Keep up with data and AI.

Executive read

This week’s GitHub momentum is shifting away from novelty apps and toward tools for making AI agents, local inference, containers and cloud-native operations more governable.

The pattern to watch is operationalisation. Repos gaining attention are tackling security scanning for agent skills, reusable engineering playbooks for coding agents, token-cost reduction, codebase memory, LLM cache infrastructure and developer runtime portability.

Repo shortlist

NVIDIA/SkillSpector

• What it is: A Python security scanner for AI agent skills, detecting vulnerabilities, malicious patterns and security risks.
• Why it is gaining traction: Agent skills are becoming a software supply-chain surface; teams need inspection before they allow reusable agent capabilities into production workflows.
• Why it matters: Strong fit for AI governance, developer-platform security and agent rollout controls. Apache-2.0 licence; active recent pushes.
• Watch-out: Still a specialist control-plane component. It is most valuable when paired with broader policy, audit and approval workflows.

addyosmani/agent-skills

• What it is: Production-grade engineering skills for AI coding agents.
• Why it is gaining traction: Teams want reusable, opinionated workflows for coding agents rather than ad-hoc prompting.
• Why it matters: Useful as a template library for standardising agent-assisted engineering practices across teams. MIT licence and very high community momentum.
• Watch-out: Treat as patterns to adapt, not a turnkey operating standard; internal security and coding conventions still matter.

apple/container

• What it is: A Swift tool for creating and running Linux containers via lightweight virtual machines on Apple silicon Macs.
• Why it is gaining traction: Mac-based developer environments still need more reliable container workflows, especially as local AI/dev infrastructure gets heavier.
• Why it matters: Relevant for platform teams supporting Apple silicon fleets and secure local development environments. Apache-2.0 licence; highly active.
• Watch-out: Apple-silicon optimisation narrows the audience; test carefully against existing Docker/Colima/OrbStack workflows before standardising.

chopratejas/headroom

• What it is: A library/proxy/MCP server for compressing tool outputs, logs, files and RAG chunks before they reach an LLM.
• Why it is gaining traction: Token cost and context overflow are becoming practical blockers for agentic workflows.
• Why it matters: Strong fit for cost control, RAG efficiency and agent observability. Apache-2.0 licence; active recent commits.
• Watch-out: Compression can hide edge-case detail. Teams should evaluate answer quality, auditability and failure modes before deploying broadly.

DeusData/codebase-memory-mcp

• What it is: A code-intelligence MCP server that indexes codebases into a persistent knowledge graph for fast agent queries.
• Why it is gaining traction: Coding agents need persistent codebase context without repeatedly burning tokens on whole-repo scans.
• Why it matters: Useful for large-repo developer productivity, onboarding and agent-assisted maintenance. MIT licence.
• Watch-out: Security review matters: code indexers can expose sensitive structure, secrets-adjacent context and internal architecture.

LMCache/LMCache

• What it is: A KV-cache layer for accelerating and reducing the cost of LLM serving.
• Why it is gaining traction: Inference efficiency is now a board-level AI cost issue, not just an infra detail.
• Why it matters: Good fit for teams running high-volume self-hosted or private LLM workloads. Apache-2.0 licence and active development.
• Watch-out: Infrastructure-grade adoption requires benchmarking under real workload patterns, not only headline latency claims.

Watchlist

• meshery/meshery: mature cloud-native management project showing continued activity; relevant for Kubernetes/platform teams, but not a new breakout.
• Panniantong/Agent-Reach: strong agent-data-access interest, but teams should scrutinise data-source terms, permissions and compliance before use.
• kenn-io/agentsview: local-first analytics for coding-agent sessions; useful category, but still needs maturity testing.

What this says about the market

The open-source signal is moving from “build an agent” to “operate an agent estate”. Security scanning, context compression, codebase memory, local runtime infrastructure and inference caching are all signs that companies are starting to care about the boring but necessary layer around AI systems.

The most durable projects this week are the ones that help teams control cost, risk, developer workflow and runtime complexity. Pure demos may still trend, but the stronger signal is infrastructure that makes AI adoption repeatable.

Editorial read

For Column readers, the practical takeaway is to watch the operational layer around AI. The near-term winners may not be the flashiest agent apps; they may be the scanners, caches, gateways, context tools and platform utilities that make AI safe enough and cheap enough to use across a company.

Executive read

• The week split into two tracks: frontier labs are improving deployment, safety and enterprise packaging, while open/local AI is moving fast on long-context, agentic and serving performance.
• OpenAI’s most important signal was not just product distribution: its deployment-simulation work points to more realistic pre-release safety testing for agents and tool-using models (OpenAI).
• Enterprise AI is becoming an operating layer: Microsoft is talking about governance and agent control planes, Databricks is bundling agents with data governance, and Snowflake is adding observability for guardrails and AI-generated artifacts.
• The open-model story is practical rather than ideological: GLM-5.2, vLLM, llama.cpp, Optimum Intel and new eval harnesses all point to more teams being able to run, serve, test and govern non-frontier models themselves.
• Product launches are converging around AI coworkers, search/visibility, home assistants, analytics agents and governed enterprise workflows.

Frontier model moves

• OpenAI introduced deployment simulation: replaying prior real conversations against candidate models to estimate undesired behavior before launch, especially relevant as models gain tools and agentic workflows (OpenAI).
• OpenAI also launched an enterprise partner network with a $150m ecosystem investment and a target to certify 300,000 consultants by the end of 2026, a clear push to make model adoption look more like cloud/SaaS rollout (OpenAI).
• Anthropic’s Claude Code analysis found that agentic coding works best when users bring domain expertise: the user increasingly decides “what” while Claude handles much of the “how” (Anthropic).
• Anthropic also disabled access to Claude Fable 5 and Mythos 5 for some customers after a US government export-control directive, a reminder that frontier-model availability is now policy-sensitive infrastructure (Anthropic).
• AWS added Google DeepMind’s Gemma 4 family to Amazon Bedrock, including multimodal input, reasoning and native function calling; this matters because open-weight-style models are being folded into managed enterprise model catalogs (AWS).
• Microsoft framed its AI push around “Intelligence + Trust,” including Microsoft IQ and Agent 365 as governance, observation and cost-control layers for agents across organisations (Microsoft).

Open and local models

• GLM-5.2 landed as a major open/MIT long-context model: Z.ai says it targets long-horizon coding and agent work with a 1m-token context, adjustable effort and improved long-context efficiency (Hugging Face).
• Artificial Analysis ranked GLM-5.2 as the leading open-weights model on its Intelligence Index, reporting 744B total parameters, 40B active parameters, 1m context and an MIT license (Artificial Analysis).
• Small models remain worth watching: Bosun-XS is a 600m-parameter relevance/warrant judge aimed at agent memory and RAG graph workflows, while SLM-10M shows how far the sub-10m-parameter tier can be pushed for tiny deployments (Bosun).
• Local deployment tooling continues to mature: llama.cpp released new multi-platform binaries and fixes, while vLLM v0.23.0 shipped a major serving update with 408 commits and broader optimisation for production inference (llama.cpp, vLLM).

Benchmarks and evals

• Artificial Analysis updated its Intelligence Index to v4.1 with more agentic workloads and cost/time/token-per-task metrics; this is more useful than pure answer accuracy because it connects model quality to operating cost (Artificial Analysis).
• Hugging Face published an “agentic enough?” evaluation workflow that measures turns, time, tool usage, errors and token consumption rather than only final-answer correctness (Hugging Face).
• AllenAI released olmo-eval, a model-development evaluation workbench for repeated checkpoint comparisons, reproducible suite definitions and agentic/multi-turn evaluation support (Hugging Face).
• The benchmark lesson this week: evaluate models as systems. For products, the useful question is not “which model is top of the board?” but “which model reliably completes the task, at acceptable latency and cost, with observable failure modes?”

Research worth reading

• MODE-RAG proposes a multi-agent approach to reducing hallucinations in multimodal RAG using outlier diagnosis, routing, causal reasoning and correction agents (arXiv).
• Agents-K1 argues for research agents that ingest full scientific papers into multimodal knowledge graphs, rather than relying on abstract-level retrieval (arXiv).
• Doctor-RAG / DR-RAG focuses on diagnosing and repairing the broken step in a multi-hop retrieval/reasoning trajectory instead of rerunning the entire agent path (arXiv).
• OpenAI’s deployment-simulation paper/post is also a research signal: safety testing is moving from static benchmark prompts toward replayed, distribution-aware simulations of real product behaviour (OpenAI).

Products people are launching

• Databricks launched Genie One, an agentic coworker that connects to Databricks plus tools such as Google Drive, Jira, Slack, Confluence and SharePoint through a Genie Ontology context layer (Databricks).
• Google launched a Gemini-first Home speaker, showing AI assistants moving from app surfaces back into ambient hardware and voice workflows (Google).
• Meta added AI Mode and creative tools inside Facebook, using Meta AI to answer from public content across surfaces such as Groups and Reels (Meta).
• Adobe introduced Brand Visibility for monitoring how brands appear in AI search surfaces such as ChatGPT, Google AI Mode, Copilot and Perplexity, a sign that “AI visibility optimisation” is turning into a software category.

Developer ecosystem

• vLLM and llama.cpp remain the two practical poles of open inference: vLLM for high-throughput, multi-user GPU serving; llama.cpp for portable local and edge deployments (vLLM, llama.cpp).
• The new evaluation tooling around agentic behaviour — Hugging Face’s harness and AllenAI’s olmo-eval — is important because agent products fail in process, not just in final answer quality (Hugging Face, AllenAI).
• Databricks announced Lakebase Search, a hybrid vector and full-text retrieval layer inside Lakebase Postgres with agent-native retrieval positioning; that is notable because it brings RAG infrastructure closer to operational data stores (Databricks).

Enterprise and data-platform angle

• Databricks expanded Agent Bricks with governed data access, memory, sandboxes, tracing, Unity AI Gateway and LakeWatch integrations, positioning agents as managed enterprise data apps rather than standalone chatbots (Databricks).
• Databricks also pushed Lakeflow as “agentic data engineering,” adding more managed design, orchestration and real-time pipeline capability under Unity Catalog (Databricks).
• Snowflake added Cortex AI Guardrails usage observability, giving teams account-usage views for scans, flagged content, credits, tokens, roles and agentic sources (Snowflake).
• Snowflake Intelligence artifacts reached GA, enabling live AI-generated charts and tables that refresh under the viewer’s credentials and preserve data permissions (Snowflake).

Editorial read

The centre of gravity is shifting from “which model is smartest?” to “which AI system can be trusted inside real workflows?” Frontier labs are building safer release processes, distribution channels and enterprise control planes. Open/local models are becoming credible enough for many controlled tasks, especially where cost, privacy or deployment flexibility matter. The next useful frontier is not just a better chatbot; it is an observable, governed, task-specific AI system with clear routing, evals, memory, permissions, cost controls and rollback paths.

For builders, the practical takeaway is to design around the stack, not the model: pick a model portfolio, instrument it with task-level evals, connect it to trusted data, and make governance visible from day one. The products that matter will not simply “add AI”; they will make AI reliable enough to sit inside everyday work.

What shipped

• Genie and AI/BI experiences: Providers can now share a Genie Space with users outside their organization using OpenSharing. (17 Jun)
• Governance and compliance controls: Rolling out starting June 22, the statement_text column in system.query.history returns for users who are not account admins or members of the databricks_pii_access account-level group. (17 Jun)
• Agentic development workflows: Omnigent is now available in Beta. Omnigent is a coding agent meta-harness that wraps coding agents such as Claude Code and Codex with a common platform: a web UI, persistent and shareable sessions, team… (17 Jun)
• Lakeflow pipeline operations: Lakeflow Designer will be available by default for workspaces with the compliance security profile enabled in late July 2026. (16 Jun)
• Lakeflow pipeline operations: Lakeflow Designer is now generally available. Lakeflow Designer is a visual, no-code experience for preparing and transforming data on a drag-and-drop canvas, with all workflows backed by production-ready… (16 Jun)
• Genie and AI/BI experiences: You can now connect Genie to Microsoft Copilot Cowork through the Genie managed MCP server, so users can ask natural-language questions about their Azure Databricks data without leaving Copilot Cowork. (16 Jun)
• Marketplace and app extensibility: You can now discover, install, and run third-party data and AI applications from Databricks Marketplace directly in your own Unity Catalog-governed workspace. (16 Jun)
• Secure data sharing: Azure Databricks will soon charge providers using OpenSharing SecureConnect for cross-region and public-internet data egress. (15 Jun)
• Genie and AI/BI experiences: You can now use the Azure Databricks Genie mobile app to access Genie One from iOS and Android devices. (15 Jun)
• Genie and AI/BI experiences: Chat in Genie One (GA) : Chat in Genie One is now generally available. Chat provides a unified interface for asking data questions in natural language using your Genie Spaces, dashboards, queries, and metric… (15 Jun)
• Databricks Runtime 19: June 15, 2026: Databricks Runtime 19 is now available in Beta, powered by Apache Spark 4.2.0. Behavioral changes JDK 17 removed : Databricks Runtime 19 only supports JDK 21. (15 Jun)
• Databricks Runtime 19 and Databricks Runtime 19 for Machine Learning are in Beta: Databricks Runtime 19 and Databricks Runtime 19 for Machine Learning are now in Beta, powered by Apache Spark 4.2.0. (15 Jun)

Why it matters

• Genie and AI/BI experiences: Databricks is expanding conversational analytics and workspace-native AI experiences for business users.
• Governance and compliance controls: The platform is tightening controls around sensitive metadata, audit trails and compliance-profile behaviour.
• Agentic development workflows: Databricks is adding more agent and Copilot-adjacent tooling around coding, analytics and workspace productivity.
• Lakeflow pipeline operations: Lakeflow updates continue the push toward lower-friction data engineering and managed pipeline design.
• Lakeflow pipeline operations: Lakeflow updates continue the push toward lower-friction data engineering and managed pipeline design.
• Genie and AI/BI experiences: Databricks is expanding conversational analytics and workspace-native AI experiences for business users.
• Marketplace and app extensibility: Databricks is making more partner and third-party functionality available directly inside the platform.
• Secure data sharing: Sharing updates point to more governed external collaboration around data products and AI/BI assets.
• Genie and AI/BI experiences: Databricks is expanding conversational analytics and workspace-native AI experiences for business users.
• Genie and AI/BI experiences: Databricks is expanding conversational analytics and workspace-native AI experiences for business users.
• Databricks Runtime 19: June 15, 2026: Databricks Runtime 19 is now available in Beta, powered by Apache Spark 4.2.0. Behavioral changes JDK 17 removed : Databricks Runtime 19 only supports JDK 21.
• Databricks Runtime 19 and Databricks Runtime 19 for Machine Learning are in Beta: Databricks Runtime 19 and Databricks Runtime 19 for Machine Learning are now in Beta, powered by Apache Spark 4.2.0.

How data teams could use it

• Genie and AI/BI experiences: Analytics teams can test where Genie reduces ad hoc report queues while keeping governed metrics and permissions in place.
• Governance and compliance controls: Security teams should check whether monitoring, troubleshooting or audit workflows need adjusting as redaction and compliance defaults change.
• Agentic development workflows: Engineering leads can trial these features on low-risk workflows and define review gates before allowing production code or pipeline changes.
• Lakeflow pipeline operations: Data teams can assess whether visual design, compliance-profile support or GA readiness changes the build-vs-code path for new pipelines.
• Lakeflow pipeline operations: Data teams can assess whether visual design, compliance-profile support or GA readiness changes the build-vs-code path for new pipelines.
• Genie and AI/BI experiences: Analytics teams can test where Genie reduces ad hoc report queues while keeping governed metrics and permissions in place.
• Marketplace and app extensibility: Platform owners should review app install controls, procurement flow and security review before enabling broad adoption.
• Secure data sharing: Teams with partner analytics or customer-facing data products should review billing, access and external-sharing controls.
• Genie and AI/BI experiences: Analytics teams can test where Genie reduces ad hoc report queues while keeping governed metrics and permissions in place.
• Genie and AI/BI experiences: Analytics teams can test where Genie reduces ad hoc report queues while keeping governed metrics and permissions in place.
• Databricks Runtime 19: June 15, 2026: Assess whether it affects platform governance, data engineering productivity, AI/BI adoption, sharing controls or workspace operations.
• Databricks Runtime 19 and Databricks Runtime 19 for Machine Learning are in Beta: Assess whether it affects platform governance, data engineering productivity, AI/BI adoption, sharing controls or workspace operations.

Editorial read

Databricks is continuing to package the lakehouse as an operating layer for data products: more AI/BI entry points for business users, more managed pipeline tooling for engineers, and more governance around sharing, audit data and external access. The practical theme is adoption with controls: turn on the useful productivity features, but pair them with app governance, metric ownership and security review.

What shipped

• Monthly Power BI feature summary: Author: Katie Murray, Senior Program Manager - Power BI continues to evolve with updates that make it easier to explore data, generate insights, and build more polished reports. (17 Jun)
• Workspace outbound access protection: Workspace outbound access protection (OAP) is a workspace-level control in Microsoft Fabric that lets you constrain where the data inside a workspace can flow. (17 Jun)
• Org apps with audiences: Announcing general availability of org apps in Power BI and Fabric, including one of the most requested capabilities: audiences. (16 Jun)
• Simplified Oracle connectivity: Connecting Power BI to Oracle has historically meant extra provider installations and data gateway deployment — even for cloud-hosted databases. (15 Jun)
• Connector migration from ODBC to ADBC: If you use connectors like Databricks, Snowflake, or BigQuery in Power BI or Fabric, there’s an important change coming. (15 Jun)

Why it matters

• Monthly Power BI feature summary: The feature summary collects reporting, modeling, Copilot and service updates into one release note surface.
• Workspace outbound access protection: Power BI reports can inherit workspace-level controls that restrict where data is allowed to flow.
• Org apps with audiences: Power BI and Fabric org apps can now target different audiences from one governed app package.
• Simplified Oracle connectivity: Oracle connections become easier by reducing dependency on extra providers and gateway-heavy setup paths.
• Connector migration from ODBC to ADBC: Power BI and Fabric connectors are moving toward Arrow Database Connectivity for several major data platforms.

How analytics teams could use it

• Monthly Power BI feature summary: BI leads can use it as a checklist for capabilities to test in report authoring, semantic modeling and self-service workflows.
• Workspace outbound access protection: Security and BI platform teams can tighten exfiltration controls for sensitive reporting workspaces without redesigning every report.
• Org apps with audiences: Analytics product teams can ship one app with role-specific navigation and content rather than maintaining duplicate distribution surfaces.
• Simplified Oracle connectivity: Teams can connect BI models to Oracle-backed operational data with less platform plumbing and faster proof-of-value cycles.
• Connector migration from ODBC to ADBC: BI teams should audit Databricks, Snowflake and BigQuery connection paths so driver changes do not surprise production refreshes.

Editorial read

Power BI is moving in two parallel directions: stronger governance for enterprise distribution, and more AI-assisted creation for reports and semantic models. For analytics products, the practical opportunity is to turn recurring BI assets into governed products: targeted audiences, safer data-flow controls, reusable DAX logic, and faster authoring loops.

What shipped

• OneLake + ServiceNow zero-copy querying: OneLake in Microsoft Fabric makes a simple promise: provide a single, unified data foundation for analytics, AI, and BI. (17 Jun)
• AI-assisted Spark failure diagnosis: An AI-powered skill that turns Spark troubleshooting from a multi-tab investigation into a single natural-language command. (17 Jun)
• Faster Spark History Server loading: As Spark workloads scale in size and complexity, fast access to Spark application execution metrics and logs is essential for debugging, performance analysis, and operational confidence. (16 Jun)
• Fabric service-issue notifications: Fabric uses several message types to communicate service issues, depending on how an issue is detected and where the information is surfaced. (16 Jun)
• Faster Spark History Server loading: SAP systems sit at the center of many enterprises’ core business operations, powering processes across finance, supply chain, manufacturing, procurement, and HR. (16 Jun)
• SAP Copy job with Microsoft ABAP Add-On: As organizations scale their data and AI investments, many are adopting a multi-platform approach so teams can use the tools that best fit each project. (16 Jun)
• Azure Databricks and OneLake interoperability: As organizations scale their data and AI investments, they increasingly adopt a multi-platform approach, enabling teams to use the tools that best fit their needs. (16 Jun)
• Real-Time Dashboard upgrades: Discover a new way to build visuals in Real-Time Dashboards. The redesigned tile editing experience in Real-Time Dashboards brings AI-assisted authoring, a larger preview area, and more flexible workflows to… (11 Jun)
• Real-Time Dashboard upgrades: Time series analysis is at the heart of understanding how data behaves over time. (11 Jun)
• Real-Time Dashboard upgrades: Real-Time Dashboards in Microsoft Fabric help you monitor live data and react to changes as they happen. (11 Jun)
• Rayfin shareable sites: If you work alongside AI tools, your screen is probably full of markdown files. They’re fast to write, easy for an agent to read, and great for keeping a record. (11 Jun)

Why it matters

• OneLake + ServiceNow zero-copy querying: Operational workflow teams can query governed OneLake data in place rather than copying it into a separate ServiceNow store.
• AI-assisted Spark failure diagnosis: Spark debugging moves from manual log chasing to a natural-language troubleshooting flow.
• Faster Spark History Server loading: Snapshot-based loading makes large Spark job metrics usable much faster, especially for high-volume batch and streaming workloads.
• Fabric service-issue notifications: Better service-status surfacing gives admins earlier context when Fabric incidents affect workloads.
• Faster Spark History Server loading: Snapshot-based loading makes large Spark job metrics usable much faster, especially for high-volume batch and streaming workloads.
• SAP Copy job with Microsoft ABAP Add-On: Fabric Data Factory can extract large SAP datasets through a Microsoft ABAP add-on with less custom extraction infrastructure.
• Azure Databricks and OneLake interoperability: Databricks customers can work with OneLake as a shared data foundation, including Unity Catalog managed-table scenarios.
• Real-Time Dashboard upgrades: Dashboard creation and monitoring are getting more interactive, including AI-assisted tile editing, time-series visuals and live refresh.
• Real-Time Dashboard upgrades: Dashboard creation and monitoring are getting more interactive, including AI-assisted tile editing, time-series visuals and live refresh.
• Real-Time Dashboard upgrades: Dashboard creation and monitoring are getting more interactive, including AI-assisted tile editing, time-series visuals and live refresh.
• Rayfin shareable sites: Fabric-adjacent AI/markdown workflows gain a more shareable presentation layer.

How analytics teams could use it

• OneLake + ServiceNow zero-copy querying: Incident, field-service, supply-chain and AI-agent workflows can be enriched with analytics context while OneLake remains the shared foundation.
• AI-assisted Spark failure diagnosis: Data engineering teams can shorten triage loops for failed notebooks, pipelines and production Spark jobs.
• Faster Spark History Server loading: Platform teams can inspect executions, tune performance and support long-running jobs without waiting on huge event logs to render.
• Fabric service-issue notifications: Analytics teams can distinguish platform disruption from pipeline or model defects and communicate impact faster.
• Faster Spark History Server loading: Platform teams can inspect executions, tune performance and support long-running jobs without waiting on huge event logs to render.
• SAP Copy job with Microsoft ABAP Add-On: Finance, supply-chain, procurement and HR analytics can land SAP operational data into OneLake for reporting, enrichment and AI use cases.
• Azure Databricks and OneLake interoperability: Mixed Fabric/Databricks estates can reduce duplicated data and let teams use preferred engines against a more consistent governed layer.
• Real-Time Dashboard upgrades: Product and operations teams can build live monitoring surfaces for telemetry, customer behaviour and business KPIs with less hand-coded UI work.
• Real-Time Dashboard upgrades: Product and operations teams can build live monitoring surfaces for telemetry, customer behaviour and business KPIs with less hand-coded UI work.
• Real-Time Dashboard upgrades: Product and operations teams can build live monitoring surfaces for telemetry, customer behaviour and business KPIs with less hand-coded UI work.
• Rayfin shareable sites: Teams can package analysis outputs for stakeholders without turning every insight into a custom app or slide deck.

Editorial read

Fabric is continuing to push OneLake as the connective tissue: less copy-and-paste data engineering, more open-table interoperability, and more operational workflows that can consume governed analytical data directly. For teams building analytics products, the practical angle is to look for places where Fabric now removes a separate platform, manual troubleshooting step, or bespoke dashboard layer.