AI Employees and Decision Systems
Research on when role-specific AI agents should act, ask for help or stop, and how JARVIS and System Brain can coordinate accountable work.
When should an AI employee decide, ask or stop?
Our research frames DECIDE as action within an explicit role, adequate evidence and granted authority. ASK requests missing evidence or a required approval. STOP blocks a prohibited action. The decision study tests these recommendations on synthetic business scenarios; the JARVIS report documents the proposed architecture.
Read the dataset methodEditorial explanation by ZEISMO Research & Lab. Published 29 September 2026. Follow the source for its version and evidence scope.
How can an AI employee make useful decisions while remaining inside explicit authority, evidence and safety boundaries?
ZEISMO develops this programme through versioned reports, measurable evidence, reproducible methods and clearly labelled study stages.
Read across the programme.
JARVIS & System Brain
Open this resource within the AI Employees and Decision Systems research programme.
Building a Decision Dataset for AI Employees
Open this resource within the AI Employees and Decision Systems research programme.
When Should an AI Employee Decide, Ask for Help, or Stop?
Open this resource within the AI Employees and Decision Systems research programme.
When Should an AI Agent Forget?
Open this resource within the AI Employees and Decision Systems research programme.
After the Last Human: Could AI Keep Working?
Open this resource within the AI Employees and Decision Systems research programme.
After the Restart: memory, output contracts and safe resumption
Open this resource within the AI Employees and Decision Systems research programme.
Research vocabulary
From executable code
to coordinated decisions.
Led by Engineer Mohanad Osama Zahran, the programme investigates how an AI system can combine specialised skills, contextual evidence and memory to carry out increasingly capable roles.
JARVIS
The proposed English-speaking interface turns a request into a defined task and communicates the system’s response, uncertainty and next step.
System Brain
The research architecture coordinates role-specific agents, evidence and permissions. It separates a proposed decision from authority to execute it.
Skills and cells
Each proposed software cell combines a role contract, curated skills and scoped tools. Cells represent software responsibilities, not biological neurons.
Memory with evidence
Proposed memories retain their source, time and review state. Candidate lessons are tested before promotion into a role’s working procedures.
What would a hundred-specialist system need to prove?
Our ambition is to coordinate complementary specialist capabilities and exceed human performance on defined tasks. The research must measure decision quality, completion time, cost, disagreement and unsafe actions against a specified human comparison under equivalent conditions.
“100 times smarter than a human” is a research ambition, not a measured result. The published JARVIS report examines a prototype and controlled simulations; the follow-on decision study defines 48 synthetic scenarios. Neither establishes a general human-intelligence multiplier.
Explore the decision datasetFrom routine work to AI employees
We investigate the phased replacement of suitable routine tasks with role-specific AI employees. Each expansion needs evidence about task quality, approval boundaries, exception handling and recovery.
Context before action
In this programme, “feel” describes appraisal of operational context: changing conditions, incomplete evidence, risk and consequences. It does not establish consciousness or subjective experience.
Evidence you can inspect.
Follow each marker to the method, record or policy behind it.
ZEISMO practice markers. Verification scope is recorded with each publication.