The unified platform experience is what keeps me on Databricks. Having notebooks, pipelines, SQL warehouses, ML, and governance all in one place under Unity Catalog means I’m not constantly stitching together five different tools just to get work done.

Lakeflow Pipelines (formerly DLT) makes it straightforward to build medallion-architecture pipelines, and the Photon engine delivers real performance gains on SQL workloads without requiring any code changes. Recent additions like Genie Code and background agents also show they’re serious about agentic AI—it doesn’t feel like a bolt-on copilot, because it can actually understand your data context through Unity Catalog. Serverless compute has been another big quality-of-life improvement as well, since I no longer have to wait for cluster spin-up when I just want to run quick, ad hoc queries. Review collected by and hosted on G2.com.

Cost management can be tricky—DBUs add up quickly if you’re not careful with cluster sizing and warehouse auto-scaling. The pricing model also isn’t always transparent, especially when you’re mixing serverless and classic compute.

Unity Catalog is powerful, but the initial setup and the migration from legacy HMS can be painful, particularly for large orgs with years of existing Hive metastore objects. The documentation is generally good, yet it sometimes lags behind new feature releases. On top of that, the workspace UI can feel sluggish at times, especially when you’re working with a large number of assets. Review collected by and hosted on G2.com.

The all-in-one platform eliminates tool sprawl. Unity Catalog gives you governance, lineage, and discoverability without bolting on a separate catalog. The notebook UI is clean and makes iterating on PySpark fast. Genie is the standout AI feature it turns curated tables into natural language interfaces for business users, and the SDK lets you configure it programmatically so it stays maintainable. DLT handles pipeline orchestration well. Performance on Spark workloads is solid, especially with Photon. Integrations with Airflow, S3, and the broader ecosystem are straightforward. For the ROI, consolidating what used to require multiple tools into one platform pays for itself in reduced complexity. Review collected by and hosted on G2.com.

Pricing can be hard to predict. Compute costs scale quickly if you're not careful with cluster sizing and SKU selection, and it's not always obvious which workload tier you actually need until you see the bill. The notebook IDE, while functional, still lags behind a real editor for refactoring, multi-file navigation, and code review workflows Review collected by and hosted on G2.com.

What I like most about Databricks is how its features have consistently matched the evolving needs of engineering teams. Over the years, I’ve seen it grow from a solid data platform into a workspace that genuinely streamlines how we build and manage data and AI solutions. Unity Catalog has been one of the biggest improvements for us having a single place to manage permissions and lineage has removed a lot of manual steps we used to handle separately across systems. Genie AI and BI have also become part of my regular workflow; being able to generate SQL or explore datasets through natural conversations helps teams get to answers faster, especially when we’re under time pressure. The Apps capability has added unexpected value by letting us create and share simplified internal tools directly within the platform, eliminating the need to stand up extra infrastructure. And with Lakebase, we’ve been able to support more transactional-style use cases without losing the flexibility of a lake, which has made certain pipelines far easier to maintain. Altogether, these improvements have removed a lot of friction from day‑to‑day work and made the platform something I genuinely enjoy using as it continues to evolve. Review collected by and hosted on G2.com.

What I dislike about Databricks is that some of the newer AI experiences especially Genie for code generation can feel unstable at times and may lose context during longer development sessions. It disrupts my workflow when the assistant can’t retain earlier logic or maintain continuity across multiple iterations.

I’ve also noticed a gap in native connectors for certain enterprise systems like DFS, SMB shares or windows-based source systems, and platforms such as DB2 on AS/400, which many customers still rely on. Even though Databricks continues to expand its ecosystem, the lack of direct connectivity in these areas often means we need extra middleware or custom pipelines to bridge the gap.

None of these are deal-breakers, but they’re areas where the platform’s otherwise smooth experience can still feel a bit incomplete. Review collected by and hosted on G2.com.

As a data engineer, I use Databricks Genie to interact with data in natural language, while still relying on the same governed tables, metrics, and semantic models that my team has built. Instead of jumping straight into SQL notebooks for every exploratory ask, I or business users can phrase questions in plain language and let Genie translate them into structured, catalog‑aware queries. This keeps self‑service fast but also secure and governed. Review collected by and hosted on G2.com.

Laptop stability when multitasking

My laptop can hang or become noticeably sluggish when I’m working with multiple Genie tabs and dashboards at the same time, especially during heavier queries or more demanding visualizations. This hurts the overall user experience and can slow down iterative development and analysis.

Latency with complex data models

With very wide schemas or more complex semantic models, Genie sometimes selects suboptimal joins or an overly broad/narrow level of granularity. As a result, I still need to review the generated SQL and optimize it myself. In that sense, it remains a helpful assistant rather than a fully autonomous query engine. Review collected by and hosted on G2.com.

Databricks Genie's contextual understanding of Unity Catalog is genuinely impressive. While working through a complex UC migration, navigating three-level namespaces, volume paths, security modes, and widget-driven SQL execution, Genie reasoned through the specifics instead of falling back on generic answers. It really speaks the UC migration language, which cuts down on a lot of back-and-forth and makes troubleshooting feel more direct. Overall, the platform is powerful for managing large-scale data engineering work across Python, Scala, and notebook-based pipelines, all in one place. Review collected by and hosted on G2.com.

My biggest frustration with Genie is the lack of persistent session memory. On a long-running migration project with 60+ test cases and multiple interconnected components, having to re-establish context every session creates real overhead. Genie also struggles with cross-component reasoning: it handles individual notebooks well, but tracing issues across multiple layers of a framework is still largely a manual effort. Occasionally, the responses feel overly cautious when what’s needed is a more direct, confident answer. Review collected by and hosted on G2.com.

I use Databricks for ETL, Reporting, and AI, and I appreciate that it works as one unified solution for all data and AI needs. It makes it easier to track data and create insights, helping us deal with data silos. I like the Unity Catalog as it helps us manage and govern data in one place. I also like using AgentBricks as a multi-agent system for creating AI applications from PDFs and other documents. I find Genie valuable as it allows business users to ask questions in natural language and get exact answers. The initial setup of Databricks was very easy, making the transition smooth. Review collected by and hosted on G2.com.

I think workflow could be improved by adding multiple triggers to the same pipeline, as for now, if we want to schedule the same pipeline multiple times in a day, we have to clone it for each time. Review collected by and hosted on G2.com.

Genie Code (Databricks Assistant Agent) — I’m currently working on migrating existing workloads from ADF and SQLMI to Databricks. As part of that, I need to convert stored procedures and ADF dataflows into Databricks notebooks. Initially, we refactored all the code manually, but once Agent Mode was available in preview, we tried using it to convert the stored procedures and dataflows into Databricks PySpark code. I was impressed by the accuracy: it handled about 90% of the code conversion without errors, aside from some case-handling and similar adjustments.

Also, Lakeflow Connect helped me connect SharePoint and SFTP data to Databricks more easily. Review collected by and hosted on G2.com.

It’s not a major issue, but in my project the client asked us to generate table and column descriptions using AI in Unity Catalog. For each environment, these descriptions vary, and I have around 300 tables just in the Bronze zone. Having to click into each table and generate AI descriptions one by one is very time-consuming, and the results are not consistent across environments.

It would be much more efficient if we had an option to generate descriptions at the schema level, and if there were an information schema or system tables that stored table and column descriptions as metadata. That way, we could easily replicate them across environments. In some cases, clients also have source system documentation we could leverage to generate more accurate table and column descriptions. Review collected by and hosted on G2.com.

What I like best about Databricks is how it finally delivered what every data engineer/data professional has been wishing for — a true unified analytics and AI platform.

I remember working across five different tools just to get a single pipeline from ingestion to reporting. Databricks collapsed all of that into one environment, and that changed everything for me.

Delta Lake was the first breakthrough. When it arrived around 2020, ACID transactions and time‑travel immediately eliminated the operational pain we used to consider “normal.” If a job corrupted a table, I could roll back to a previous version in seconds instead of spending hours restoring backups. That reliability alone saved multiple downstream failures.

Before Delta existed, our pipelines relied heavily on overwrite patterns because there was no reliable way to apply updates or handle late‑arriving data safely. Overwrites were slow, expensive, and risky — especially for large tables. A single failure during overwrite could leave the table in a half-written, inconsistent state. Processing took longer, compute costs shot up, and recovery often meant manually rebuilding partitions from scratch.

The ROI became obvious as soon as we used Databricks end‑to‑end. Because one platform handles ingestion → transformation → ML → BI → governance, we retired entire categories of legacy tools and reduced operational overhead dramatically.

Then Genie arrived — and it genuinely transformed my day‑to‑day work.

I once needed a PySpark module for data quality checks. Genie generated the full logic — null checks, schema validation, aggregations — in seconds. Instead of spending 30 minutes writing boilerplate, I spent 3 minutes refining the logic. It shifted my focus from syntax to decisions.

Integrations are another strength. Connecting Databricks to S3, SQL Server, and especially Power BI has been seamless. Publishing Delta tables directly to BI models removed the need for brittle extracts and sped up refreshes. Unity Catalog made everything even cleaner with consistent permissions and lineage.

Performance is consistently strong when it matters — heavy joins, window functions, multi‑stage pipelines, or streaming workloads. Serverless compute starts instantly, and workloads scale predictably even under pressure.

Finally, onboarding surprised me. Features like serverless compute, natural‑language queries, AI‑generated code suggestions, and automatic comments make Databricks intuitive even for engineers new to Spark. It feels like the platform actively helps you learn.

In short: Databricks lets me work faster, recover instantly, integrate seamlessly, and scale confidently — all in one place. It’s the rare platform that improves both speed and reliability at the same time. Review collected by and hosted on G2.com.

What I dislike most about Databricks is the cost visibility and predictability.

Even as an experienced engineer, it can be difficult to get a straight, real‑time view of what a workflow will cost before running it. Photon vs. standard runtime, autoscaling behaviour, shuffle-heavy operations, DBUs—these can stack up quickly, and cost surprises happen unless you actively monitor and tune everything. A simple pipeline misconfiguration can quietly double your spend.

Another challenge is the rapid pace of new features and changes.

Databricks innovates incredibly fast, which is great, but it also means features may land before documentation, best practices, or governance patterns are fully mature. Sometimes functionality behaves differently across runtimes or cloud providers, and staying on top of everything requires continuous learning and refactoring. This can create team friction and technical debt.

In short: Databricks is exceptional, but the cost model isn’t always transparent, and the rapid feature rollout can introduce operational complexity that teams must actively manage. Review collected by and hosted on G2.com.

I appreciate how Databricks helps our different teams to collaborate together. A shared notebook project allows our marketing analysts, supply chain engineer, and data scientist to work together in real-time. The version history within Databricks prevents any confusion about the latest developments. I like that it helped us build predictive models to forecast demand more accurately, and we created visual dashboards that we shared with our leadership team, giving them clear regional insights. Within a few months, we saw measurable lifts in sales and improved profit margins. Databricks streamlined processes that used to take weeks of manual spreadsheets and emails, now happening in days with fresh and reliable data. The cross-departmental data sourcing it provides breaks down silos in our organization, enabling smarter and faster decision-making based on a complete picture rather than fragmented departmental views. The setup was straightforward right on the cloud, avoiding messy offline setups. Review collected by and hosted on G2.com.

There were delays in obtaining permissions for enterprise security and approval workflows while sharing data. Review collected by and hosted on G2.com.

Databricks is great because it brings everything you need for data and AI into one place.

Instead of switching between different tools for data engineering, data cleaning, analytics, and machine learning, you can do it all in a single environment. That makes life a lot easier. Review collected by and hosted on G2.com.

Databricks is not beginner-friendly. You often need solid data engineering skills to use it effectively.

Reviews point out that while Databricks is extremely capable, it’s “a high‑end workshop” that requires expertise and is not easy for less technical teams.Databricks uses cost units (DBUs), which many people find difficult to estimate and manage.

Even expert reviews highlight that its pricing is famously complicated and can hide unexpected costs. Review collected by and hosted on G2.com.