What we have on
production right now.

Default for any new web/mobile/Node work. Strict mode on, no exceptions.

Default for marketing, content-heavy, and BFF-style apps.

Data, ML, and back-office services. We've moved off Flask.

Native iOS work. Composable architecture for non-trivial apps.

Native Android. Compose has won — we no longer ship XML layouts.

Strong fit when teams need one codebase + product velocity wins over native polish.

Picking it up for high-throughput services. Used selectively, not as a default.

Useful, but we still prefer explicit /api routes for testability and tracing.

The boring choice that keeps winning. RDS, Aurora, Supabase, or self-hosted.

Default cloud for greenfield. We'll meet you where you are otherwise.

For Next.js marketing/web product where DX outweighs cost.

Analytics workloads where Postgres OLAP starts to hurt.

Edge logic, AI gateways, light APIs. Quietly excellent.

We maintain it on existing engagements, but no new projects.

Default E2E. Replaces the Cypress estate we used to maintain.

Standard CI for every repo we ship.

Tracing, metrics, logs as a first-class concern from sprint one.

Used internally and on most engagements. Jira when the client mandates it.

Primary path for Ask Prometheas and new customer-facing assistants.

Fallback or fit-specific provider for selected agent and copilot work.

Component foundation for new web work. Owned, not vendored.

Vector + keyword + structured filters. Pure-vector RAG is no longer a default.

Every production LLM app ships with offline + online eval. No exceptions.

Trunk-based dev + flags. Long-lived branches retired across our practice.

We've found the operational cost rarely pays back outside the largest estates.

Modular monoliths first. Split when forced, not for fashion.