J Jarvis collaboration + learning loop
WorkSession WS-1042 bounded policy created

Human-agent collaboration and learning loop

Jarvis

Human-agent collaboration and learning-loop protocol. Watch one HumanWorker and AgentWorker move through shared goals, policy, autonomous execution, request approval, review, contribution, evidence, and governed learning.

Current phase

Start the WorkSession

01 / 09
HumanWorker goals + judgment review, approve, teach
AgentWorker autonomous worker plan, execute, learn
Jarvis protocol WorkSession policy requests review evidence
Host executes outside Jarvis UI storage tools sandbox
Request record human gate approve narrow deny takeover

Jarvis creates the collaboration record before the agent starts.

    01

    WorkSession is the surface

    Every serious interaction has a durable collaboration record: goals, events, requests, reviews, contributions, evidence, and learning.

    02

    Policy creates autonomy

    The AgentWorker works inside human-defined policy. Outside policy, it creates a Request instead of silently acting.

    03

    Memory is governed

    Human learning, agent learning, and pair learning become governed proposals with policy, provenance, and review.

    04

    Evidence travels with work

    Jarvis records what happened as it happens. EvidenceManifest gives the human worker and agent worker a portable evidence record.