Writing21
Notes from the work: agentic AI, backend patterns, performance, and the tools I use every day. Articles are published on Medium and in Towards AI; each link opens there.
Latest
Claude’s new message threads let an agent send the conversation once and then only the new part.
New models are smarter than ever, yet most of an app's calls don't need that. A case against reaching for an LLM by default.
A plain answer to a question many teams are stuck on: one agent, or a team of them? Why Anthropic now says one agent is enough.
An AI engineer roadmap built from nine real job ads: what employers actually ask for, what to learn first, and what you can skip.
What an agent harness is, where the term came from, and why the same model can do much better or much worse depending on it.
A step-by-step guide to building an autonomous coding agent that reads files, edits code, runs commands, and searches the web.
An engineer's read of OpenAI's post on scaling a single PostgreSQL instance to support 800 million ChatGPT users.
How to choose an embedding model, prompted by a conversation about picking one for the legal sector.
Every web application faces the primary key question. Why random IDs hurt database performance, and what to do about it.
When a critical production record is changed or gone, you need its history. Building audit logs with Prisma, Postgres, and NestJS.
Connecting an LLM to more and more tools feels like progress. Why it can make an AI agent perform worse.
How to handle transactions in enterprise backend APIs, with best practices for NestJS and Prisma.
A .NET test app and k6 load tests comparing In-Memory Cache, Hybrid Cache, and Fusion Cache on cache stampede, system overload, and data consistency.
Lessons from a project built heavily on AWS Lambda: as the serverless architecture grew more complex, we looked at whether secret caching really works.
Indexing has a large effect on query performance. A guide to the PostgreSQL index types and when each one helps.
The story behind meem, a side project for managing digital clutter with a simple, visual approach to bookmarks and notes.
A comparison of two popular development workflows, trunk-based development and GitFlow, from years of working with both.
Big O notation applied to everyday C# code: the time complexity of common List and LINQ methods.
Validating data matters more as an application grows. How to use Joi for data validation in Node.js.
Sorting items by their distance from a given location, a common task with location-based data, in TypeScript.
Six practical ways to make a React application faster.