LLaManchaAI Enablement
Checking access…
← All modules
Module 7Audience: manager

Team adoption and working agreement

Helps teams decide what to standardize, what to prohibit, and how to run a 30-day adoption pilot.

Outcomes

What you will be able to do

  • Pick pilot workflows that produce a clear adoption or failure signal.
  • Write a one-page team working agreement.
  • Coach team members at different AI fluency levels.
  • Measure adoption by outcome, not by usage, and run a 30-day retro.
Completion check

How this module is approved

Submit a one-page team working agreement and a 30-day pilot plan with a baseline, an outcome signal, and an expand/adjust/stop rule.

Pass criteria

  • • One bounded pilot workflow named
  • • Week-1 baseline captured
  • • Outcome signal with a target (not a usage count)
  • • Safety counter-metric present
  • • Standardize/prohibit/review lines are concrete
  • • Expand/adjust/stop decision rule stated
Loading personalization context…
Lesson35–45 minutes self-guided, or 60 minutes with a team planning session

What you should take away

By the end of this module, as a manager you will have a one-page team working agreement and a realistic 30-day pilot plan that tells you whether AI is actually helping your team — not just whether they are using it.

Prerequisites: Complete Module 3 · Complete Module 5

Part 1

Two pilots, one team — only one produces a decision

Two managers each run a 30-day AI pilot with the same team. You are reading this for your own decisions, not to relay it to the team — so judge these two choices the way you would judge a plan you have to defend. One produces a decision; the other produces a vibe.

  • Pilot A: 'Everyone, try AI on whatever you want this month. We'll see how it goes.'
  • Pilot B: 'For 30 days, everyone uses AI for the weekly status draft only. Success = manager edits drop and nobody reports a data-boundary near-miss. We review every Friday.'
  • Pilot A generates anecdotes and a few enthusiastic users. Pilot B generates a yes/no answer about one workflow you can then expand or kill.

Quick check · <30 sec

Why is Pilot B more useful to a manager than Pilot A?

  • A. It is more ambitious
  • B. It is scoped to one workflow with a defined success signal and a review cadence
  • C. It uses more AI
  • D. It involves the whole team
Show answer
A pilot is an experiment. Pilot B has one variable, a success measure, and a cadence — so it produces a decision. Pilot A has none, so it produces stories.

Part 2

Picking pilot workflows that surface signal

A good pilot workflow is narrow, frequent, low-risk, and instrumented. The point is not to find where AI is most exciting — it is to find where you can clearly see whether it helped. Pick workflows that produce either a strong adoption signal or a clear failure signal fast.

  • Frequent: it happens enough times in 30 days to show a trend, not a coincidence.
  • Reviewable: you can tell good output from bad without specialist judgment.
  • Bounded: a Module 3 'safe' or low 'caution' workflow, never 'prohibited'.
  • Instrumented: you decided before the pilot what signal you'll watch.
  • Avoid: high-stakes, rare, or sensitive workflows as a first pilot — the signal is too slow and the downside too large.

Quick check · <30 sec

Which is the better first pilot workflow: (A) drafting the weekly internal status update, or (B) drafting customer-facing contract language?

  • A. A
  • B. B
  • C. Either
  • D. Neither
Show answer
A is frequent, reviewable, and low-risk — fast clear signal. B is high-stakes and would sit in Module 3's caution/prohibited zone; a poor first pilot.

Part 3

The one-page team working agreement

A working agreement is a single page the team can actually remember. It answers three questions: what we standardize, what we prohibit, and what we review. It is not a policy (that is Module 8) — it is the team's local norm, consistent with company policy but specific to this team's work.

  • Standardize: the shared prompt patterns, the persona cards, and the one pilot workflow everyone runs the same way.
  • Prohibit: the team's specific 'never' list, inherited from Module 3 and the personas — concrete, not 'be careful'.
  • Review: what gets a human check before it leaves the team, and the weekly cadence for looking at the pilot.
  • One page or it will not be used. If it needs a second page, the team is trying to boil the ocean.

Quick check · <30 sec

What is the difference between a team working agreement and a company AI policy?

Show answer
The working agreement is a one-page local norm specific to this team's work and consistent with company policy. The company policy (Module 8) is the org-wide governance document. The agreement is downstream of, and narrower than, the policy.

Part 4

Coaching at different fluency levels

Your team is not at one level. A useful model has four: new (hasn't really used it), curious (experimenting, no method), regular (uses the prompt pattern, sometimes over-trusts), and advanced (builds artifacts and briefs, needs guardrails not encouragement). Coaching that ignores the level wastes everyone's time.

  • New: pair them on one safe workflow; remove the blank-page fear before teaching nuance.
  • Curious: give them the Module 2 pattern and a persona card; convert energy into method.
  • Regular: coach verification and the data boundary; this is where automation bias appears.
  • Advanced: review their briefs adversarially; their risk is over-permissioning, not under-use.

Quick check · <30 sec

A team member produces polished AI output fast but rarely checks it. Which level, and what do you coach?

  • A. New — coach basic prompting
  • B. Regular — coach verification and the data boundary
  • C. Advanced — coach artifact building
  • D. Curious — coach enthusiasm
Show answer
Fast polished output plus weak checking is the classic 'regular' automation-bias pattern. The coaching need is verification discipline, not more prompting technique.

Part 5

Measuring adoption — not usage

Usage is how often the tool is opened. Adoption is whether the work got better. They are not the same, and tracking only usage is how programs declare victory while nothing improved. Measure the outcome of the pilot workflow, plus a safety counter-metric so speed never hides a boundary problem.

  • Outcome signal: e.g., manager edits on the status draft trending down, cycle time down, rework down.
  • Safety counter-metric: data-boundary near-misses reported — this should stay at or near zero, and reporting one is a good sign, not a bad one.
  • Qualitative signal: in the Friday review, can people point to a specific task that got better?
  • If usage is up but the outcome signal is flat, the pilot is not working — say so and change it.

Quick check · <30 sec

Usage is up 300% but manager edits on the piloted draft are unchanged. What does this mean?

  • A. Strong adoption
  • B. The pilot is not improving the work; investigate or change it
  • C. Success — keep going
  • D. Nothing; usage is what matters
Show answer
High usage with a flat outcome signal means activity without improvement. Adoption is measured by the outcome, not the activity.

Part 6

The 30-day plan and the retro

A 30-day pilot has a shape: week 1 set up and baseline, weeks 2–3 run and coach, week 4 retro and decide. The retro answers one question: expand, adjust, or stop. A pilot that cannot be stopped was never an experiment.

  • Week 1: pick the workflow, write the agreement, capture a baseline of the outcome signal.
  • Weeks 2–3: run it, coach by level, log near-misses, hold the Friday review.
  • Week 4: retro — what the signal says, what the team says, expand/adjust/stop.
  • Decide in writing. 'It felt fine' is not a decision.
Activity · ~12 minadoption plan

You are planning a 30-day AI pilot for your team. You will lay out the four weeks with a workflow, an agreement, signals, and a decision point.

  • Team size: Use your real team size, no names
  • Constraint: Exactly one pilot workflow; it must be Module 3 'safe' or low 'caution'

Your task

Lay out the 30-day plan: (Week 1) the one pilot workflow, the one-page agreement's standardize/prohibit/review lines, and the baseline value of your outcome signal. (Weeks 2–3) the coaching focus per fluency level and the weekly review cadence. (Week 4) the outcome signal target, the safety counter-metric, and the expand/adjust/stop criteria. Name one way this pilot could produce a false positive.

Show a hint
If you cannot name the baseline number in Week 1, you will not be able to claim improvement in Week 4. Pick a signal you can actually measure cheaply.
Compare with a strong answer
Week 1: pilot = weekly status draft for all 6 team members; standardize = shared persona + Module 2 pattern; prohibit = no headcount/vendor pricing in any outside tool; review = Friday 15 min. Baseline: manager rewrites ~40% of each draft. Weeks 2–3: coach 'regular' members on verifying figures; log near-misses. Week 4: target manager rewrites under ~15%, near-misses at zero (reported ones count as healthy); expand if both met, adjust if signal flat, stop if any boundary breach. False positive: edits drop because drafts got shorter, not better — also track completeness.

Why this matters: Writing the decision criteria before the pilot starts is what prevents a manager from rationalizing whatever happened as success.

Quick check · <30 sec

What single question must the week-4 retro answer?

  • A. Did people enjoy it?
  • B. Expand, adjust, or stop?
  • C. How much was it used?
  • D. Who used it most?
Show answer
A pilot is an experiment with a decision at the end: expand, adjust, or stop — decided against the pre-set signal, in writing.

End-of-module quick check

Five short retrieval questions. Answer from memory first, then reveal each explanation.

  1. 1. The main reason to scope a pilot to one workflow is to…

    • A. Limit cost
    • B. Get a clear, fast signal you can act on
    • C. Reduce training time
    • D. Keep it simple for legal
    Show answer
    One variable means the pilot can actually tell you whether that workflow improved — it produces a decision instead of anecdotes.

  2. 2. Adoption is best measured by…

    • A. How often the tool is opened
    • B. Whether the piloted work measurably got better
    • C. Number of licenses
    • D. Enthusiasm in the retro
    Show answer
    Usage is activity. Adoption is the outcome signal on the piloted workflow, paired with a safety counter-metric.

  3. 3. A 'regular'-fluency team member most needs coaching on…

    • A. Basic prompting
    • B. Verification and the data boundary
    • C. Artifact building
    • D. Enthusiasm
    Show answer
    Regular users produce fast polished output and tend toward automation bias; verification discipline is the gap.

  4. 4. True or false: a pilot with no defined stop condition is still a valid experiment.

    • A. True
    • B. False
    Show answer
    False. If it cannot fail and be stopped, it is not an experiment — it will always be rationalized as a success.

  5. 5. Name the three lines of a one-page team working agreement.

    Show answer
    What we standardize, what we prohibit, and what we review (with a weekly cadence).

Further reading

Worked examples by role

Operations manager

Pilot: weekly status draft

Workflow: every team member drafts their weekly status with the shared persona + Module 2 pattern. Standardize: the prompt and persona. Prohibit: headcount and vendor pricing in any outside tool. Review: Friday 15 min. Outcome signal: manager rewrite share, baseline ~40%, target <15%. Counter-metric: boundary near-misses at zero. Decision: expand/adjust/stop in week 4, in writing.

Sales coordinator

Pilot: follow-up email first drafts (sales team)

Workflow: first-draft follow-ups from sanitized account context, owner always sends. Standardize: the draft prompt and the fact-check step. Prohibit: competitor/pricing claims, customer identifiers in unapproved tools. Review: weekly pipeline sync. Outcome signal: factual corrections per ten drafts, target under one. Counter-metric: zero customer-data near-misses. Decision documented at day 30.

Training manager

Pilot: comprehension-question drafting (L&D team)

Workflow: first-draft comprehension questions from approved objectives. Standardize: the bias + accessibility review step. Prohibit: publishing without that review. Review: weekly content standup. Outcome signal: reworked items per set, target under one. Counter-metric: bias/accessibility flags caught before publish, not after. Decision: expand to rubric drafting only if signal holds.

Before / after

Pilot definition: open-ended vs. instrumented

Before: 'Let's all try AI this month and regroup.' No workflow, no baseline, no success measure, no stop condition.

After: 'One workflow (weekly status draft), baseline manager-rewrite share captured, target under 15%, zero boundary near-misses, Friday reviews, expand/adjust/stop decided in writing at day 30.'

What changed: The instrumented version can be falsified — it can fail and tell you so. The open-ended one can only ever feel like a success.

Completion artifact

Submit a one-page team working agreement (standardize / prohibit / review) plus a 30-day pilot plan with one workflow, a baseline, an outcome signal with a target, a safety counter-metric, and an expand/adjust/stop decision rule.

ExercisePilot-ready artifact

Produce your team's one-page working agreement and a 30-day pilot plan. Keep it to one page of substance. No employee names; describe the team by size and function.

Participant template

  • Team (size + function, no names):
  • Pilot workflow (one, Module 3 safe/low-caution):
  • Agreement — Standardize:
  • Agreement — Prohibit (concrete categories):
  • Agreement — Review (what + weekly cadence):
  • Outcome signal + week-1 baseline:
  • Safety counter-metric:
  • Coaching focus by fluency level:
  • Week-4 decision rule (expand/adjust/stop):
  • One way this pilot could give a false positive:

Example submission

Team: 6-person shared-services ops team. Pilot: weekly status draft. Standardize: shared persona + Module 2 pattern. Prohibit: headcount, vendor pricing in any outside tool. Review: Friday 15-min, manager checks figures trace to notes. Outcome signal: manager rewrite share, baseline ~40%, target <15%. Counter-metric: boundary near-misses (target 0; reported ones healthy). Coaching: pair the one 'new' member, push 'regular' members on verification. Decision: expand if signal + counter-metric met, adjust if flat, stop on any breach. False positive: shorter drafts cut edits without improving quality — also track completeness.

Role-flavored variants

Same exercise, framed for different roles. Use the one closest to your work.

Front office coordinator

You lead a front-desk team. Pick a documentation or scheduling workflow. The prohibit line must address personal/contact data; the decision rule must include a quality counter-metric so 'faster' never hides a worse handoff.

See a sample submission
Front-desk lead plan (excerpt): Pilot: end-of-day callback list drafted from sanitized voicemails. Standardize: the draft prompt + morning lead review. Prohibit: contact details in any outside tool. Outcome signal: lead reorders per list, baseline ~5, target under 2. Counter-metric: zero mis-prioritized clinical items. Decision: expand only if both hold for the full month.

Software developer

You lead an engineering team. Pick a low-risk delivery workflow (test scaffolding or PR summaries). The prohibit line must forbid pasting proprietary code/secrets into unapproved tiers; the signal must be reviewable in CI or review.

See a sample submission
Eng lead plan (excerpt): Pilot: AI-suggested unit tests as PR comments. Standardize: approved IDE tier only, tests authored by the dev. Prohibit: proprietary code/secrets in consumer tools, AI approving PRs. Outcome signal: suggested-test acceptance rate, target >50%. Counter-metric: zero secret-exposure incidents. Decision documented at day 30.

Learner checklist

Use this as a final check before submitting. Program leads use a separate review guide when they approve or coach submissions.

  • • One bounded pilot workflow named
  • • Week-1 baseline captured
  • • Outcome signal with a target (not a usage count)
  • • Safety counter-metric present
  • • Standardize/prohibit/review lines are concrete
  • • Expand/adjust/stop decision rule stated
Loading your module status…

Previous module

← From role/persona to agent workflow brief

Review the prior step in the path.

Next module

Business AI policy path →

Keep momentum with the next completion check.