Hi. I am an engineer at Activepieces, an open source automation platform (think Zapier, but you can self host it and read the source code). I write...
For further actions, you may consider blocking this person and/or reporting abuse
Very Helpful!!
this is great. I was just reading about agent harness on langchain blog -- learning about it since the last week.
AI doesn’t magically understand your codebase. Tests, naming, architecture, and documentation are the real instruction layer.
I'm not at 1.9m LOC yet, more like 100k and one developer, but it's reached the point where I need to rethink my strategy, a simple guidelines.md has served me well but it's no longer enough.
Specialised subagents looks like a really good idea.
But there's lots of great advice here.
I’m really happy to see you back with new blogs! I love how you organize them—they’re both fun and clear, and the information really stands out. Big applause for the structure of the blog and even the title, AI Harness.
hit this with our biggest repo - suggestions were technically valid but wrong for our patterns. the fix was not prompt engineering, it was 3 days of context doc cleanup. repo hygiene as AI prep work is underrated.
Thank you very much now i have motivation to go and buy subscription.
I promised myself by the end if the day I would have AI subscription but I had no idea what I would do with it beside well writing same things I do now with free tier on the browser.
I intend to do creative learning so i must adjust the harness to do more planning than yours.
I have found that system design heavy prompt generate better and easier to use code and its easier to locate what need to be fixed for example if I say "decoupled and modular with self explanatory naming convention".
Great post! I just released a Claude Code plugin to do this: dev.to/tacoda/introducing-bridle-a...
Here's the repo (open to create issues to improve): github.com/tacoda/bridle
I like a lot of what you have in here. Would love to collaborate with you on this!
I feel like part of the struggle isn’t even technical. For a long time, we were made to feel like using AI is cheating or doing something wrong, so a lot of us held back. Now it’s more about unlearning that mindset and actually embracing it as a tool.
extremely detailed
Keep going 👏
And only as smart as the services you connect it to. That's why I built querybear.com. To connect your coding agents to your DB safely.
Indeed helpful
Tests are the secret sauce in the mix.
Too many treat them as an afterthought.