Tracing is the only way to debug evals and keep some sanity 🫠

Go is a great match for agentic systems. They're mainly about orchestrating API calls and concurrent operations, and it's so easy with goroutines 👌

I could easily think this is about webscale MongoDB, if not the years mentioned 😆

People will claim vibe coding frees you from writing the code, only to mention best practices, rules, and context management. Writing all the instructions and reiterating on errors can be just as tedious. Both have their place, but it's not easy to know beforehand.

I guess we'll never agree on this. If you’ve worked with overengineered crap, you’ll swear by keeping things simple. If tech debt made your life a nightmare, you’ll try to foresee it and react early. In the end, it's all about how you use the tools in context.

The catch-22 of any creative work: you need to produce to get better, but it never feels good enough to share.

When making games, there's no way to come up with a fun mechanic right away — you need to try things, play test, iterate, and throw away what doesn't work. That works great with AI-supported coding!

I'm making a game with Claude Code, and it's been fun so far. It feels a bit like writing a book with a spec that anyone could pick up and implement.

I'm having lots of fun writing "tasks" for Claude code while working on a game in Ebitengine. It's not bulletproof, of course, but I like writing down my ideas anyway. Previously, I would write them down in a notes app and gradually move them to code. Now, I can write a clear

"Just wants to build" is a weird take It was always about building, unless you solved code puzzles for fun. There's a whole spectrum between "just coding" and "describing features to an AI bot".

We've spent a few weeks this year building an LLM-based feature. Here's a new post about what we learned: threedots.tech/post/ai-agent-…

Case study of how to work in a real production environment with unreliable AI agents which can hallucinate. An interesting engineering challenge and there are some good lessons here with examples in #Golang. #AI #artificialintelligence #machineLearning #AIResearch #ML

Usual obstacle to learn new skills or libraries: invent a project where you can apply them. You no longer have this excuse for Watermill. Introducing Watermill Quickstart - our new free training. Learn all basic concepts required to build a simple event-driven application by

x.com/roblaszczak/st…

We launched Watermill Quickstart this week: a short hands-on training to get started with event-driven apps. watermill.io/learn/quicksta…

An AWS outage today brought down what seems to be half of the internet. Very often, your application can handle outages gracefully. Dima Kotik shows on our blog how you can use SQLite as a lightweight Pub/Sub with a focus on durable execution. threedots.tech/post/sqlite-du…

threedots.tech/post/common-an…

Durable Background Execution with Go and SQLite: threedots.tech/post/sqlite-du…