Skip to content
Blueprint Catalog

    ↑↓ move · Enter open · Esc close

    11 — Technology · 114 stances · updated

    Blueprint Radar

    Mike Reams's stance on every technology in the Technology Reference Model, each with a one-line reason and the date it was set. It is one architect's opinion for a mid-to-large enterprise starting new work, not a fact about any product and not an endorsement by anyone named here.

    Standard64
    The default for its job. Use it unless there is a specific reason not to.
    Pilot31
    Worth a bounded trial on real work, with an exit plan.
    Watch15
    Track it. Not yet for production commitments.
    Retire4
    Start nothing new on it, and plan to move existing use off it.
    123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114
    StandardPilotWatchRetire

    blueprint-radar.csv ↗ for Build Your Own Radar · JSON ↗

    Languages & Frameworks

    Standard · 16

    1. .NET — Default platform for Microsoft-stack services; cross-platform and on a predictable LTS cadence.
    2. Angular — Opinionated full framework that suits large teams wanting one structure; fine where already established.
    3. axe-core — Default automated accessibility rules engine; wire it into CI, knowing it finds only part of WCAG issues.
    4. Django — Default Python web framework for data-backed apps; built-in admin and auth save weeks of work.
    5. Expo — Default toolchain for React Native: builds, updates and native modules without hand-maintained native projects.
    6. Go — Default language for infrastructure tools and lean network services; simple deployment as a single binary.
    7. Next.js — Default React application framework for server-rendered web apps; large ecosystem, though it pulls you toward one host.
    8. Node.js — Default JavaScript server runtime; use only even-numbered LTS releases in production.
    9. Python — Default language for data, ML and automation work, with the deepest library ecosystem in those areas.
    10. PyTorch — Default framework for training and fine-tuning models; most research code and model releases target it.
    11. React — Default UI library for web front ends; widest talent pool and component ecosystem.
    12. React Native — Default cross-platform mobile framework when your team already writes React.
    13. React Navigation — Default navigation library for React Native apps; Expo Router builds on it.
    14. Spring Boot — Default framework for Java services; mature, well documented and widely staffed.
    15. Tailwind CSS — Utility CSS that maps cleanly onto design tokens and stays consistent across teams without a custom CSS layer.
    16. TypeScript — Default language for browser and Node.js code; static types catch interface drift before it ships.

    Pilot · 6

    1. Apache Camel — Mature JVM integration library with hundreds of connectors; trial it where code-first integration beats an iPaaS.
    2. Astro — Good fit for content-heavy sites that ship little JavaScript; trial it for documentation and marketing sites.
    3. Backstage — Open developer portal framework; flexible, but it is a product you build and staff, not a box you install.
    4. LangGraph — Explicit graph-based agent workflows with state and human review; trial it on one bounded agent use case.
    5. Ruby on Rails — Productive full-stack framework for small teams; trial it where Ruby skills exist, since enterprise talent is thinner.
    6. Sidekiq — Mature background job processor for Rails; adopt it with Rails, and compare with the built-in Solid Queue.

    Watch · 5

    1. DeepEval — Test-style evaluation for LLM outputs that fits CI; the eval tooling field is still settling, so track before standardizing.
    2. Google Agent Development Kit — Young open agent framework tied closely to Google Cloud; track it unless Gemini platform services are already in use.
    3. Hotwire — Server-rendered interactivity with little JavaScript; strong inside Rails, less proven elsewhere, so track it.
    4. Mastra — TypeScript agent framework that is young and moving fast; track it before production commitments.
    5. Svelte — Lean compiled UI framework that developers like; smaller ecosystem and talent pool than React for enterprise work.

    Platforms

    Standard · 19

    1. Apache Airflow — Default open orchestrator for batch data pipelines; every major cloud offers it managed.
    2. Apache Kafka — Default backbone for durable, replayable event streams; heavy to operate yourself, so prefer a managed service.
    3. Apache Spark — Default engine for large-scale batch processing; most managed data platforms run it under the hood.
    4. Grafana — Default open dashboarding layer that reads from almost any metrics, logs or tracing source.
    5. HashiCorp Vault — Default dedicated secrets manager across clouds; note the source-available license since 2023.
    6. Jira Service Management — Practical ITSM for small to mid-size IT teams already on Jira; plan on Cloud, since self-managed has an end date.
    7. Kubernetes — Default container orchestration when you run many services; use a managed offering, not a hand-built cluster.
    8. Microsoft SQL Server — Default relational database in Microsoft shops; license cost argues for PostgreSQL where the stack is open.
    9. MLflow — Open default for experiment tracking and a model registry; avoids locking model metadata into one cloud.
    10. MySQL — Acceptable relational choice where an application requires it; PostgreSQL is the default for new systems.
    11. NGINX — Default web server and reverse proxy; stable, fast and understood by every operations team.
    12. PostgreSQL — Default relational database: open license, every major cloud runs it managed, and the extension ecosystem is deep.
    13. Prometheus — Default open metrics system for Kubernetes and services; its query language and exposition format are widely adopted.
    14. RabbitMQ — Default broker for task queues and request routing where you need messaging, not a replayable log.
    15. Red Hat Enterprise Linux — Default Linux where certified vendor support or long lifecycles matter; RHEL 10 supported to 2035.
    16. Redis — Default in-memory cache and fast data structures; Redis 8 added an AGPL option, and Valkey is the BSD-licensed fork.
    17. ServiceNow — Default enterprise ITSM and CMDB platform at scale; keep customization low to protect upgrades.
    18. Ubuntu — Default Linux for cloud VMs and containers when you do not need a paid enterprise subscription; LTS releases only.
    19. Windows Server — Required for Windows-only workloads; prefer Linux for new services that do not depend on it.

    Pilot · 13

    1. Apache Flink — Strongest open engine for stateful stream processing, but costly to run; use a managed service and a real use case.
    2. Confluent Schema Registry — Use a schema registry with Kafka so contracts evolve safely; check its license terms against Apicurio first.
    3. Debezium — Log-based change data capture avoids dual writes; trial on one source system before relying on it broadly.
    4. Elasticsearch — Strong full-text search, but weigh its licenses and the Apache-licensed OpenSearch fork before a new estate.
    5. Grafana Loki — Lower-cost logging that indexes labels only; trial it where full-text search across logs is rarely needed.
    6. Grafana Tempo — Low-cost trace storage that pairs with OpenTelemetry; trial it if Grafana is already your observability front end.
    7. Istio — Adds mTLS and traffic policy across services, but it is real operational weight; trial it where those needs are proven.
    8. Keycloak — Self-hosted OIDC and SAML provider for cases a SaaS identity provider cannot cover; you own patching and uptime.
    9. Kong Gateway — Capable gateway on an open-core model; many enterprise features need the paid license, so price the full need.
    10. Langfuse — Open LLM tracing and evaluation you can self-host; trial it to see what your prompts and costs actually do.
    11. LiteLLM — One API across model providers with spend controls; trial behind pinned versions given its 2026 supply-chain incident.
    12. MongoDB — Fits document-shaped data with changing schemas, but the SSPL license and relational alternatives argue for a trial first.
    13. Trino — Federated SQL across lakes and databases is useful, but run it where one query layer clearly beats moving data.

    Watch · 8

    1. Apache APISIX — Fully open API gateway with dynamic config; track it as an alternative to open-core gateways.
    2. DataRobot — Automated ML suite with a premium cost; hyperscaler ML platforms now cover most of its ground.
    3. Feast — Feature stores pay off only with many models sharing features; most teams do not need one yet.
    4. IBM watsonx.ai — Relevant mainly to IBM-centric estates; the hyperscaler platforms have broader model choice and ecosystem.
    5. KServe — Open model serving on Kubernetes; most teams are better served by managed inference until volume justifies it.
    6. Kubeflow — Broad but complex ML platform on Kubernetes; managed ML services cover most needs with far less upkeep.
    7. OpenText Content Management — Enterprise records management for regulated content; heavy to run, so justify it with retention needs.
    8. SigNoz — Single OpenTelemetry-native backend for logs, metrics and traces; track it as a lighter alternative to larger stacks.

    Retire · 2

    1. OpenLDAP — Start new applications on OIDC and SCIM via your identity provider; keep LDAP only for systems that cannot change.
    2. Oracle Database — Start new systems on PostgreSQL; keep Oracle only where a packaged application or existing license demands it.

    Cloud Services

    Standard · 14

    1. Amazon API Gateway — Default managed API front door where you are on AWS; little to operate and integrates with Lambda and IAM.
    2. Amazon Bedrock — Default foundation-model access where you are on AWS; several model families behind one API and AWS IAM.
    3. Amazon DynamoDB — Default key-value store on AWS when access patterns are known up front; poor fit for ad-hoc queries.
    4. Amazon S3 — Default object storage where you are on AWS, and its API is the de facto interface other stores copy.
    5. Amazon SageMaker AI — Default managed ML training and serving where you are on AWS.
    6. Amazon SQS — Default managed queue on AWS for decoupling work between services; nothing to run or patch.
    7. AWS Lambda — Default serverless compute where you are on AWS, for event-driven work with spiky or low traffic.
    8. Gemini Enterprise Agent Platform — Default AI and agent platform where you are on Google Cloud; it is the renamed Vertex AI with agent services added.
    9. GitHub — Default source control and code review platform; the largest developer ecosystem and integration surface.
    10. GitHub Actions — Default CI/CD where code lives on GitHub; pipelines sit next to the code and the action ecosystem is large.
    11. Microsoft 365 — Default productivity and collaboration suite for most enterprises; identity, mail and files share one tenant.
    12. Microsoft Foundry — Default model and agent platform where you are on Azure; formerly Azure AI Foundry.
    13. Okta — Default workforce identity provider for multi-cloud estates; Microsoft Entra ID is the natural choice in Microsoft shops.
    14. Snowflake — Proven managed warehouse with low operational effort; watch consumption costs and prefer Iceberg tables for portability.

    Pilot · 5

    1. Amazon Bedrock AgentCore — Managed runtime, memory and identity for agents on AWS; trial with one agent before standardizing.
    2. Datadog — Broad managed observability with quick time to value, but costs climb with hosts and data; model the bill first.
    3. GitHub Copilot — Easiest AI coding assistant to roll out where you already license GitHub; measure delivery outcomes before going wide.
    4. Port — Hosted developer portal with less build effort than Backstage; trial it if you lack a team to run a portal.
    5. Vercel — Fast path for Next.js front ends, but usage-based pricing and platform-specific features need a cost and exit review.

    Retire · 1

    1. Cloudflare Pages — Cloudflare now points new sites to Workers with static assets and publishes a Pages migration guide; start there.

    Tools & Standards

    Standard · 15

    1. Alertmanager — Default alert routing and grouping for Prometheus-based monitoring.
    2. Ansible — Default agentless configuration management for servers and network devices; complements Terraform.
    3. Apache Iceberg — Open table format that keeps lake data portable across engines and vendors instead of locked to one warehouse.
    4. Argo CD — Default GitOps controller for Kubernetes: declared state in Git, drift visible, rollbacks by revert.
    5. cert-manager — Default certificate automation for Kubernetes; removes manual renewals, a common cause of outages.
    6. dbt — Default way to manage SQL transformations as versioned, tested code in the warehouse.
    7. Helm — Common packaging format for Kubernetes applications; most vendor software ships as a chart.
    8. Markdown — Plain-text format for documentation that lives with code, diffs cleanly and moves between tools without conversion.
    9. Model Context Protocol — Open protocol for connecting AI agents to tools and data; now broadly supported across model vendors.
    10. OAuth 2.0 — Open standard for delegated API authorization; follow current best-practice profiles rather than older grant types.
    11. OpenAPI Specification — Open standard for describing HTTP APIs; contract-first design, generated clients and gateway import all depend on it.
    12. OpenID Connect — Open standard for sign-in; use it for every new application instead of custom auth or legacy protocols.
    13. OpenTelemetry — Open standard for traces, metrics and logs; instrument once and keep the freedom to change observability vendors.
    14. SCIM — Open standard for user provisioning; require it from SaaS vendors so joiners and leavers are automated.
    15. Terraform — Most widely supported infrastructure-as-code tool; note the BSL license and keep OpenTofu as the open option.

    Pilot · 7

    1. AsyncAPI — Open standard for documenting event interfaces; tooling is thinner than OpenAPI, so trial it on one event domain.
    2. Claude Code — Agentic coding that works across the repository; trial with clear review rules and measure delivery, not keystrokes.
    3. Cursor — AI-first editor with strong codebase context; trial it alongside other assistants and compare on real work.
    4. Dev Containers — Open spec for reproducible development environments; new joiners and CI get the same toolchain from one file.
    5. Fastlane — Still useful for scripting store releases, though hosted build services now cover much of what it did.
    6. Open Policy Agent — General-purpose policy engine; useful for admission control and authorization, but Rego has a learning curve.
    7. OpenAI Codex — Agentic coding from OpenAI in CLI and cloud forms; trial it where OpenAI models are already approved.

    Watch · 2

    1. Detox — End-to-end mobile testing that works but is flaky to maintain; compare with Maestro before committing.
    2. DuckLake — Lakehouse format that keeps metadata in a SQL database; promising, but Iceberg has the wider engine support.

    Retire · 1

    1. Jenkins — Plugin sprawl and self-managed controllers cost more than hosted CI; build new pipelines on GitHub Actions.