LLaManchaAI Enablement
Checking access…
← All modules
Module 5Audience: all

Role/persona/onboarding context

Turns real roles and workflows into reusable AI context for training, onboarding, prompts, and future assistants.

Outcomes

What you will be able to do

  • Explain why a role persona is operational memory, not a job description.
  • Interview yourself against the twelve canonical persona fields.
  • Reuse a persona card to seed prompts without re-typing your context.
  • Keep a persona current as the role changes.
Completion check

How this module is approved

Build a sanitized role persona card for your role using all twelve canonical fields.

Pass criteria

  • • All twelve fields present
  • • Recurring workflow is concrete and sanitized
  • • Sensitive-data categories and prohibited AI uses are specific
  • • Human review rule and escalation path named
Loading personalization context…
Lesson30–40 minutes self-guided, or 45–60 minutes with a guided self-interview

What you should take away

By the end of this module, you will have a sanitized role persona card you can paste at the top of future prompts so AI tools understand your work without you re-explaining it every time.

Prerequisites: Complete Module 1 · Complete Module 2

Part 1

Start here: two descriptions of the same job

Read these two descriptions of one role. The first is a job description. The second is a role persona. The second is the one an AI tool can actually use, because it describes the work, not the title.

  • Job description: 'Operations coordinator. Supports cross-functional teams, maintains reporting, and drives process improvements.'
  • Role persona: 'Every Monday I turn five teams' sanitized status notes into one leadership update. I work in a spreadsheet and a wiki. I never put headcount or vendor pricing in an outside tool. My manager reviews the update before it goes to the leadership channel.'
  • The job description could describe a thousand people. The persona could only describe this person's actual Monday.

Quick check · <30 sec

Which description gives an AI tool enough to draft a useful first version of the Monday update?

  • A. The job description
  • B. The role persona
  • C. Both equally
  • D. Neither
Show answer
The persona names the recurring workflow, the inputs, the tools, the data boundary, and the review gate. The job description names none of those, so the model would have to guess all of them.

Part 2

Why operational context beats the job description

A persona is operational memory: the specific, recurring, sanitized facts about how you actually work. Job descriptions are written for hiring and HR. Personas are written for getting useful, safe AI help. The more concrete and current the persona, the less the model guesses and the easier the output is to review.

  • Concrete beats abstract: 'weekly leadership update from five sanitized status notes' beats 'maintains reporting.'
  • Recurring beats one-off: a persona is most valuable for work you do many times.
  • Boundaries are part of the context: what must never go into a tool is as important as what the work is.
  • A persona is sanitized by construction: it describes categories ('vendor pricing'), never the actual sensitive values.

Quick check · <30 sec

In one phrase, rewrite 'drives process improvements' as something a persona would actually say.

Show answer
A persona-grade answer names the specific recurring action and its boundary, e.g. 'monthly, I draft a process-change proposal from sanitized retro notes; my lead approves before it changes the runbook.' Abstract verbs like 'drives' are the thing a persona is supposed to replace.

Part 3

Interview yourself: the twelve canonical fields

A role persona has twelve fields. You build it by interviewing yourself, not by writing prose. Answer each field with a concrete, sanitized sentence. If a field is hard to answer, that is usually the field that matters most.

  • Role name · Department or team · Core responsibilities · Recurring workflows
  • Tools and systems · Inputs · Outputs · Sensitive data categories
  • Good AI use cases · Prohibited AI uses · Human review rules · Escalation path
  • The four hardest fields for most people are Sensitive data, Prohibited AI uses, Human review rules, and Escalation path — they are also the four that keep a persona safe.
Activity · ~10 minpersona interview

You are going to interview yourself to populate a role persona card. Ten questions cover all twelve fields — a couple of answers fill two fields at once (role + team, good + prohibited uses), which is expected, not a gap. Answer out loud or in a scratch doc. Keep every answer sanitized — describe categories, never real names, numbers, or records.

  • Time budget: 10 minutes, ~45 seconds per question
  • Rule: Concrete and recurring beats broad and aspirational

Your task

Answer these ten questions in one sanitized sentence each: (1) What is your role and team? (2) What are your two or three core responsibilities? (3) What is one workflow you repeat weekly or monthly? (4) What tools/systems do you use for it? (5) What inputs start that workflow? (6) What output ends it? (7) What categories of data must never leave an approved tool? (8) Where could AI safely assist this workflow today? (9) What must AI never decide here? (10) Who reviews or approves the output, and when do you escalate?

Show a hint
If an answer could describe a coworker in a different role, it is too vague. Add the detail that makes it yours.
Compare with a strong answer
Role: front office coordinator on the patient-intake desk. Core: scheduling, intake documentation, handoffs. Weekly workflow: turn the day's voicemails into a callback list. Tools: scheduling system, shared inbox. Inputs: voicemails, walk-in notes. Output: a prioritized callback list for the team. Never leaves an approved tool: any personal or contact detail. Safe AI assist: drafting neutral confirmation wording from sanitized notes. AI must never: decide clinical urgency or who gets seen first. Review/escalate: the front-desk lead reviews the list each morning; anything that sounds clinical goes straight to a nurse, not AI.

Why this matters: The interview format forces specificity. Prose lets you hide behind abstractions; ten direct questions do not.

Quick check · <30 sec

Which two fields most directly keep a persona safe rather than just useful?

  • A. Role name and Department
  • B. Core responsibilities and Outputs
  • C. Sensitive data categories and Prohibited AI uses
  • D. Tools and Inputs
Show answer
Useful fields tell the model what to do. Sensitive-data categories and prohibited AI uses tell it (and you) what not to do — that is what prevents the common data-boundary mistake from Module 3.

Part 4

How a persona seeds every future prompt

The payoff is that you stop re-typing your context. Once the persona exists, you paste a short, sanitized version of it above any prompt, or save it as a custom instruction / project setting in an approved tool. The model now starts every task already knowing your role, boundaries, and review gate.

  • Manual: paste a 4–5 line persona summary above the Goal/Context/Constraints/Output/Validation prompt from Module 2.
  • Built-in: store the same summary as a custom instruction, project, or space in an approved tool so it applies automatically.
  • Safety carries over: because the persona names prohibited uses and the review gate, every seeded prompt inherits them.

Quick check · <30 sec

Why does seeding a prompt with a persona make the output easier to review, not just faster to produce?

Show answer
The persona states the boundaries and the review gate up front, so the output arrives already shaped to your constraints and you are checking against a known standard instead of re-deriving it each time.

Part 5

Keeping the persona current

A persona is operational memory, and memory goes stale. Roles drift: new tools, new boundaries, a workflow that moved teams. A stale persona is worse than none, because it confidently seeds prompts with wrong context. Treat it like a runbook: short, owned, and reviewed on a cadence.

  • Re-read the persona when a tool, an approval rule, or a recurring workflow changes.
  • Review it on a fixed cadence — a quarterly five-minute pass is enough for most roles.
  • Version it lightly: note what changed and when, so a reviewer can trust it.
  • If the Sensitive-data or Prohibited-uses fields change, update before you reuse the persona, not after.

Quick check · <30 sec

True or false: a slightly outdated persona is harmless because the model will fill in the gaps.

  • A. True
  • B. False
Show answer
False. A stale persona seeds every downstream prompt with confident-but-wrong context — especially dangerous if the outdated field is a data boundary or a review rule.

End-of-module quick check

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

  1. 1. A role persona differs from a job description mainly because it…

    • A. Is longer and more formal
    • B. Describes recurring operational work, boundaries, and review gates
    • C. Is written by HR
    • D. Lists salary band and reporting line
    Show answer
    A persona is operational memory: concrete recurring work plus the boundaries and review gate that make AI help safe.

  2. 2. How many canonical fields does a role persona card have?

    • A. Six
    • B. Nine
    • C. Twelve
    • D. As many as you want
    Show answer
    Twelve: role name, department, core responsibilities, recurring workflows, tools, inputs, outputs, sensitive data, good AI uses, prohibited AI uses, human review rules, escalation path.

  3. 3. Which field most directly prevents the Module 3 data-boundary mistake?

    • A. Core responsibilities
    • B. Sensitive data categories
    • C. Tools/systems
    • D. Role name
    Show answer
    Naming the sensitive-data categories up front means every prompt seeded with the persona inherits that boundary.

  4. 4. True or false: once written, a persona does not need maintenance.

    • A. True
    • B. False
    Show answer
    False. Roles drift; a stale persona confidently seeds wrong context. Review on a cadence and before reuse if a boundary changed.

  5. 5. What is the main day-to-day payoff of having a persona card?

    Show answer
    You stop re-typing your role and boundaries into every prompt — you paste or store the persona once and every task inherits your context and review gate.

Further reading

Worked examples by role

Front office coordinator

Persona card: front office coordinator

Recurring workflow: turn the day's sanitized voicemails into a prioritized callback list. Tools: scheduling system, shared inbox. Sensitive data: any personal or contact detail — never leaves the approved system. Good AI use: drafting neutral confirmation wording. Prohibited: AI decides clinical urgency or priority. Review: front-desk lead each morning; anything clinical escalates to a nurse, not AI. Fork-ready prompt seeded by this persona: "Draft a neutral confirmation message for a rescheduled internal meeting using only the sanitized details below."

Operations manager

Persona card: operations manager

Recurring workflow: weekly leadership update from five teams' sanitized status notes. Tools: spreadsheet, wiki. Sensitive data: headcount, vendor pricing — categories only, never values, in any outside tool. Good AI use: first-draft narrative and risk list. Prohibited: AI invents metrics or owners. Review: manager approves before the leadership channel; budget-impacting items escalate to finance. Fork-ready prompt seeded by this persona: "From these sanitized status bullets, draft a 150-word weekly update with progress, risks, and next steps. Do not invent dates or owners."

Software developer

Persona card: software developer

Recurring workflow: draft first-pass unit tests from a described function contract. Tools: IDE assistant (approved tier), repo, CI. Sensitive data: proprietary source, secrets, internal URLs — never pasted into a consumer tool. Good AI use: edge-case test generation, sanitized log triage. Prohibited: AI approves PRs or alters access controls. Review: peer review + CI before merge; security-relevant changes escalate to the security reviewer. Fork-ready prompt seeded by this persona: "Given this sanitized behavior description, propose unit tests and edge cases. Do not invent external APIs. List assumptions."

Before / after

Same task, no persona vs. persona-seeded

Before: Prompt: 'Write my weekly update.' The model invents a structure, a tone, and placeholder metrics, and has no idea what must stay out of the tool.

After: Prompt: persona summary pasted on top ('operations manager; weekly leadership update from five sanitized status notes; never include headcount or vendor pricing; manager approves before posting') followed by the Module 2 pattern. The model returns a correctly scoped draft with a validation note.

What changed: The persona supplied audience, recurring structure, the data boundary, and the review gate — four things the bare prompt forced the model to guess.

Completion artifact

Submit a sanitized role persona card with all twelve fields. Each field should be one concrete sentence. No real names, numbers, records, or credentials.

ExercisePilot-ready artifact

Build your sanitized role persona card with all twelve fields. Answer the self-interview, then write one concrete sentence per field. Use categories, never real values.

Participant template

  • Role name:
  • Department/team:
  • Core responsibilities:
  • Recurring workflows:
  • Tools/systems:
  • Inputs:
  • Outputs:
  • Sensitive data categories:
  • Good AI use cases:
  • Prohibited AI uses:
  • Human review rules:
  • Escalation path:

Example submission

Role: operations manager. Team: shared services. Core: weekly reporting, process upkeep, cross-team coordination. Recurring workflow: Monday leadership update from five sanitized status notes. Tools: spreadsheet, wiki. Inputs: team status notes. Outputs: one leadership update. Sensitive data: headcount, vendor pricing (categories only, approved tools only). Good AI use: first-draft narrative and risk list. Prohibited: AI invents metrics, owners, or dates. Review: manager approves before posting. Escalation: budget-impacting items go to finance before the update ships.

Role-flavored variants

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

Sales coordinator

Frame the persona around an outreach or follow-up workflow. The Sensitive-data field must name customer identifiers and deal terms; the Prohibited field must forbid AI inventing competitor or product claims.

See a sample submission
Sales coordinator persona (excerpt): Recurring workflow: first-pass follow-up email from sanitized account context. Sensitive data: customer names, deal value, contract terms — categories only, approved CRM only. Good AI use: tone and structure of a follow-up I will fact-check. Prohibited: AI states competitor claims or commitments. Review: account owner approves before send; pricing questions escalate to the deal desk.

Training manager

Frame the persona around a curriculum or evaluation workflow. The Prohibited field must address assessment bias; Human review rules must include an accessibility pass.

See a sample submission
Training manager persona (excerpt): Recurring workflow: draft comprehension questions from learning objectives. Tools: authoring tool, LMS. Sensitive data: learner records — never in an outside tool. Good AI use: first-draft questions and rubric calibration. Prohibited: AI ships assessment items without a bias and accessibility review. Review: I check reading level and bias before publish; flagged items escalate to instructional design.

Learner checklist

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

  • • All twelve fields present
  • • Recurring workflow is concrete and sanitized
  • • Sensitive-data categories and prohibited AI uses are specific
  • • Human review rule and escalation path named
Loading your module status…

Previous module

← Build something useful with AI

Review the prior step in the path.

Next module

From role/persona to agent workflow brief →

Keep momentum with the next completion check.