Deep-dive software architecture teardowns, AI engineering paradigms, and system design analysis — the kind of technical depth that takes real research, not a quick take. Every dispatch traces a real production pattern down to why it exists, not just what it does. From caching layers and API gateways to the ethics of the systems we build, this is engineering writing meant to be studied, not skimmed.
21+
Dispatches
Tech #021
Latest Issue
19 min
Avg Read
AI agents can execute commands you never explicitly wrote. A sandbox decides what those commands can touch—and where they can go.
Rajnish Kumar
29 September 2026 , Tuesday · 17 min read
A flag decides which path executes. It does not decide what got deployed, migrated, scheduled or left behind — and that gap is where production incidents live.
An idempotency key protects the client-facing boundary. It does nothing for the boundary that actually breaks — the one between your own database and the system you're calling next.
A retry is technically safe only when repeating the request is business-safe — and most systems retry first, ask that question never.
A 120-millisecond API response and a five-second page load can both be true at the same time — and the gap between them is exactly where almost every performance dashboard stops looking.
As agents gain access to hundreds of tools across dozens of MCP servers and APIs, the bottleneck stops being what an agent can do and becomes whether it can find, choose, and safely use the right capability at the right moment.
The next challenge for Model Context Protocol isn't what agents can do — it's how MCP infrastructure survives when millions of requests start arriving.
A progress bar reaching 100% only proves the browser finished sending bytes — it says nothing about whether the system on the other end has any reason to trust those bytes yet, and treating the two as the same event is where most document upload pipelines quietly go wrong.
How slow dependencies, retries, queues, and cascading failures bring healthy-looking distributed systems to their knees.
GET puts query data in the URL. POST puts it in the request body. HTTP QUERY introduces a safe and idempotent method designed for queries that need a request body.