Munawar Shah
Hi! I’m Munawar. I’m passionate about building software that removes friction from our lives and gives us more time, agency, and control. I believe our data should belong to us, stay private, and be governed by us.
My approach to AI is: less is more. I believe AI is an incredible force multiplier when used with careful consideration to limit the surface area requiring review. It also has however a tendency to fill our field of vision with slop - something one must actively fight against.
Recently I've been finding ways to make AI-use feel less like pulling on a slot machine. I believe predictability is a prerequisite to improving outputs over time, and to this end context passed to an AI agent should be easy to inspect. More-so, the system should make it easy to predict beforehand what would land in the context .
If you’re curious about my work, try codedecorum or watch the flitterbot demo below - both MIT-licensed.
Writing
scroll back up-
Only a wrapper for GPT-4, but it takes 90% of the grunt work out of support ticket handling.
-
In the long view, Artificial Intelligence (AI) will affect the economy and create new winners and losers. What’s worth putting your chips into?
-
If it isn’t about what you know, then what is it about, and what should it be about?
-
I think there might just be significant potential in the custom solution space.
Demos
scroll back upTwo short walkthroughs of systems I am proud to have built.
-
Flitterbot
An orchestrator for Pi, Codex and Claude Code sessions that makes parallel AI work easier to direct, inspect, and resume.
3:16Transcript available for Flitterbot -
Schema-driven workflows
An event-driven workflow system where agents and scripts work against shared state while the business logic stays visible in one place.
3:49Transcript available for Schema-driven workflows