Write the change-management pack for three real personas — a burned skeptic holding legitimate reliability evidence, a threatened senior whose expertise feels devalued, and an over-enthusiast shipping unreviewed AI output. Three 1:1 scripts, one team-wide message, and the policy changes each persona's feedback should actually trigger.
Tomas Lindqvist is an EM at Verdant Grid, a fictional 40-engineer clean-energy software company mid-rollout. His adoption dashboard looks fine; his skip-levels do not. Three conversations from the past week convinced him the rollout's real risks are human, not technical — and that each of the three people involved is telling him something true. The standard change-management failure is treating resistance as an attitude problem to be messaged away. The bar in this lab is higher: every persona's underlying signal must produce a policy response, not just a persuasive script — because two of the three are right about something, and the third is a liability wearing an advocate's face.
PERSONA 1 | "Burned skeptic" - Ingrid, senior backend, 11 yrs exp
From skip-level notes: "I logged every AI suggestion I took for
two weeks in March. Four of nineteen introduced subtle bugs; one
hit staging before we caught it - the null-handling one. I have
the spreadsheet. Nobody asked to see it. Then the kickoff email
told us the tools 'boost quality.' I'm not anti-AI. I'm anti
being told my own eyes are wrong."
Behavior: uses tools minimally, correctly. Votes no in retros.
Influence: high - three mid-levels quietly follow her lead.
PERSONA 2 | "Threatened senior" - Rob, staff engineer, 16 yrs exp
From 1:1: "I spent fifteen years getting fast at exactly the
things this tool now does in seconds. Juniors ship in an
afternoon what used to take me a day. What exactly am I senior
AT anymore? Maybe I should just wait this wave out."
Behavior: disengaging from reviews; stopped mentoring; two
recruiters' InMails open in another tab.
Influence: high - he trained half the team.
PERSONA 3 | "Over-enthusiast" - Kai, mid-level, 4 yrs exp
From review-queue data: 31 PRs last month (team median 12), review
turnaround on his PRs now 3x team median, two reverts, one
incident postmortem naming his unverified AI-generated migration
script. Quote from retro: "Velocity is the point, right? The AI
writes better code than most of us. Reviewers are the bottleneck."
Influence: rising - two juniors have started copying his style.
Diagnose before drafting — ten minutes, no AI. For each persona write two lines: what they are right about, and what they are wrong about. This split drives everything: Ingrid is right about the evidence and wrong that it was ignored deliberately; Rob is right that his role is changing and wrong that it is disappearing; Kai is right that AI raises velocity and wrong about literally everything else. If you cannot fill in the "right about" line for a persona, you are about to write propaganda, not change management.
Draft Ingrid's 1:1 script — evidence first. Structure: open by asking for the spreadsheet (and mean it — schedule the review in the meeting); acknowledge the kickoff email's "boosts quality" claim overstated what the org had measured; commit her data into the measurement panel (Lab 08) and the review standards' risk-tier logic (Lab 05); and make the ask — not "get on board," but "run the verification workstream." The script must contain zero sentences that manage her feelings about being right; validate the evidence, then hire it.
Draft Rob's 1:1 script — the skill path. Structure: name the real shift without euphemism (pattern-recall and boilerplate speed are depreciating; judgment, system design, review of machine output, and teaching are appreciating); show him where seniority now bites — he is precisely the person who can tell plausible AI output from correct output, which is the scarcest skill in the building; give him a concrete path with stakes: co-author the review standards, own the AI-code review guild, mentor on verification. End with the honest line: the wave is not waitable-out, and you want him surfing it from the front. Follow with a written 90-day skill-path outline.
Draft Kai's 1:1 script — guardrails without humiliation. Structure: start with genuine credit (his energy is an asset; his instinct that the org must speed up is correct); then the data, framed as impact, not accusation (3x review turnaround = he is spending the team's time; the migration incident = the risk is real, not theoretical); the boundary (review standards apply to him fully — author-explains-every-line, provenance labels, no unverified merges); and the redirect — channel the enthusiasm into a legitimate fast lane (sandbox experiments, the eval battery, demo slots at office hours). No punishment framing: the goal is converting a liability into a power user, and the two juniors copying him are the real audience.
Write the team-wide message. One message that speaks to all three constituencies without naming anyone: what the org has actually measured so far (honest, including the March-class failures); the review standards now in force and why; the invitation for evidence in both directions (failure logs are as welcome as wins — cite the new verification workstream); and where judgment and mentorship sit in the org's future (the Rob constituency reads this paragraph). One page maximum.
Extract the policy changes — the graded core. For each persona, write the systemic response their signal should trigger: from Ingrid — a standing verification workstream, suggestion-defect sampling in the metrics panel, and a correction norm for overclaiming comms; from Rob — review-standards co-authorship, a recognized AI-code-reviewer role in the career ladder, and mentorship explicitly valued in calibration; from Kai — enforcement teeth in the review standards (unverified-merge consequences), per-author review-burden visibility, and a sanctioned experimentation sandbox so velocity energy has a safe outlet. If a persona's section contains only conversation and no policy, the lab is incomplete.
Role-play the hard one. Have the AI play Ingrid: "You are a senior engineer with a defect spreadsheet the org ignored. I am your EM. Push back hard on my script." Run three exchanges. If the AI-Ingrid stays angry, your script is validating feelings instead of evidence — revise. Repeat for whichever other script felt easiest to write; easy scripts are usually the ones that would fail live.
Assemble and sequence. Final pack: change plan (the three-persona diagnosis + policy changes), three scripts, team message, and a sequencing note — Ingrid's 1:1 happens before the team message (her evidence must be in the panel before the org is told the panel is honest), Rob's before any ladder announcement, Kai's before the next sprint starts.
In most rollouts the loudest resistance carries the best data and the loudest advocacy carries the worst habits — a comms pack that flatters adopters and manages skeptics has it exactly backwards.
A complete lab has all of the following:
Resistance is data. The skeptic's spreadsheet, the senior's identity question, the enthusiast's review-queue trail — each is a system signal wearing a personality. Scripts open the conversation; only policy changes end it.