Stage 3.5

Operationalization

From certified agent to running program -- the organizational infrastructure for deployment, adoption, maintenance, and governance

The Problem This Solves

Organizations need to know how to run an AI program: who owns what, how work moves between stages, what happens when something breaks, and how to get stakeholders on board.

The operationalization layer closes that gap. It generates documentation the organization needs to deploy, adopt, maintain, and govern the agent -- using data the pipeline already collects. No new interviews. No extra data collection. One Claude API call produces four artifacts from structured pipeline data.

Working Agreement (Pre-Engagement Gate)

Before any build work begins, AESOP OS generates a Working Agreement from the completed discovery session. This is a signed document establishing:

Scope -- what will be delivered (6 artifacts from discovery brief through launch package)
Stakeholder map -- executive sponsor, project owner, affected groups
Autonomy classification -- fully autonomous, semi-autonomous, or read-only
Commitment readiness -- team capacity, response time SLA, change appetite
Response commitments -- who responds to what, how fast

The Working Agreement is a static markdown template with variable injection -- not LLM-generated. It appears in the sidebar between Discover and Design. Design is gated behind a signed agreement, with a skip option for internal projects.

How to Access

After Discovery

Complete a discovery interview. On the brief review page, click "Create Working Agreement" to generate the agreement from your session data.

Sign and Proceed

Review the agreement, download as markdown if needed, then click "Mark as Signed" to unlock the Design stage.

After Build Completes

On the Build page, a "Generate Launch Package" button appears after the build pipeline completes. Click it to generate all four artifacts in a single pass.

Review and Download

Each artifact can be expanded inline, downloaded as markdown, and marked as reviewed. Tiger Team findings can be pasted into the Repair page for the next repair cycle.

Data Sources (No New Collection)

Every artifact is generated from data already captured by the pipeline:

Discovery session -- stakeholder map, autonomy level, commitment readiness, problem statement, constraints, risks
Build summary (Phase 5) -- State Change Hypothesis, Translation Debt Assessment, Ripple Map, gap analysis findings
Evaluation output (optional) -- certification level, dimension scores, known weak areas

The Operationalization Agent receives this structured data and produces all four artifacts in a single API call (max 16K tokens output).

The Four Artifacts

Governance

RACI Matrix

Role-responsibility mapping across the full agent lifecycle, from discovery sign-off through decommission.

Covers: 10-15 activities including discovery, design approval, KB preparation, build review, evaluation, repair, launch, training, KB maintenance, issue triage, re-evaluation, drift response, decommission

Rule: Every activity has exactly one Accountable (A). Executive sponsor is A for GO/NO-GO, I for operations. Project owner is R or A for most operational tasks.

Rendered as: Markdown table (parsed to structured JSON for interactive rendering)

Stakeholder Alignment

Tiger Team Brief

Plain-language stakeholder briefing -- what changes, what we need, and a feedback mechanism.

Format: 5-minute read. Role-specific "what changes for you" scenarios. Concrete asks with time commitments. Thumbs up/down feedback.

Tone: Direct, respectful. No corporate jargon. Written for busy people who need to understand impact, not architecture.

Feedback loop: Tiger Team findings from stakeholder review can be pasted into the Repair page. The Repair agent treats them as real-world user testing data alongside evaluation findings.

Sustainability

Maintenance Plan

Evaluation cadence, KB review schedule, repair protocol, drift monitoring, and decommission criteria.

Evaluation cadence: Tied to autonomy level. Higher autonomy = more frequent evals. Includes triggers for unscheduled evaluation (score drop, KB change, process change).

Thresholds: References AESOP certification levels -- Platinum (90+), Gold (85+), Silver (80+), Bronze (75+). Defines what score drop triggers action.

Drift monitoring: Metrics to track, alert thresholds, escalation path.

Rollout

Adoption Plan

Phased rollout calibrated to commitment readiness scores, with measurable adoption metrics.

Phases: Pilot (subset of users), Expansion (broader group with gate criteria), Full Deployment (all users with handoff).

Calibration: If change appetite is low (1-2/5), recommends longer pilot and more hands-on support. High readiness enables faster rollout.

Metrics: Measurable adoption targets with specific measurement methods -- not just "monitor usage."

• Operationalization Pipeline

Data Sources

Stage 1

Discovery Session

Stakeholder map, autonomy level, commitment readiness, problem statement

Stage 3

Build Summary

State Change Hypothesis, Translation Debt, Ripple Map, gap analysis

Agent

Operationalization Agent

Single-pass generation of all four artifacts from structured pipeline data

Launch Package

RACI

Role matrix

Tiger Team

Stakeholder brief

Maintenance

Eval cadence

Adoption

Phased rollout

Feedback Loop

Tiger Team Findings to Repair

Stakeholder feedback feeds into the Repair agent alongside evaluation reports for the next repair cycle

Architecture principle: No new data collection. Every artifact is generated from structured data already captured during Discovery (stakeholder map, commitment readiness) and Build (State Change Hypothesis, Translation Debt, Ripple Map). The Operationalization Agent is a generation pass, not a new interview.