@ahmetcantryk
6 min read
LLM Structured Outputs: JSON You Can Trust
Field notes on getting reliable JSON from an LLM: schema modes vs tool-calling, Zod/Pydantic validation at the boundary, and the failure modes that still bite.
Tag
Field notes on getting reliable JSON from an LLM: schema modes vs tool-calling, Zod/Pydantic validation at the boundary, and the failure modes that still bite.
Four levers — prompt caching, model routing, batch APIs, and context hygiene — stack to cut real LLM bills 50–70%, with a one-page cost-audit template.
Most agent failures on long tasks aren't model failures — they're context failures. Seven patterns, a context budget, and a checklist to fix an existing agent.
TypeScript 7 rewrites the whole compiler in Go for a ~10x speedup with the same type system. Project Corsa, RC status, and a safe way to adopt it today.
A category-by-category roundup of 12 open-source, self-hostable replacements for Notion, Airtable, Datadog, Vercel, 1Password, and Google Photos in 2026.
The 10 Kubernetes mistakes we still see wreck production clusters in 2026, from missing resource requests to plaintext secrets, each with the exact manifest that fixes it.
Write unit tests that catch real bugs: name for behavior, use Arrange-Act-Assert, mock only boundaries, and skip brittle implementation-detail tests.
With AI writing half the code, clean code principles matter more, not less. A checklist to audit naming, function size, comments, and tests in 30 seconds per PR.
Docker best practices for production: a practical, 2026-current checklist covering multi-stage builds, non-root users, image scanning, health checks, and resource limits.