Framework 04

Capabilities

Everything RSUA does, mapped against value path, Human Workbench design, implementation, and governed expansion.

Capabilities organized by AI-native operations phase

RSUA work follows the four phases of the Process Framework: Define, Convert, Install, and Govern. Each capability is named, scoped, and tied to an owner, output, value path, Human Workbench, and measurement gate. No vague retainers. No scope creep without explicit re-scoping.

Define

01

Workflow mapping

Map current-state workflow, decision points, exception paths, data sources, and ownership. Output: workflow map and friction inventory.

02

Operating readiness screen

Score the operating environment across data, process, ownership, and governance. Output: readiness score, blockers, gates, and first-workflow evidence plan.

03

First-workflow selection

Rank candidate AI-native workflows by value, feasibility, and risk. Output: first-workflow sequence, value lever, ROI test plan, and evidence needed before build.

Convert

01

System architecture

Choose the AI pattern, tool stack, and integration approach for the converted workflow. Output: phase-one system spec with build, buy, or partner calls per component.

02

AI-native workflow design

Define confidence thresholds, escalation paths, reviewer roles, audit requirements, correction capture, and rollback boundaries. Output: human-role and control plan.

03

Human Workbench design

Specify what the accountable person sees before deciding: evidence, confidence, risk, validation errors, allowed actions, reviewer burden, and what each decision does next. Output: Human Workbench spec.

04

ROI and capacity modeling

Translate the AI workflow into measurable business metrics with named owners. Output: value model, payback hypothesis, and gating criteria.

Install

01

AI workflow system implementation

Implementation of agents, tools, and integrations against the converted workflow. Output: deployed system with evals and acceptance gates.

02

Guardrail implementation

All five layers: input, retrieval, generation, action, audit. Output: documented control set and red-team report.

03

Shadow-mode rollout

Run the system in production parallel without affecting live operations. Output: accepted-output report, reliability readout, and edge-case inventory.

Govern

01

Continuous evaluation

Recurring eval runs, drift detection, and threshold tuning. Output: monitoring cadence and performance readout.

02

Staged autonomy expansion

Reduce human review on workflow segments where reliability has been demonstrated. Output: updated control matrix.

03

Workflow extension

Add new tools, data sources, or adjacent workflows to the existing system. Output: incremental release notes and updated runbook.