The hard part of global IoT is rarely first connectivity alone. It is linking eSIM remote provisioning, device registration, policy delivery, acknowledgements, and diagnostics into one operating loop. This article explains why SGP.32 and LwM2M work better together.
Author: Zed IoT
The Core Architecture of an IoT Device Management Platform
Many IoT teams build device management as device registration plus online status plus a detail page. That works for demos, but it breaks under fleet operations, command tracking, version control, and troubleshooting. This article lays out a safer five-part architecture: registry, state, command plane, fleet index, and ops console.
How to Design a Local-First Smart Home with Home Assistant
A local-first Home Assistant architecture is not the same as trying to remove every cloud service. The stronger pattern is to keep device control, critical automations, state coordination, and recovery paths local while treating cloud services as optional enhancement layers.
The Hard Part of Multimodal Edge Systems Is Latency, Sync, and Operations
In multimodal edge systems, the hardest part is rarely whether a model can run. It is whether voice, video, and event streams stay aligned, low-latency, diagnosable, and recoverable under real hardware and real networks. This article offers a more practical decision framework.
Why Edge AI Fails More Often on Observability Than on Model Accuracy
Edge AI deployments rarely fail first on model accuracy. They fail when teams cannot see input health, inference health, version context, or diagnostic evidence. This article explains why observability should be designed as a core Edge AI capability from ESP32-class devices to Linux edge boxes.
Brownfield to Cloud: A Practical Path for Legacy Industrial Equipment
Legacy industrial equipment projects usually fail when teams push PLCs, meters, and serial devices straight into the cloud without a stable edge boundary. This article outlines a safer brownfield-to-cloud path built around asset inventory, edge normalization, reliable uplink, and controlled write-back.
How to Separate Model, Firmware, and Config Versioning in Edge AI
Edge AI fleets become hard to operate when firmware, model, and config are hidden behind one bundle version. This article explains how to separate those version planes so rollout, rollback, and troubleshooting stay controllable.
The Hard Part of Global IoT Is Not Connectivity, but Lifecycle Control
Global IoT deployments rarely fail because devices cannot connect. They fail because eSIM provisioning, device identity, regional policy, config versions, and operations feedback are not controlled as one lifecycle system. This article explains why lifecycle control matters more than connectivity alone.
Layering A2A, MCP, OPC UA, and Modbus for Agentic IoT
Agentic IoT becomes fragile when A2A, MCP, OPC UA, and Modbus are treated as interchangeable layers. A more stable architecture uses A2A for agent coordination, MCP for controlled tool access, OPC UA for asset semantics, and Modbus for field execution.
Why Industrial Edge Gateways Need Store-and-Forward
Industrial edge gateways that only forward data usually lose control of buffering, replay order, duplicate writes, and acknowledgment recovery during weak network conditions. This article shows how a practical store-and-forward design should work.