Skip to content

Stage 04: Share & Next Steps, 30 minutes

Show the work. Then choose one person and one move.

You have a working prototype. Show what it can do, what you tested, and what needs review before real use.

Two-minute demo

Problem, capability, next step

Do not narrate the build. Show one input going in and one output coming out, then say what you would do next.

  1. 20 seconds

    The problem

    The dependency, in one sentence. Who waits, on what.

  2. 60 seconds

    The capability

    Paste the fictional input. Show the output. Point at one thing you tested and one thing the tool refused to invent.

  3. 40 seconds

    The surprise and the next step

    One thing that surprised you while building. Then the colleague who will test it and the first improvement you would make.

Observers: how to contribute in the share-out
  • Ask the builder to show where one claim in the output came from in the source.
  • Say what you tried as a tester and what happened.
  • Name the HUMAN check you think is most at risk for that tool, and why.
  • Offer to be the first colleague tester when it goes back to the office.

Before anything goes anywhere

HUMAN review

Run every output through the Human Stewardship Model. Facts, math, tone, bias, policy fit. The AI drafts; the human decides.

The Human Stewardship ModelAI can draft, calculate, and build. It cannot be accountable. That part stays with you.HHumanAccountabilityYou own the outcomes.AI assists, but you areresponsible.UUnderstandthe SourceKnow where data comesfrom. Fact, orhallucination?MMaintainContextAlign with the policies,plans, and audiencethat AI lacks.AAuditthe OutputRigorously check facts,math, and tone,every time.NNever DelegateJudgmentDecisions about peoplestay humandecisions.JOSH LEMON AI
H Human Accountability
You own the outcomes. AI assists, but you are responsible.
U Understand the Source
Know where the data came from and where it goes.
M Maintain Context
Give the model your rules, plan, and philosophy; it does not know them.
A Audit the Output
Facts, math, dates, citations, tone. Assume one thing is wrong until checked.
N Never Delegate Judgment
Decisions about people are made by people.
Before you paste anything in
  • Is this PII or confidential? Names, pay, health, SSNs, anything your company protects.

    Understand the Source, Maintain Context

  • Do I have the right to put it here? Just because you can paste it does not mean you are allowed to.

    Human Accountability

  • Is this tool on my company's approved list? If you do not know, that is your answer for today: ask first.

    Human Accountability

  • Where does this data go, and is the tool training on my input? Check the setting. Business and enterprise tiers usually let you turn training off.

    Understand the Source

  • Can I use a de-identified sample instead of the real thing? Almost always, yes. Strip the identifiers, keep the structure.

    Maintain Context

Before you ship anything out
  • Did I audit the output? Facts, math, dates, citations, and tone. Assume one thing is wrong until you have checked.

    Audit the Output

  • Does it fit our actual policy, plan, and philosophy? The model does not know your rules unless you gave them to it.

    Maintain Context

  • If this touches a decision about a person, did a person make the call? Hiring, pay, performance, exits: the AI drafts, the human decides.

    Never Delegate Judgment

  • Did I check for bias or unequal impact? Especially anything that sorts, scores, or screens people.

    Never Delegate Judgment, Audit the Output

  • Does this need real sign-off before it goes live? Employee-facing pay, benefits, and policy content usually does. Get it.

    Human Accountability

After the workshop

From prototype to something people rely on

Today's builds run on fictional data. Before one is used for real work it needs an owner, approved sources, repeatable tests, access controls, sign-off, and monitoring. Each step below is a concrete task you can start this month.

  1. 01

    Owner

    Name one person accountable for the tool, its outputs, and turning it off.

  2. 02

    Authorized sources

    Replace fictional samples with approved, current documents, and record who approved them.

  3. 03

    Tests

    Turn today's test checklist into a repeatable set, including edge cases and adversarial pastes.

  4. 04

    Access controls and server secrets

    Login for the right people only; AI credentials stay server-side, never in the browser.

  5. 05

    Approvals

    Security, privacy, legal, and HR leadership sign-off before any employee-facing use.

  6. 06

    Monitoring

    Review outputs on a schedule, log failures, and decide in advance what triggers a pause.

Your next action

Name one colleague who will test what you built, and one first step you will take this week.

A tester, not an audience. Someone who has the actual problem and will tell you where it breaks. Write both into the plan below before you leave the room.

Fill in my action plan

Action plan

Keep building

Saved in this browser only. Print it, or save it as a PDF, and take it back to the office. Booklet: p. 15.

Questions after today: josh@joshualemon.com. Prompts, FRAME, and links live under Resources.