A complete SAP HANA + BO/SAC to Databricks migration guide — business case, TCO model, 5-phase technical playbook, and 30-day action plan. Built from a real Fortune 500 engagement.

Databricks Data + AI Summit 2026 made one thing clear: this is no longer just a data platform. From sub-millisecond reads with Lakehouse//RT to CustomerLake, App Builder, and ZeroOps, the announcements tell a single architectural story. Here's what stood out, and why the most important AI news wasn't about AI at all.

Databricks now supports 12 official Docker base images, and you can build your own on top of them. Custom containers let data engineering teams control the full runtime (including native binaries, enterprise certificates, internal packages, and offline environments) not just Python dependencies. This post walks through how to build, test, and deploy a custom Databricks container, with nine production use cases from compliance runtimes to DuckDB integration.

Unity Catalog catalog commits move Delta table commit coordination from filesystem-level operations into Unity Catalog, making UC the authoritative source of table state. This enables strict concurrency control, attribute-based access for external engines, and multi-statement multi-table transactions — while leaving data in open Delta format on cloud storage.

Declarative Automation Bundles mutators let you define job parameters once in a central config file and inject them into every job automatically at deploy time — no more copy-pasting catalog names, schema paths, and environment variables across your bundle. This post walks through a working implementation with full code examples.

Earning a Databricks BrickBuilder Specialization in Financial Services requires more than technical credentials — it demands domain knowledge built across banks, insurers, and investment firms operating under strict audit and regulatory requirements. SunnyData's specialization reflects years of production-ready implementations where data errors have real consequences. For financial services leaders evaluating data partners, it's a signal worth understanding beneath the surface.

Most Databricks teams using Bundles are only scratching the surface. The real value isn't in the deployment syntax — it's in the engineering discipline Bundles makes enforceable: explicit dependency management, reproducible local environments, automated quality gates, and CI/CD as the only path to production. This post breaks down what a production-ready Databricks project structure actually looks like, and the software engineering practices that make it ship with confidence.

Databricks Lakeflow Connect now supports query-based capture — a watermark-based incremental ingestion pattern for teams that can't or won't enable Change Data Feed. Using a cursor column, soft deletes, and optional hard-delete sync, teams can achieve true incremental processing from SQL sources like PostgreSQL, Azure SQL, and MySQL — without a gateway, and without full table reloads that silently drain your compute budget.

Informatica's new LTS pricing means staying put is no longer free — it's a recurring tax on a platform with a shrinking capability roadmap. This guide covers the full migration path from PowerCenter to the Databricks Lakeflow stack: the CFO-ready business case, a construct-by-construct translation guide, and the five-phase M5 methodology SunnyData runs on every engagement. If your team is weighing IICS versus a full re-platform, the architecture decision and cost model are here.

Naming conventions only work if they're enforced — and a Confluence page nobody reads isn't enforcement. This post walks through using Databricks Workspace Skills to make naming rules executable in Genie Code, then scaling that to a catalog-wide audit agent built with Databricks Apps and DABS. The result is automated, repeatable governance that runs without requiring engineers to opt in.

Databricks AUTO CDC isn't just the easiest way to process Change Data Feed, it's also the cheapest. A benchmark across 25 million INSERT, UPDATE, and DELETE operations found AUTO CDC outperformed Structured Streaming and SQL Warehouse on cost in every run, even as the target table grew to 72 million records. Structured Streaming remains the right choice for custom logic; AUTO CDC wins on standard SCD Type 1/2 patterns at scale.

As teams migrate infrastructure definitions into Declarative Automation Bundles, Terraform still owns the Azure layer — Key Vaults, resource groups, networking. This post walks through a clean, CI/CD-ready pattern for passing Terraform outputs directly into bundle variable overrides, eliminating manual config steps and the environment drift that follows them.

Databricks Genie Code hit a 77.1% task success rate in production data science workflows — more than double what general-purpose coding agents achieve. But that performance is entirely conditional on the quality of your Unity Catalog metadata. SunnyData's two-week evaluation breaks down what works, what doesn't, and the governance layer you need before you go live.

Five common Databricks coding patterns — including undocumented API calls, manual SparkSession instantiation, and hardcoded Spark configs — that pass code review but fail silently in serverless environments or during platform migrations. For each anti-pattern, this post explains why it breaks and shows the correct native Databricks approach using DABS, the Databricks SDK, and dynamic job parameters.

Databricks Lakewatch replaces the traditional SIEM model with an Open Security Lakehouse — storing 100% of telemetry in open formats at up to 80% lower TCO. AI agents reason across years of unified data to detect and respond at machine speed, closing the visibility gap that legacy SIEMs were structurally forced to create. Early customers include Adobe and Dropbox, with broader availability following Private Preview.