Skip to main content

From Doing Agile to Being Agile

McKinsey Rewired — Chapter 13

Based on McKinsey Rewired Chapter 13: "From doing agile to being agile" — E. Lamarre, K. Smaje, R. Zemmel (Wiley, 2023)

TL;DR — Being Agile, Not Just Doing Agile

ADLC automates agile ceremonies with 26 AI agents operating under constitutional governance. Layer 1 (bash scripts) collects data in 5 seconds. Layer 2 (Claude agents) synthesizes recommendations autonomously. The result: decisions in minutes, not days.

McKinsey identifies the gap between "doing agile" and "being agile": most enterprises adopt the ceremonies and call that transformation. Being agile means agents exercise autonomous judgment within constitutional guardrails — reprioritizing work from evidence, not just reporting status. The ADLC framework closes that gap by making AI agents first-class ceremony participants, not just automation scripts.

Why Being Agile Matters​

The bottleneck in the digital, cloud, data and AI age is not automation — it is decision quality at speed. Three markers distinguish "being agile" from "doing agile" (McKinsey Rewired, Chapter 13):

  1. Autonomous decisions within guardrails — ADLC agents (26 specialists) operate under constitutional governance with 35 enforcement hooks, not manual approval for every action.
  2. Evidence-based reprioritization — every ceremony reads real data (CSV, SQLite, JIRA, git log) and produces scored recommendations, compressing the OODA loop from days to minutes.
  3. Embedded learning loops — sprint retrospectives write git-tracked evidence that future ceremonies consume, so the system improves across sprints without human memory.

ADLC ceremonies are not process theater. They are the operating system that turns a single manager and 26 AI agents into a product team that outcompetes in the age of digital and AI.

Sprint Model​

1-Week Sprint Cadence

Product: xOps (entire digital platform). Sub-products: CloudOps, FinOps, DevOps, ADLC — components within xOps, not separate product lines. Sprint: 1-week time-box. ONE active sprint for the entire product (xOps-S{N}). Rationale: AI-augmented pod (1 HITL + 26 agents) delivers at 3-5x velocity. Scrum Guide 2020: sprints are "one month or less" — 1 week is the right cadence for rapid PDCA. Boards: SPM (product development, Launch Pod) + OPS (ITSM operations). Sub-products are JIRA Components (CO, FO, TF, AD, XC), not boards.

Agile Cadence​

The following cadence maps McKinsey Exhibit 13.2 to ADLC ceremony automation.

Agile Cadence — Exhibit 13.2

Ceremony-to-Component Mapping​

Every McKinsey ceremony has a corresponding ADLC automation layer:

WhatWhy It MattersCadence
Daily StandupTeam health + blocker visibilityDaily
Sprint ReviewWorking product demo + stakeholder feedbackWeekly
QBRVelocity trends + OKR progress + strategic directionQuarterly

2-Layer Ceremony Model​

The ADLC framework implements a 2-layer model. Layer 1 (scripts) is "doing agile" — deterministic data collection. Layer 2 (agent synthesis) is "being agile" — autonomous judgment that recommends, not just reports. Claude commands and skills are the primary operating mode; Taskfile provides the offline fallback:

Ceremony Layers — Exhibit 13.4

Layer 1 -- Deterministic Collection (bash scripts, zero LLM):

  • collect-daily-standup.sh reads CSV, SQLite, coordination logs
  • collect-sprint-review.sh enumerates demo items, DORA actuals
  • collect-sprint-retro.sh gathers story completion, agent consensus
  • Output: raw JSON in tmp/<project>/ceremonies/

Layer 2 -- Agent Synthesis (observability-engineer):

  • Renders terminal template from raw JSON
  • Generates 4L retrospective from sprint patterns
  • Writes git-tracked evidence to framework/ceremonies/
  • Produces HITL email deliverable (email.md)

Quarterly Business Review (QBR)​

The QBR aggregates sprint-level data into quarterly trends for leadership.

QBR Implementation

QBR inputs:

  • DORA metrics trend (sprint-over-sprint from dora.csv)
  • Story velocity (completed stories per sprint from stories.csv)
  • OKR key result progress (from team mission JSON)
  • Cost variance (from FinOps evidence)

ADLC Component Architecture​

The full stack powering each ceremony:

Component TypeCountExamples
Agents2observability-engineer (execution), product-owner (planning)
Commands6/ceremony:standup, plan, review, retro, qbr, /metrics:update-dora
Collector Scripts3collect-daily-standup.sh, collect-sprint-review.sh, collect-sprint-retro.sh
Skills2session-retrospective-protocol.md, production-ready-scoring.md
Hooks1enforce-coordination.sh (exempt for ceremonies)
Evidence Paths3tmp/, framework/ceremonies/, framework/retrospectives/

Daily Workflow​

Governance​

Governance

All ceremonies are exempt from PO+CA coordination requirements (they read existing data -- no architecture decisions). However, ceremony commands must delegate to the designated agent per the Ceremony Delegation table.

RuleEnforcement
Delegation requiredagent: frontmatter in command file
Coordination exemptcoordination: exempt in command file
Sequential scoringRules-layer (no parallel re-scoring)
Evidence dual-writeJSON (agent) + MD (HITL) required

From Process to Competitive Advantage​

The shift from doing agile to being agile is not incremental — it is a step change in how fast an organization learns. When ceremonies produce autonomous recommendations instead of status reports, the time from data to decision compresses from days to minutes.

Get Started

Explore the full ADLC framework at adlc.oceansoft.io. Run your first ceremony: task ceremony:standup (5 seconds, zero config).

References​