Blog

Notes on backend, cloud and AI engineering from real projects.

Articles: 29
Topics: 14
Latest: 2026.04.01

Multi-Agent AI in Production: Architecture Patterns for Enterprise Scale

AI agents are moving from demos to production. This post covers the architecture patterns, failure modes, and operational realities of running multi-agent systems at enterprise scale. Covers orchestrator-worker patterns, tool use, state management, observability, cost control, and a real production pipeline example.

Read article2026.04.01 · 10 min read

29 articles

2026.03.288 min readCloud & Serverless ArchitectureFinOps for Architects: Controlling Cloud Spend Before It Controls Your BusinessCloud costs are the second largest engineering conversation in most organizations after delivery speed. Architects who understand FinOps can design infrastructure that performs well and stays cost-efficient at scale. Covers tagging strategy, right-sizing, Kubernetes cost optimization, reserved capacity planning, and the tooling that gives you real visibility.
2026.03.159 min readSoftware ArchitecturePlatform Engineering in 2026: Building an Internal Developer Platform Your Teams Will Actually UsePlatform engineering has moved from a conference buzzword to a genuine competitive advantage. Companies that build good Internal Developer Platforms ship faster, with fewer incidents, and at lower operational cost. Covers the architecture, tooling choices, and organizational patterns that make IDPs work in practice, including Backstage, Terraform, and golden path templates.
2026.02.158 min readKubernetes & InfrastructureNative Sidecar Containers in Kubernetes: Production LessonsNative sidecar containers (alpha in 1.28, on by default in 1.29, stable in 1.33) replace years of init container workarounds. This changes how we handle service mesh, logging agents, and security proxies in production clusters. A practical guide from 17 years of platform engineering.
2026.01.207 min readPython & BackendMigrating a FastAPI Service to Pydantic V2: Lessons from a 50-Endpoint APIMigrating a production FastAPI service from Pydantic V1 to V2 surfaced gotchas that only show up under real traffic: stricter type coercion, model_dump() output that isn't JSON-safe, and changed validator semantics. Practical lessons on performance and a safe migration path.