AI Operating System

A public-facing case study of Sararin's AI-enabled working method: governance, evidence, decision flow, routing discipline, and practical execution.

AIOS Case Study Map

These public pages group the current AIOS artifacts without crowding the top navigation.

Choose a path below to explore the AIOS case study.

Executive Summary

Start here

What this case study is, why it exists, and how the AIOS is framed.

Lean Value Tree

Strategy-to-execution logic connecting value, work, and proof.

How We Build

Delivery method, gates, review loops, and evidence discipline.

Org Roles

Role boundaries for humans, AI workers, review, and escalation.

Workstreams

Current operating lanes, status, owners, and next actions.

Agent Review Evidence Dashboard

Public evidence snapshot showing manual evidence, confidence, missing states, and explicit gaps.

AI Agent Launch Checklist

Portfolio framework for AI launch-readiness, human gates, evidence, and rollback criteria.

Architecture

System structure, boundaries, and public-safe technical view.

Evidence Discipline for AI-Assisted Delivery

In progress

An in-progress case study examining how to separate work-in-progress from claims that are ready for a public audience, using an ongoing surface recovery as the working example.

Work in progress. Shared while the underlying engagement is still active; not a final writeup.

Path to Production

A deterministic release path that binds a scoped candidate to independent validation, non-force promotion with preflight validation, exact deployment identity, live surface evidence, and a final claim gate.

Runner Gang

Execution: Runner Gang
Independent authority: Prime Gate
Candidate implementation: Big Crew (bounded)
Surface meaning: Surface Guild
Route/render proof: Surface Runner

Reusable runtime accepted. This Cockpit release is the production proof attempt; the page does not certify its own deployment.

What It Is

A layered system for managing AI-assisted work: task routing, AI workforce allocation, governance gates, evidence capture, and execution traceability. It answers three questions: What work needs doing? Who should do it? How do we know it was done correctly?

Public pages show proof and method. Private operating notes, budget/backlog context, and draft work stay unpublished until they are public-ready.

Recruiter-facing case study

AI Delivery Measurement & Governance

This project combines existing AI tools with customized governance, retrieval, review, and validation components. The value is not the number of tools used. It is the operating model that defines what each tool may do, what evidence should be retained, and where human judgment remains required.

What this demonstrates

Sararin can structure ambiguous AI-adoption problems, assign bounded responsibilities across tools, protect sensitive context, challenge weak evidence, and ship validated improvements without overstating system maturity.

Coordinated existing AI tools without confusing configuration with custom development.
Designed governance boundaries across source-of-truth, retrieval, validation, and human approval.
Built public-safe evidence surfaces while clearly labeling what is measured, missing, or not proven.
Preserved human approval where automation was not yet proven.

Existing tools selected and configured

These are third-party tools and access routes. They were selected, configured, tested, and assigned bounded responsibilities; they were not built as part of this project.

ChatGPT / GPT

Existing tool

Reasoning, synthesis, planning, and executive review support.

Project use: Configured with role boundaries, review gates, and source-of-truth rules.

Codex

Existing tool

Existing coding agent for bounded repository implementation.

Project use: Used with scoped tasks, stop conditions, validation steps, and Git discipline.

Claude Opus

Existing tool

External high-judgment reviewer for selected claims and architecture decisions.

Project use: Used as an independent skeptical critique step before relying on evidence.

Hermes Agent

Existing tool

Existing orchestration and stage-management tool.

Project use: Tested for profiles, routing continuity, handoffs, and context-access patterns.

OpenRouter / ChatX

Existing tool

Existing model-access routes.

Project use: Evaluated for provider routing, quota limits, cost visibility, and telemetry gaps.

Ollama

Existing tool

Existing local-model runtime.

Project use: Tested for bounded local execution and runtime evidence capture.

Systems designed and customized

These working methods adapt existing tools into a controlled delivery process.

GPT KB + Git

Active

Git-backed knowledge base and change history for reviewed decisions, evidence, and working context.

AIOS workflow

Working prototype

Human-gated operating model coordinating planning, implementation, critique, validation, and release decisions.

Evidence classification

Used in audit

Rules separating measured, derived, inferred, missing, invalid, and not-applicable evidence.

Validation gates

Active

Bounded checks covering typecheck, lint, build, route smoke tests, boundary scans, and explicit-path Git staging.

Tool-routing rules

Partially proven

Evolving guidance for matching tools and review depth to the task, risk, and evidence required.

Components built locally

These components were designed or implemented as part of the AIOS prototype.

Supernova

First version complete; bounded Stakeholder Core candidate.

Custom strategy and critique concept. Stakeholder Core is a canonical candidate, not an active-role claim; POC validation and R01 start are not yet authorized or proven.

Data Team

Evidence lens active

Supports evidence structure, aggregation logic, interpretation, and dashboard-readiness review. It does not act as final claim authority.

Curated AI-readable context layer

Implemented

Reviewed context subset that separates approved AI-readable material from broader working knowledge.

Custom read-only MCP server

Local smoke-tested prototype

Allowlisted context retrieval with file listing, approved-file reads, keyword search, and protections against broad or unsafe access.

Evidence Inventory Snapshot

Implemented

Public-safe historical measurement view with explicit caveats. It is not live telemetry.

Normalized receipt and reconciler

Future slice

Proposed evidence-capture layer for timing, route, cost, validation, and human-gate reconciliation.

Current maturity boundary

The custom MCP layer is a local smoke-tested read-only prototype, not a production RAG system. Vector search, embeddings, semantic retrieval, live telemetry, automated approval, and enterprise-scale outcome claims remain outside the proven scope.

View the evidence and observability appendix →

Architecture Summary

The AIOS public architecture is organized into four executive groups. The detailed map keeps governance, evidence, durable state, and business runway visible without turning this overview into a dashboard or operations manual.

Executive & Intent

Routing & Governance

Execution & Validation

Evidence, State & Business Runway

Evidence, Audit & Observability is one review layer. Phase 1 telemetry production and storage remain implemented in optimize-worker while AIOS owns governance meaning and evidence boundaries. Protected Internal Telemetry is a separate authenticated view, not a public or continuous observability platform.

View the full executive architecture map →

Interaction explainer

How the system works across layers

Step 01 of 08

Human Decision

Sararin defines intent, priority, risk appetite, and approval boundaries.

Manual governance support, not full automation.

Public page is a presentation surface, not source of truth.

GPT KB + Git remain durable truth records.

Step 1 of 8

AI Workforce

AI Workforce is the role-based execution capability within the AI Operating System. Each role has explicit responsibilities, boundaries, and operating relationships.

Explore AI Workforce & Org Roles →

Strategic Framework

The Lean Value Tree connects strategic goals with operational outcomes, from control to leverage to monetization.

North Star

Sustainable economic leverage through an AI-assisted personal operating system.

AIOS Capability Layer

This North Star is supported by measurement-grade evidence discipline, source-of-truth control, policy, governance, and Prime Gate / Gate PM review.

The goal is not blind automation. It is controlled delegation: routine validation, evidence checks, and escalation boundaries are progressively moved through the system under Prime Gate / Gate PM review, while strategic control stays with the owner.

AIOS reduces low-value decision load so the owner can operate more like a CEO — setting direction, approving material trade-offs, and focusing attention on strategy, judgment, and leverage.

Strategic Control

Active

Maintain decision authority and scope discipline across AI-assisted work

Execution Leverage

Active

Increase output per hour without losing review discipline

Workforce Efficiency

Drafted

Operate as a one-person system with AI leverage, before scaling

Portfolio Proof

Drafted

Demonstrate governance maturity through visible artifacts

Opportunity Intelligence

Planned

Detect money-creation triggers and route to right module

Monetization

Planned

Explore whether governance maturity can support sustainable revenue

Current Workstreams

Operating status for active work: owner, next action, gate decision, and proof level.

optimize-worker

Active

Owner: Sararin + Codex

Keep benchmark trace automation parked until a thin slice is chosen.

Proof: Documented

Fallback routing

AIOS Enforcement v0.2 canonical and AVAILABLE

Owner: Sararin

Keep v0.2.1 checker tooling additive under v0.2 and v0.3 candidate/inactive.

Proof: Smoke-proven

Profile positioning

Active

Owner: Sararin + Codex

Keep homepage focused on differentiation; avoid operations-dashboard drift.

Proof: Regression-gated

Supernova

First version complete; bounded Stakeholder Core candidate.

Owner: Sararin

Keep POC validation and R01 start unauthorized and unproven until their gates pass.

Proof: Documented

Data Team

Evidence lens active

Owner: Sararin + Data Team

Support evidence structure, aggregation logic, interpretation, and dashboard-readiness review without acting as final claim authority.

Proof: Documented

Big Crew

Started

Owner: Sararin + Codex

Keep Big Crew usage bounded to implementation and verification tasks with explicit handoff.

Proof: Documented

Researcher

Started

Owner: Sararin + Researcher

Use for bounded positioning and claim-safety briefs before public copy changes.

Proof: Documented

Why These Boundaries Exist

The AIOS design is shaped by operational lessons and governance rules learned through practice. These boundaries aren't theoretical — they exist because something broke, drifted, or became unmaintainable without them.

Human Review Gates

Every artifact passes through human review before being committed as truth. AI workers can draft, implement, and validate, but the human still owns the final gate.

Why: Early runs without this gate produced confident outputs with silent errors. The review gate catches logic gaps, context drift, and misaligned assumptions.

Capability Routing

Tasks are routed by capability, not by convenience. Repo edits go to Codex. Evidence scans go to Researcher. Architecture critique goes to Big Crew.

Why: Generic routing led to budget waste (expensive models doing cheap work) and quality gaps (cheap models attempting senior tasks).

Privacy Boundaries

Private raw context (decisions, finances, health, personal notes) stays in the committed knowledge base. Public-facing content is curated separately and reviewed for safety.

Why: Early attempts at auto-publishing from the KB leaked context that was fine for personal notes but not for external visibility.

Thin-Slice Delivery

Work is broken into thin slices with explicit validation gates between each slice. Build → test → commit → stop for review. No multi-stage runs without checkpoints.

Why: Long autonomous runs without checkpoints produced large diffs that were hard to review, debug, or rollback. Thin slices keep errors local and recovery cheap.

Budget Discipline

Mechanical work routes to cheap workers. Senior synthesis routes to expensive models. No delegation without a named decision, cost cap, and abort condition.

Why: Undisciplined routing burned budget on tasks that didn't need senior reasoning. Role-first routing treats AI budget like workforce salary.

Evidence Before Claims

Public-facing metrics require baseline, date, scope, caveat, and evidence source. No quantified claims without the full five-field standard.

Why: Early portfolio attempts included impressive-sounding numbers that lacked context. Reviewers flagged them as unsubstantiated. The evidence standard restores credibility.

“AI orchestration is program management. Tools change. Governance discipline doesn't.”

Component Glossary

Component	Definition
GPT KB + Git	Source of truth. Master archive of decisions.
GPT	Executive orchestrator and reviewer.
Hermes + Opus + custom MCP*	Context-aware operator surface — Active trial — structured context pull, task continuity, governance-aware operation.
optimize-worker	Execution routing and task packaging layer. In Phase 1, it also implements telemetry production and store behavior for observability evidence.
Codex	Bounded implementation executor for repo edits.
ai-os-profile	Curated external visibility layer and deployable cockpit/app surface (this site).
Evidence, Audit & Observability	Governance evidence layer for traces, routing decisions, cost/usage evidence, reliability signals, and deployment readiness checks. Not a claim that AWS app deployment is complete.

*Current operational trial surface. MCP is used as a read-only context bridge for controlled context retrieval. It is not yet approved as an autonomous production workflow.