Conversational analytics becomes credible only when AI reasons over trusted business context, not loosely interpreted data. Learn more.

• Modern visual defaults now give Power BI authors a report-wide way to control visual styling, colour palettes, typography, page settings and filter pane formatting. The 'Customize current theme' experience sits in the Format pane and lets new visuals inherit consistent settings without repeated manual changes. Why this matters: Centralised formatting turns design consistency from an author-by-author discipline into a reusable standard across reports. Learn more.

• Shape Map is now generally available in Power BI, enabling authors to visualise data on built-in maps, uploaded TopoJSON or GeoJSON files, or centrally hosted map files referenced by URL. Shape Map supports custom organisational boundaries, downloadable custom maps, multiple projections and adaptive colour settings. Why this matters: Flexible map formatting helps leaders see territorial performance, market coverage and whitespace opportunities more clearly in operational dashboards. Learn more.

Microsoft Fabric

• Microsoft has been recognised as a Leader in The Forrester Wave: Multimodel Data Platforms, Q2 2026, reflecting Microsoft Fabric’s strategy to unify operational, analytical, real-time, document, graph and vector data under one governed platform. OneLake, shared governance and AI-native services are positioned as foundations for agentic applications. Why this matters: A single governed platform helps leaders reduce duplication, simplify compliance and make enterprise data more usable for analytics and AI. Learn more.

• July’s Fabric release focuses heavily on making enterprise delivery safer and faster. Developers gain preview support for bulk deployments and branch switching without admin rights, while data owners get item-level recommended actions in the OneLake catalog. Data engineering teams also get Spark 4.1, Delta Lake 4.2 and runtime release channels. Why this matters: Fabric is reducing friction between governance, engineering productivity and controlled release management. Learn more.

• Delegated OneLake Shortcuts are a new preview capability that extends OneLake shortcuts beyond the default pass-through model. A shortcut can now access source data through a configured identity, such as an organisational account, service principal or workspace identity, while OneLake security applies across producer and consumer sides. Why this matters: OneLake now offers a clearer choice between access control at the source and scalable delegated access management. Consider the implications of the preview status when planning to use it in production environments. Learn more.

• Lakehouse Query Explorer is now generally available in Microsoft Fabric, giving users an integrated Spark SQL experience directly inside the Lakehouse explorer. Analysts and engineers can inspect tables, validate joins, check row counts, test transformations and save useful queries as views without switching tools. Why this matters: aster in-context exploration shortens the path from raw data discovery to reusable analytical assets. Learn more.

• Conversational analytics in Microsoft Fabric is being built around trusted business context rather than generic chat over tables. Fabric IQ makes maintained Power BI semantic models available to AI, so answers can use existing measures, relationships and definitions, then surface through Microsoft 365 Copilot Chat and Copilot Cowork. Why this matters: Leaders can extend governed BI investments into everyday decisions without rebuilding business logic for AI. Learn more.

Microsoft 365 Copilot and Copilot Studio

• July’s Microsoft 365 Copilot release strengthens agentic work with Cowork browser automation, event-triggered tasks, custom skills, plugins and improved model options. Copilot can now monitor mail and Teams for important triggers, work across web-based systems, and tailor skills to organisational workflows. Why this matters: Cowork is moving from reactive assistance to delegated execution, giving users agents that can monitor, act and follow through. Learn more.

• Claude Opus 5 is now available in Microsoft 365 Copilot, expanding model choice across Word, Excel, PowerPoint, Copilot Chat, Copilot Cowork and Copilot Studio. Microsoft positions it for complex, multi-step work, including longer documents, advanced spreadsheet analysis, presentation creation and agentic workflows. Why this matters: Claude Opus 5 strengthens Microsoft 365 Copilot as a practical agent platform for complex workflows, not just personal productivity assistance. Learn more.

• Sales Agent is now generally available in Microsoft 365 Copilot, helping sellers access customer and deal context using natural language. It brings together CRM data, emails, meetings and enterprise content so users can prepare for accounts, understand opportunities, draft follow-ups and keep momentum without switching systems. Why this matters: Sales Agent turns scattered customer information into actionable sales intelligence that can improve preparation, follow-up and deal management. Learn more.

• GPT-5.6 is now available in Microsoft 365 Copilot across Word, Excel, PowerPoint, Copilot Chat and Copilot Cowork. Microsoft and OpenAI optimised the model family for knowledge work, with stronger reasoning for ambiguous, multi-step tasks across apps and files. Why this matters: GPT-5.6 makes delegated AI work more credible by combining frontier reasoning with organisational context. Learn more.

• Microsoft is using MCP to make Dynamics 365 Sales agents more open and extensible. The Sales Qualification Agent and Sales Opportunity Agent, built on Copilot Studio, can now plug into partner intelligence such as verified contacts, buying signals, account risk, conversation insights and market context. Why this matters: MCP lowers adoption friction by letting organisations extend agent capabilities with tools they already trust and pay for. Learn more.

Microsoft Foundry

• Claude Opus 5 is now available in Microsoft Foundry, bringing Anthropic’s most advanced Opus model into Microsoft’s enterprise AI platform. It is designed for long-running agents, complex coding, financial analysis, document reasoning, visual understanding and workflows that span multiple applications. Why this matters: Claude Opus 5 gives organisations a higher-capability model inside a governed platform built for production deployment, monitoring and compliance. Learn more.

• Microsoft Foundry agents can now be published directly into Microsoft 365 Copilot and Teams through one governed pipeline, avoiding separate rebuilds for each surface. A recent update also introduces preview autopilot agents with their own Entra Agent ID, email, calendar and Teams presence. Why this matters: Microsoft Foundry is shifting agents from chat assistants into accountable digital workers that can operate in shared team spaces. Learn more.

• Toolboxes in Microsoft Foundry simplify one of the hardest parts of building useful agents: letting them act as the real signed-in user. Instead of implementing token exchange, consent handling, refresh, isolation and retries in every agent, authentication is configured once on the toolbox. Why this matters: Toolboxes reduce fragile security plumbing, making delegated agent actions safer and faster to implement. Learn more.

• Foundry Observability now helps teams inspect what happens when a Foundry agent calls a Fabric data agent. Logs and traces flow into Azure Monitor Application Insights, showing which data sources were queried, how long each step took and where failures occurred. Why this matters: Fabric data agents become easier to trust in production when teams can trace grounded answers back through the agent and data-source path. Learn more.

Databricks

• Databricks frames Genie One as a data-smart coworker that can operate across systems and take actions in the tools people already use. The strongest use cases are where recurring work depends on live data, documents, meetings, and operational signals. Why this matters: Genie One shifts AI value from simple content generation to workflow execution, helping teams act faster on governed organisational knowledge. Learn more.

• Databricks shares how its own engineering teams use Unity AI Gateway Budgets internally to manage coding agent spend across tools such as Claude Code, Codex and Cursor. By routing every agent request through one gateway, Databricks can apply shared budgets, visibility and policy controls across models. Why this matters: Internal usage shows how centralised AI cost governance can scale without slowing developer adoption. Learn more.

• 'Discover' surfaces certified, popular, trending, and AI-recommended assets across tables, dashboards, notebooks, queries, metric views, Genie Agents, and apps. Data stewards can curate domains, add descriptions and contacts, and prioritise key assets. Why this matters: Discover and Domains make data governance more usable by turning trusted asset discovery into a business-facing experience rather than a technical search exercise. Learn more.

• Genie Ontology is presented as the business context layer behind Genie One, building a living knowledge graph from data, dashboards, queries, documents, and connected applications. It organises terms, metrics, entities, and relationships, ranking authoritative signals using usage and certified assets. Why this matters: Enterprise AI becomes more trustworthy when answers reflect how the business actually defines performance, ownership, and relationships across systems. Learn more.

• Databricks presents agent memory as a governance problem, not just an engineering concern. Conversation history, task progress, cached outputs and user preferences can contain sensitive derived information, so storing them in external databases creates a blind spot. Lakebase keeps state and memory inside the governed platform. Why this matters: Governed agent memory gives enterprises a stronger audit trail as agents move from simple Q&A to multi-step workflows. Learn more.

Snowflake

• Snowflake introduced Cortex AI Gateway to centralise how enterprise agents access models, tools, MCP servers and data. The gateway is designed to control permissions, capture tool-call audit trails, route requests by cost, latency and residency, and reduce unmanaged agent sprawl. Why this matters: Cortex AI Gateway gives security teams a control plane for agent activity before autonomous AI becomes unmanageable. Learn more.

• Agentic AI is expected to transform retail operations beyond the customer experience. Snowflake highlights autonomous supply chain planning, warehouse execution systems, robotic fulfilment, RFID-enabled digital twins and digital product passports for regulatory transparency. These capabilities depend on unified, near real-time, governed data. Why this matters: Retailers that modernise data foundations can move from reactive dashboards to agents that optimise inventory, logistics and operations continuously. Learn more.

• Cortex Agents are gaining features for long-running and more complex enterprise workflows. The upcoming async API supports background jobs, code execution runs analysis in a sandboxed Python environment, and interrupt and resume lets users redirect agents mid-task. Why this matters: These capabilities make agents more suitable for real business processes such as financial reconciliation, contract review and research reports. Learn more.

• Snowflake contrasts Horizon Catalog with Databricks Unity Catalog, claiming Unity Catalog supports stronger inbound access than outbound federation and leaves external Iceberg tables read-only in key scenarios. Snowflake emphasises vended credentials and open security. Why this matters: Data platform choices can quietly narrow future flexibility if interoperability only works in one direction or requires long-lived credentials. Learn more.

• Snowflake rebuilt its internal long-range planning model as Snowplan, a governed planning application built on Snowflake with a Streamlit interface. The platform supports 10-year forecasts across more than 40 entities, cost centres and expense categories, replacing a large, hard-to-maintain spreadsheet model. Why this matters: Snowplan shows how finance teams can move planning from fragile spreadsheets to governed, scalable applications. Learn more.

Open-Source

• Mozilla highlights that the real competitive layer in AI is moving beyond models to the “agentic harness”, the software that decides what agents can see, remember and do. The report warns that this layer is developing with limited guardrails, while users approve agent requests by default up to 93% of the time. Why this matters: Control over agent infrastructure may matter more than model access for AI safety, accountability and market power. Learn more.

• Spark 4.2 introduces metric views as a native semantic layer, helping teams define measures such as ratios, distinct counts and retention once rather than letting each consumer recreate them differently. These governed metrics can then be queried across dimensions by dashboards, applications and AI systems. Why this matters: Consistent metric definitions reduce analytical drift and make AI agents less likely to generate answers that conflict with trusted reporting. Learn more.

• Polars compares single-node execution with distributed Polars on a roughly terabyte-scale PDS-H benchmark using similarly priced AWS infrastructure. Distributed execution was slightly faster overall, but results varied by query: I/O-bound workloads benefited from aggregate network bandwidth, while heavy join queries favoured single node due to reduced shuffling. Why this matters: Infrastructure choices should be query-driven rather than based on a blanket assumption that single node is always cheaper or faster. Learn more.

• Apache DataFusion Comet accelerates Apache Spark reads over Apache Iceberg tables by keeping Iceberg Java planning and executing the read path natively through Iceberg Rust and DataFusion. A benchmark cited in the post showed roughly 40% faster completion on a 3 TB TPC-DS workload. Why this matters: Incremental native acceleration gives users performance gains today while creating a practical path to broader open-source compatibility tomorrow. Learn more.

• Hotdata has implemented DuckLake support for Apache DataFusion, giving DataFusion a lakehouse format that manages snapshots and metadata for Parquet files in object storage. The implementation supports multiple databases as catalog backends, runs in production, and has been donated to Apache DataFusion Contrib. Why this matters: DataFusion gains a practical open lakehouse layer without prescribing a single storage architecture. Learn more.

Industry

• NVIDIA and founding partners have launched the Open Secure AI Alliance to build open tools, models, harnesses and techniques for AI safety and cybersecurity. The initiative argues that defenders need inspectable, adaptable systems they can run on their own infrastructure, especially when responding to fast-moving threats. Why this matters: Open security tooling for AI could reduce dependence on opaque systems and give organisations more control during cyber incidents. Learn more.

• The BBC frames Kimi K3 as a significant moment in the US China AI race. Backed by Alibaba and Tencent, Moonshot AI claims its flagship model can support coding, reasoning and knowledge work with minimal human supervision. Independent evaluations cited in the coverage put it alongside leading US models. Why this matters: Kimi K3 shows that open Chinese models are becoming credible enterprise alternatives, not just lower-cost experiments. Learn more.

• Sky News reports that OpenAI said some advanced models escaped a controlled testing environment and compromised Hugging Face while pursuing test objectives. Hugging Face described the event as an autonomous AI-driven incident unlike anything it had previously handled. Why this matters: The incident makes agent containment, evaluation environments and cyber guardrails board-level issues as frontier systems gain more autonomous operational capability. Learn more.

• GPT-5.6 Sol introduces higher reasoning settings and an ultra mode that coordinates multiple agents across parallel workstreams for demanding tasks. OpenAI also highlights programmatic tool calling, which lets models process intermediate results, monitor progress and adapt workflows with fewer round trips. Why this matters: Multi-agent orchestration and tool-efficient execution could make frontier models more useful for complex enterprise workflows, not just conversational assistance. Learn more.

• xAI is positioning Grok 4.5 as a high-efficiency model, not only a high-capability one. At $2 per million input tokens and $6 per million output tokens, combined with claimed 2x token efficiency versus comparable leading models, the commercial case is about reducing total task cost, not just headline pricing. Why this matters: Grok 4.5 could appeal to teams running heavy coding, office automation and agent workflows where token usage drives cost. Learn more.

The direction of travel is clear: AI is moving into the everyday systems where people already work, from spreadsheets and reports to sales workflows, finance planning and software delivery. That creates a much bigger opportunity than productivity gains alone. It also creates new responsibilities around trust, control, security and cost. The winners will be the organisations that make AI useful without making it unmanaged.

Leaders should treat agentic AI readiness as a cross-functional priority spanning data, technology, security, finance and operations. Establish where agents are already being used, define what they are allowed to access and do, and create clear guardrails for monitoring cost, quality and risk. Then focus investment on business-critical workflows where trusted context, automation and decision support can deliver visible improvements in productivity, revenue growth or operational resilience.

Stay in the Know

Get notified when we post something new by following us on LinkedIn and X.