Resources

Prompts you can copy and use today

Every prompt follows FRAME. Bracketed items are yours to edit: swap in your audience, your word count, your document. Nothing here needs a login.

Prompt library

Find a prompt

Search by task or filter by category.

7 of 7 prompts shown.

Building your own tool

Turn an idea into a build brief (one AI prompts another)

When to use: Before you build a tool in Lovable or a similar builder.

I want to build [describe the tool in one or two sentences]. Interview me one question at a time to pull out the inputs, the outputs, the edge cases, and the data it needs and where that data lives. When you have enough, write a clear, structured build brief I can paste into an app builder. Do not start building; just write the brief.

Communications

Manager talking points for a change

When to use: Rolling out a policy, comp, or benefits change and managers need to speak to it.

You are an HR communications partner. From the update below, write manager talking points for a [15-minute] team conversation: a two-sentence summary, the three things employees will care about most, three likely questions with plain answers, and one line on where to send people for more help. Keep it under [250] words.

[paste the update]

Benefits

Turn a policy into a plain-language FAQ

When to use: Any dense policy or plan document employees actually have to read.

You are an HR benefits writer. Turn the policy below into a plain-language FAQ for [hourly employees] at a [sixth]-grade reading level. Use a question-and-answer format, [eight] questions maximum, keep every rule intact, and flag anywhere the language is ambiguous instead of guessing.

[paste the policy]

Benefits

Interrogate a document before employees do

When to use: Pressure-test a plan document or handbook for gaps.

You are a skeptical employee reading the document below for the first time. List the ten questions you would ask that the document does not clearly answer. For each one, quote the closest relevant line, or say "not addressed."

[paste the document]

Compensation

Clean up and sanity-check a job description

When to use: A messy JD, or checking a role against your leveling.

You are a compensation analyst. Rewrite the job description below into a clean, consistent format with sections for summary, key responsibilities, and required qualifications. Remove duplication and vague filler without inventing scope, authority, credentials, or direct reports. Then compare each responsibility against the leveling descriptions I paste, quote the leveling text you relied on, and flag anything that reads above or below the stated level. If I have not pasted leveling descriptions, ask for them instead of guessing.

[paste the JD]

[paste your leveling or job architecture descriptions]

Talent

Draft consistent, role-specific interview questions

When to use: Bringing consistency and fairness to interviews.

You are a talent partner. From the role summary below, generate [eight] structured interview questions that assess the core competencies for this role. For each, add what a strong answer includes. Keep every question job-related and avoid anything that could touch protected characteristics. These support the interviewer's judgment; they do not replace it.

[paste the role summary]

Editing

Score and improve a draft against a rubric

When to use: Offer letters, comms, anything with clear quality criteria.

Score the draft below against these five criteria [list them], each out of five with a one-line reason. Then rewrite the draft to raise the two lowest scores, and show what you changed.

[paste the draft]

Booklet

AI terminology for HR

Booklet: pp. 6 to 7. You do not need to be technical. You need enough vocabulary to make good decisions about when to trust a tool, when to check it, and when to keep a human in the chair.

Open the terminology list

Large language model (LLM): The engine behind ChatGPT, Claude, Gemini, and Copilot. It predicts the next most likely words based on everything it has read. It is not a database and it is not looking things up unless you give it something to read.
Prompt: What you type in. The single biggest lever you have over the quality of the output. Most bad results are bad prompts, not bad models.
Context window: How much the model can 'hold in its head' at once: your instructions, the document you pasted, and the back-and-forth so far. Big jobs can overflow it, which is when a model starts forgetting what you told it earlier. Do not push a model to the limits of its context window. The model will produce the best results before hitting 50% of its capacity. After that point, the model may stop following instructions or lose context.
Token: The unit models read and write in, roughly three-quarters of a word. You will see it in pricing and limits. You rarely need to think about it beyond 'very long inputs cost more and can crowd the context window.'
Hallucination: When the model states something false with total confidence: a made-up citation, a wrong number, or a policy that does not exist. This is the single most important word in this booklet. Assume it can happen every time.
Grounding: Giving the model the real source material to work from, such as your actual policy or your actual data, instead of letting it guess from memory. Grounding is your best defense against hallucination.
RAG (retrieval-augmented generation): A fancy term for grounding done automatically: the tool retrieves the right document first, then answers from it. This is how a policy chatbot answers from your handbook and not the internet.
Agent: An AI that can take steps, not just answer: fill a form, send a draft, or move data between tools. More power, more need for oversight.
Guardrails: The limits you put around a tool so it behaves: what it is allowed to do, what it must refuse, and what it must never touch, such as PII.
PII / confidential data: Personally identifiable information, including names, salaries, SSNs, and health details, and anything your company treats as confidential. The thing you have to think hardest about before you paste.
Generative AI: AI that creates new content, including text, images, audio, and code, rather than just sorting or scoring what already exists. Everything you build today sits on top of it.
Bias: Systematic skew in an AI's output that reflects patterns in its training data. It matters most the moment a tool touches a decision about a person, which is why it gets its own line in the checklists later.
Model vs. app: ChatGPT is a model you talk to. Lovable is a builder you use to create your own app that other people use. Today you will use both and you will see how there is a difference between using AI and building with it.
Dependency: Anything you wait on to get your work done. Naming your dependencies out loud is step one of removing them.
The execution gap: The distance between what your team knows needs doing and its actual capacity to do it. AI does not close that gap by making you smarter. It closes it by removing dependencies so knowing turns into doing.
API: A defined way for software systems to communicate with one another and request data or actions.
MCP (Model Context Protocol): A standard way for an AI model to connect with external tools and sources so it can use context and perform actions through them.

FRAME

Five moves for a prompt that works

Booklet pp. 8 to 9.

F
Frame the job.
Give the AI a role and a clear definition of done. Not "help with this policy," but "You are an HR communications specialist. Rewrite this policy for frontline employees at a ninth-grade reading level."
R
Real context.
Add the constraints the model cannot know: your audience, your plan year, your comp philosophy, what is off-limits. This is the step people skip, and it is the one that matters most.
A
Add your source.
Paste the actual messy input: the real job description, the real policy, the real de-identified data. Grounding beats guessing every time.
M
Model the output.
Say exactly what you want back: the format, the length, the tone, and one example of "good." Showing one example beats describing ten rules.
E
Evaluate and iterate.
Treat the first answer as a draft, never a deliverable. Refine it in plain language, then audit it against the Human Stewardship Model before it goes anywhere near an employee.

Build path

From idea to shipped, in seven moves

Booklet p. 10. The same path the guided build and your own build follow.

1
Name the dependency.
What, or who, are you waiting on? Write it in one sentence. If you cannot name it, you are not ready to build.
2
Front-load the reasoning.
In a regular AI chat, think the whole thing through before you touch a builder: inputs, output, edge cases, and where the data lives.
3
Let one AI write the build brief.
Have the chat model turn your reasoning into a clear brief you can hand to the builder.
4
Build in small moves.
Paste the brief into the builder, then work in small, testable steps. Tip: If you build with an MCP connection, you can have your LLM direct the build in small steps. This may surface items for you to address while building, helping you build something more aligned to your overall vision.
5
Test with real, messy data.
Not the clean example. The ugly real thing (de-identified). That is where it breaks.
6
Audit with HUMAN.
Run the output through the Human Stewardship Model. Facts, math, tone, bias, policy fit.
7
Ship to one real person.
Hand it to a single colleague who has the actual problem. Watch them use it. Then widen.

Official links and tools

Where to go next

Official Lovable documentation and the tools mentioned in the room. All open in a new tab.

Lovable
The builder used in the workshop. Start here for your own tools.
Open
Lovable documentation
Official docs: getting started, features, and how the builder works.
Open
Credits and usage
How credits work and what counts toward them.
Open
Data opt-out guide
Turn off "Use my Lovable content for model training" in Account settings.
Open
Lovable security
Security and trust information from Lovable.
Open
Lovable Trust Center
Compliance and security documentation.
Open
Lovable Academy
Official courses and lessons for learning the builder.
Open
Gemini Notebook
Turn one trusted source into many formats: summary, FAQ, audio overview.
Open

Workshop files (ZIP)

Job Family Descriptions.csv (50 rows), Job Leveling Guide.csv (15 rows), Example Job Description.pdf, and README. Workshop example materials; no employee data.

Download

Live examples

Finished tools from Josh's practice

Josh Lemon AI Playground
Working examples from Josh's practice, including a couple of additional playground items.
Open example

Questions: josh@joshualemon.com. Back to Start Here.