AI agents are powerful but often unreliable in multi-step tasks. They drift from instructions, skip validations, expand scope, or self-declare success without evidence.
This white paper introduces a practical control layer inspired by classic computing:
- ENIAC-mode: Fixed, linear, “wired” procedures for repeatable tasks
- IAS-mode: Stored-program execution with branching, flags, recovery, and explicit program counter
It proposes turning loose prompt-based skills into strict, verifiable state-transition protocols:
input state → operation → output state → verification gate → next state
Includes:
- A concrete Strict Skill Schema (YAML-friendly)
- External validators & execution kernel design
- Real examples (SQL debugging, VBA review)
- The new role of Agent Skill Engineer
Especially relevant today: a timely opportunity for developers to build reliable agent workflows while creating enterprise-grade automation.
→ Full (AI generated) white paper in repo
ENIAC/IAS-Style State-Transition Protocols for Reliable AI Agent Execution
https://osf.io/q8egv/files/osfstorage/6a20b02ef378e08fb9a94d5a
Feedback welcome — especially from those building production agents, LangGraph/CrewAI users, or enterprise automation teams.
Let’s make AI agents not just smarter, but procedurally stable.

