Invoker

Hermetic AI Execution Workflow Orchestrator

Visit Website
May 7, 2026 A Slog

I started Invoker 2 months ago and right now and released it 3 weeks ago for an alpha.

During that time it has been quite annoying to go down every rabbit hole of technical debt/scale. I did not design Invoker to scale this hard. I thought it would be a simple task DAG or a build system monolith.

Increasingly I'm finding I have to build a lot of backpressure buffers and reuse many concept from DDIA.

Had I designed Invoker for scale to begin with, I could have saved myself 2-3 weeks.

Ontop of that, I am finding the auxillary infrastructure just as important as the core. For instance, after planning, you need to be able to generate a plan that is broken up, recoverable, and legible when you come back to it. In that regard, we are also stumbling into the problem of compressing human judgment.

I will fix all the obvious bugs I can for the core execution and then take a step back on the problem of human judgment. Ideally I'd like for some developers to use this and try and give me some feedback. But until then, I'm okay with developing this as it is.

Comment

About

Everyone is firing slop cannons blindly, leading to chaos and an inability to recover from errors. AI must be hermetic, legible, and controllable. https://medium.com/@edbertchantech/invoker-more-control-not-more-agents-