The broken feedback loop of agentic coding (which we spent months to fix)

Following up on last post, I thought I'd share a bit more on why we created Voidleap Code. We experienced basically three levels of hell which triggered us to start this journey and build something better.

1. Observability hell

Most tools give you a limited view of agent actions. So when you try to understand what happened and why, you're basically looking on a black box. Adding to that, code diffs wasn't really easily consumed, rather hidden away. This first hell really was the first block of owning and controlling your agents and the outcome.

2. Configuration hell

Then when you actually find out how you want to configure your workflow, agents or skills - you enter the second hell where you need to control all markdowns for agents. You have different versions stored in each repository. Just finding them is troublesome enough. Then getting an overview of their configuration, easily editing, or just re-use them... It was nearly impossible to own the workflow.

3. Terminal hell

Scaling up to multiple running agents in parallel didn't work either. Each thread/session was a new terminal window, and we tend to keep them open even after they was completed, because we might need to continue from that context in a new turn. The solutions to organize terminals exists, but it wasn't enough. It wasn't easy enough to start work in a thread, pause it for hours/days and resurface it again.

So we built Voidleap Code.

Deep observability over black box magic.
Visual scaffolding compatible with existing markdown files.
Agentic development environment instead of terminals.

Was built in stealth. Now we're on the go to market journey.