Driving down the gap between what I want coding agents to do vs what they actually do.

So I think it’s pretty evident now we can offload a lot of work we used to do as software engineers to coding agents. πŸ€–

It’s a reality which took some time, but with every big change in our lives, it’s becoming the norm now.

Everything which was wrong with this new change is still there, you know, for example, things like code review fatigue, as engineers not knowing the inside out of the codebase, cybersecurity, and all these problems we have to keep tackling as offloading code entirely to coding agents becomes normal day to day. πŸ”’

And there is a lot of experimentation going on right now, everybody can build very easily, so we build a lot of things which we didn’t before because of all the work we needed to do. πŸ§ͺ

I am building software every day in and out for over a decade now, and recently, one thing that really pops up a lot in my workflow is just managing these agents and driving down the rate of bugs and errors we face when we go into production.

So building a system around what coding agents outputs has become a really big chunk of what I do every day. βš™οΈ

So I have been using orch to exactly do that, and because it’s open source, I am trying to improve it as I am using more and more of it in my daily workflow and realising the gaps and improvements I can make to the system.

Orch helps me build a system around the coding agents which can help me drive down the gap between what I want coding agents to do vs what they actually do.

What systems, processes are you using to solve the same problem? πŸ‘‡