TECHNOLOGY RADAR

Our Technology Stances

Xephyr's view on the tools and techniques that matter for AI and data teams — updated every quarter.

Last updated: Q2 2026

Top Adopt

Biggest Moves

↓DuckDBTrial → Assess
↑LangChainAssess → Trial
↓Power BITrial → Hold

New Entries(5)

✦Apache Iceberg
✦Hugging Face Transformers
✦SST (Ion)
✦dbt Metrics / MetricFlow
✦Atlan

What Is a Technology Radar?

A technology radar is a structured view of the technologies we use, evaluate, and recommend — or caution against. Inspired by ThoughtWorks, our radar is organised into four rings (Adopt, Trial, Assess, Hold) and four quadrants covering the main categories we work in: Data Infrastructure, ML & AI, Cloud & DevOps, and Analytics & BI.

Unlike a generic industry radar, ours reflects actual Xephyr engagement experience. Every technology listed is one we have built with, advised on, or consciously decided not to recommend. The radar is honest — including the things we've moved to Hold.

How We Evaluate

We review the radar every quarter. Each technology is discussed by our delivery team based on recent project experience, ecosystem maturity, vendor trajectory, and the specific needs of the clients we serve. Ring placements reflect our current best judgement — not vendor relationships, sponsorships, or trend chasing.

Adopt—proven in production, recommended for most projects

Trial—worth pursuing in the right context

Assess—worth exploring, not production-ready

Hold—proceed with caution, consider alternatives

Data Infrastructure

ML & AI

Cloud & DevOps

Analytics & BI

Xephyr Technology Radar — Q2 2026
Technology	Quadrant	Ring	Change	Description
Snowflake	Data Infrastructure	Adopt	Unchanged	Snowflake remains the leading cloud data warehouse for organisations that need fast, scalable SQL analytics without managing infrastructure. Separation of storage and compute continues to offer best-in-class cost management for variable workloads. // TODO: real Xephyr opinion
dbt	Data Infrastructure	Adopt	Unchanged	dbt has become the standard for SQL-based data transformation, bringing software engineering practices — version control, testing, documentation — to analytics workflows. The modular, DAG-based approach makes complex transformations maintainable. // TODO: real Xephyr opinion
Apache Spark	Data Infrastructure	Adopt	Unchanged	Apache Spark remains the dominant framework for large-scale distributed data processing. PySpark in particular is the standard for data engineering teams handling batch transformations at scale. // TODO: real Xephyr opinion
Databricks	Data Infrastructure	Trial	Unchanged	Databricks unifies data engineering, machine learning, and analytics on a single lakehouse platform built on Delta Lake. Unity Catalog adds enterprise-grade data governance. The platform accelerates ML workflows significantly for teams already using Spark. // TODO: real Xephyr opinion
Fivetran	Data Infrastructure	Trial	Unchanged	Fivetran provides managed, schema-adaptive ELT connectors that significantly reduce the time to ingest data from SaaS sources. Pricing by monthly active rows is cost-effective for mid-market companies. // TODO: real Xephyr opinion
Apache Iceberg	Data Infrastructure	Trial	New this quarter	Apache Iceberg is an open table format that brings ACID transactions, schema evolution, and time-travel queries to data lakes. Broad multi-engine support (Spark, Flink, Trino, DuckDB) makes it the emerging standard for lakehouse storage. // TODO: real Xephyr opinion
DuckDB	Data Infrastructure	Assess	Moved out ↓	DuckDB is an in-process analytical query engine that runs SQL directly on Parquet, CSV, and JSON files with exceptional speed. Zero-dependency, embeddable, and increasingly production-ready for mid-scale analytics. // TODO: real Xephyr opinion
Trino	Data Infrastructure	Assess	Unchanged	Trino (formerly PrestoSQL) is a distributed SQL query engine for federated queries across multiple data sources. Excels at interactive queries across heterogeneous data stores without ETL. // TODO: real Xephyr opinion
PyTorch	ML & AI	Adopt	Unchanged	PyTorch is the dominant deep learning framework for research and production. Dynamic computation graphs, extensive ecosystem (Hugging Face, Lightning, torchvision), and first-class GPU support make it the default choice for ML teams. // TODO: real Xephyr opinion
scikit-learn	ML & AI	Adopt	Unchanged	scikit-learn remains the standard for classical machine learning in Python. Consistent API, excellent documentation, and broad algorithm coverage make it the first tool we reach for when deep learning is unnecessary. // TODO: real Xephyr opinion
Weights & Biases	ML & AI	Adopt	Unchanged	Weights & Biases is the leading MLOps platform for experiment tracking, model visualisation, and artifact management. Integration is minimal — a few lines of code — and the value is immediate for any team training models. // TODO: real Xephyr opinion
LangChain	ML & AI	Trial	Moved in ↑	LangChain provides a composable framework for building LLM-powered applications — chains, agents, RAG pipelines, and tool use. The ecosystem has matured significantly, and LangGraph adds robust workflow orchestration for complex agent patterns. // TODO: real Xephyr opinion
OpenAI API	ML & AI	Trial	Unchanged	The OpenAI API provides access to GPT-4o, o3, and specialised models for text, vision, and reasoning. Structured outputs and function calling make integration into production applications straightforward. // TODO: real Xephyr opinion
Hugging Face Transformers	ML & AI	Trial	New this quarter	The Hugging Face Transformers library provides access to thousands of pre-trained models for NLP, vision, and multimodal tasks. Model Hub, Spaces, and Inference Endpoints make the full ML lifecycle more accessible. // TODO: real Xephyr opinion
MLflow	ML & AI	Trial	Unchanged	MLflow is an open-source platform for ML lifecycle management — experiment tracking, model registry, and deployment. Strong Databricks integration makes it compelling for teams already on that platform. // TODO: real Xephyr opinion
Pinecone	ML & AI	Assess	Unchanged	Pinecone is a managed vector database designed for embedding storage and similarity search at scale. Simple API, good performance characteristics, and serverless tier make it accessible. // TODO: real Xephyr opinion
AWS Lambda	Cloud & DevOps	Adopt	Unchanged	AWS Lambda is the default serverless compute platform for event-driven workloads, APIs, and data processing triggers. Cold start improvements and Container image support have addressed most prior objections. // TODO: real Xephyr opinion
Terraform	Cloud & DevOps	Adopt	Unchanged	Terraform (OpenTofu fork included) is the standard infrastructure-as-code tool for multi-cloud environments. Declarative HCL, extensive provider ecosystem, and state management make it the default for infrastructure teams. // TODO: real Xephyr opinion
GitHub Actions	Cloud & DevOps	Adopt	Unchanged	GitHub Actions is the default CI/CD platform for repositories hosted on GitHub. Native integration, marketplace of reusable actions, and matrix builds make it fast to set up and powerful in practice. // TODO: real Xephyr opinion
Docker	Cloud & DevOps	Adopt	Unchanged	Docker containers remain the standard packaging unit for applications and ML models. Reproducible environments, straightforward local development, and universal deployment target support. // TODO: real Xephyr opinion
SST (Ion)	Cloud & DevOps	Trial	New this quarter	SST Ion is a framework for deploying full-stack applications to AWS using Pulumi under the hood. OpenNext integration deploys Next.js to CloudFront + Lambda with minimal configuration. Purpose-built for JavaScript/TypeScript applications on AWS. // TODO: real Xephyr opinion
Kubernetes	Cloud & DevOps	Trial	Unchanged	Kubernetes is the standard container orchestration platform for production workloads at scale. Managed flavours (EKS, GKE, AKS) reduce operational burden significantly. // TODO: real Xephyr opinion
Datadog	Cloud & DevOps	Trial	Unchanged	Datadog is the leading cloud-native observability platform — APM, infrastructure monitoring, log management, and ML model monitoring in one platform. Deep AWS integration and LLM Observability module are particularly relevant for AI workloads. // TODO: real Xephyr opinion
AWS CDK	Cloud & DevOps	Assess	Unchanged	AWS CDK lets teams define AWS infrastructure using TypeScript, Python, or Java with full IDE support and type safety. Synths to CloudFormation, giving access to every AWS resource. // TODO: real Xephyr opinion
Looker	Analytics & BI	Adopt	Unchanged	Looker (Google Cloud) is the leading semantic-layer BI tool for organisations that want a single source of truth for business metrics. LookML centralises business logic and enables consistent, governed reporting across teams. // TODO: real Xephyr opinion
dbt Metrics / MetricFlow	Analytics & BI	Trial	New this quarter	dbt MetricFlow (dbt Semantic Layer) brings metric definitions into the dbt project, enabling consistent metric computation across BI tools. Integrations with Tableau, Looker, and Mode are maturing. // TODO: real Xephyr opinion
Great Expectations	Analytics & BI	Trial	Unchanged	Great Expectations is the leading open-source data quality framework, enabling teams to define, document, and automatically validate data quality expectations as part of their pipelines. // TODO: real Xephyr opinion
Metabase	Analytics & BI	Trial	Unchanged	Metabase is a self-serve BI tool that enables non-technical users to explore data without writing SQL. Simple deployment, pleasant UX, and a generous open-source edition make it ideal for fast-moving teams. // TODO: real Xephyr opinion
Apache Superset	Analytics & BI	Assess	Unchanged	Apache Superset is an open-source BI platform with a wide database connector library and strong SQL-native exploration. Self-hosted, extensible, and increasingly production-ready. // TODO: real Xephyr opinion
Atlan	Analytics & BI	Assess	New this quarter	Atlan is a modern data catalog and governance platform that combines metadata management, lineage tracking, and collaboration. Tight integrations with dbt, Snowflake, and Databricks make it compelling for lakehouse teams. // TODO: real Xephyr opinion
Monte Carlo	Analytics & BI	Assess	Unchanged	Monte Carlo is a data observability platform that automatically detects data quality anomalies, monitors freshness and volume, and traces issues through lineage. Requires minimal setup compared to custom monitoring. // TODO: real Xephyr opinion
Power BI	Analytics & BI	Hold	Moved out ↓	Power BI is Microsoft's BI platform, tightly integrated with the Microsoft 365 ecosystem. Strong for organisations already on Azure, but the DirectQuery performance, licensing model, and semantic layer limitations create friction at scale. // TODO: real Xephyr opinion

Historical data coming soon — previous quarter comparisons will be available in a future update.

GET STARTED

Book a Technology Assessment Assessment?

Not sure where your technology investments should go? We run structured technology assessments that map your current stack against industry best practices — and give you a clear, prioritised roadmap.

Book a Technology Assessment