IThe unequal-literacy gap and the cost of inaction.
IIThe 24 SDLC personas and seven clusters.
IIIA five-stage onboarding framework.
IVSix-month roadmap and tool tracks.
VPitfalls, resistance, and KPIs.
VICenter of Excellence and what to do Monday.
Part I
I
The literacy gap.
Why most agentic AI rollouts stall and what the evidence says about fixing it.
Problem · Adoption gap
Teams with formal onboarding hit80%active adoption in 90 days. Teams without hit 23%.
3.5xgap between structured and unstructured rollouts
Source: GitHub Octoverse 2025. Active adoption defined as Copilot use on at least five workdays per week. The differentiator is not budget, not seniority, not stack. It is the program around the tool.
Problem · Why rollouts stall
Three patterns, three failure modes.
01 · Tool-only rollout
License first, learning later.
Provision Copilot Enterprise, send the welcome email, hope for the best. Active adoption peaks at week two and falls to 12% by day sixty (Forrester 2025).
02 · Engineer-only program
Devs trained, the rest left out.
QA never opens it. The PM does not see the connection. Security blocks the rollout for lack of clarity. ROI flattens because the SDLC is multi-persona.
03 · One-shot training
A workshop instead of a journey.
Three hours of slides, no follow-up. Kirkpatrick level 1 is satisfied (people liked it); level 3 (behavior change) never happens.
All three share a root: they treat onboarding as an event, not a program.
Part II
II
The 24 SDLC personas.
Seven clusters. One program that speaks each cluster's language.
Personas · Seven clusters
Twenty-four roles, organized for the program.
Personas · Tool matrix
Each cluster has a primary tool, a secondary, and a governance plane.
Engineering · Quality
Copilot Agent Mode
Primary: GitHub Copilot Agent Mode and Chat. Secondary: GitHub Actions. Governance: GitHub Advanced Security with Security Autofix.
Platform · Cloud
Copilot for IaC
Primary: Copilot for Bicep, Terraform, Bash, PowerShell. Secondary: Azure Developer CLI (azd). Governance: Defender for Cloud.
Product · Design
Copilot Studio
Primary: Microsoft Copilot Studio for backlog and feedback agents. Secondary: Microsoft 365 Copilot. Governance: Purview labels for research data.
Data · AI · Security
Foundry & GHAS
Primary: Azure AI Foundry for orchestration; GitHub Advanced Security for code. Secondary: Microsoft Fabric. Governance: Purview, Compliance Manager.
Part III
III
The five-stage framework.
From preparation to autonomy. ADKAR meets the four stages of competence.
Stages · Six-month flow
Five stages, twenty-four weeks, explicit exit criteria.
Stages 0-1 · Set up and frame
Lay the rails before the first prompt is typed.
Stage 0 · Preparation
Nine pre-requisites, no exceptions.
Publish the AI usage policy. Provision Copilot Enterprise, Foundry workspaces, Copilot Studio, GitHub Advanced Security, Defender for Cloud, Fabric, and Purview. Identify champions and put them through two weeks of advanced training before the broad kickoff. Define the program KPIs in writing.
Stage 1 · Awareness
Demos with the team's own backlog.
Frame the program as augmentation, not automation. MIT Sloan 2024 finds augmentation framing produces 2.8x the engagement of automation framing. Run 60 to 90 minute live demos, one per cluster, using a real artifact from the team's current sprint. Generic tutorials fail.
Stage 2 · First 72 hours
One concrete quick win per cluster, designed for the BJ Fogg habit window.
Engineering
Backlog item, end-to-end.
Use Copilot Agent Mode to take one low-complexity backlog item from branch to PR without leaving the editor.
QA · SRE
Full unit test suite.
Generate the unit-test suite for an uncovered class. Typically 15 minutes versus 3 to 4 hours manual.
Platform
CI/CD for a new service.
Generate the GitHub Actions pipeline for a new service, with the project config file as Copilot context.
Product · Data
Backlog Q&A agent.
Build a Copilot Studio agent that answers questions about the roadmap from a SharePoint sheet.
Security
Top-three CVE autofix.
Use GHAS plus Security Autofix to triage and propose fixes for the three highest-severity findings on a low-risk repo.
Data & AI
Notebook in 15 minutes.
Generate an exploratory analysis notebook for an existing dataset, with Copilot guiding feature checks.
Leadership
Retro summary.
Use Microsoft 365 Copilot to generate the executive summary of a recent retro from Teams or OneNote.
Champion
Pattern catalog v0.
Capture every quick win as a one-page pattern card. Becomes the seed of the prompt and recipe library.
Part IV
IV
Pitfalls, resistance, and KPIs.
What goes wrong, why it goes wrong, and how to know whether the program is working.
Pitfalls · Field evidence
Seven pitfalls account for most failed rollouts.
01One-shot training, no follow-up. Adoption falls to 12% in 60 days.
02Engineer-only program. Silos of adoption, half the ROI.
03No real use cases. Generic tutorials kill motivation in week one.
04No clear AI usage policy. CISO blocks; teams freeze.
05No internal champions. No social diffusion of practice.
06No or wrong metrics. Budget is cut at the first review.
07One-size-fits-all curriculum. High dropout, low relevance.
Sources: McKinsey Global Survey on AI 2024; Forrester AI Adoption Pitfalls 2025.
KPIs · What to measure
Six KPIs, measured every fortnight.
01 · Reach
Active user rate.
Percent of licensed users active at least five days per week. Target: 80% by day 90.
02 · Depth
Agentic actions per week.
Count of multi-step Agent Mode sessions per developer per week. Distinguishes autocomplete from orchestration.
03 · Quality
Acceptance & revert rate.
Suggestion acceptance rate, plus rate of rollback within 7 days. Together they reveal over-reliance.
04 · Throughput
DORA quartet.
Lead time, deployment frequency, change failure, MTTR. Lagging signals, but the ones that pay the bill.
05 · DevEx
Satisfaction & flow.
SPACE-aligned quarterly survey. Includes psychological safety question for skeptics and juniors.
06 · Risk
Security findings closed.
GHAS findings auto-fixed and merged per week. Defender for Cloud recommendations actioned.
Closing · Monday morning
Onboarding is not an event.Run it like a platform.
Three actions before the next sprint. Publish the AI usage policy in writing. Name two champions per team and book their two-week advanced training. Schedule the first cluster-specific demo using a real artifact from the current backlog. Then come back for Stage 2.