The failures
Cost time, focus, and years of work. Some were badly positioned. Some were too early. Some became too complicated.
Founder of SpielOS · Agent Harness Architect
I spent more than eight years designing growth systems for startups.
Then I spent years building products and agent systems. More than 20 product attempts later, SpielOS is the one that finally brought the work together.
I built SpielOS for companies that know AI should become part of how they work, but do not yet have a reliable structure for roles, instructions, knowledge, tools, human decisions, and quality control. It is the product I needed while repeatedly designing these systems myself.
Eight years of startup systems before agent architecture.
Before SpielOS, I worked as a marketing strategist and Marketing Director for Iranian startups.
Those outcomes came from designing systems around people, decisions, workflows, customer behavior, and measurement.
The same systems thinking now applies to AI agents: define the workflow, set the boundaries, measure the output.
More than 20 product attempts across marketing, automation, AI, content systems, and agent orchestration. Most failed.
Cost time, focus, and years of work. Some were badly positioned. Some were too early. Some became too complicated.
Clarified how complicated products should be structured and where systems usually collapse.
I discovered that capable models were only one part. The real challenge was everything around them.
Full-stack development with agent orchestration frameworks.
Multi-agent orchestration with roles, memory, and handoffs.
Quality gates and human judgment in automated systems.
Capable models are only one part of the system. Reliable work also needs roles, boundaries, tools, context, memory, handoffs, evaluation, and human judgment.
Most companies know AI should be part of how they work. But they do not have a reliable structure for roles, instructions, knowledge, tools, human decisions, and quality control. SpielOS is the product I needed while repeatedly designing these systems myself.
SpielOS is the product my previous work was leading toward. It is not one project in a portfolio. It is the culmination.
I built SpielOS for companies that know AI should become part of how they work, but do not yet have a reliable structure for roles, instructions, knowledge, tools, human decisions, and quality control. It is the product I needed while repeatedly designing these systems myself.
A small selection of projects from the journey to SpielOS.
Open-source AI orchestration platform. File-based agent harness with roles, skills, context management, and long-horizon execution.
View on GitHub →The first prompt-cache audit and optimization tool for AI agents. Finds context waste, prompt cache misses, and hidden agent cost leaks.
View on GitHub →Earlier version of the SpielOS content engine. Markdown-driven marketing team with 8 roles and a 12-state pipeline.
View on GitHub →Experimental vibe coding platform. Multi-role agent orchestration on LangGraph with frontend generation and PocketBase backend.
View on GitHub →Markdown-based wiki with content pipeline.
View on GitHub →