Workflow 1: meeting notes to action plan
Use this workflow after calls, sales conversations, and planning sessions. The goal is a decision-ready action plan with owners, dates, open questions, and follow-ups, not another summary.
Inputs -> prompt -> output -> QA. Every workflow in this guide uses the same four stages. If the prompt stage itself is where you are unsure, the prompt engineering for beginners framework covers how to write the instruction, context, constraints, and format that each template below assumes.
| Stage | What you supply | Why it matters |
|---|---|---|
| Inputs | The real material: a transcript, an inbox, source pages, a diff, a task list, or a draft | Vague requests let the model infer owners, dates, and claims you never gave it |
| Prompt | Role, context, constraints, and the exact output format | Constraints are what keep the model away from private data and decisions that need human judgment |
| Output | A named structure: action table, priority queue, extraction table, findings list, schedule, or content kit | A named format is far easier to check than paragraphs of prose |
| QA | A check that traces each item back to the source before you act | Names, dates, numbers, and commitments still need a human to approve them |
Ready-to-run template packs. The seven workflows below are the general shapes. These two packs are the filled-in versions, already written out prompt by prompt.
| Template | What it does | Best for |
|---|---|---|
| Founder Research Stack | Nine copy-ready prompts for market maps, competitor teardowns, pricing structure, positioning, and customer language mining, ending in a landing page draft | Founders, consultants, and operators deciding where to aim and what to charge |
| Claude Coding Prompt Pack | Twelve copy-ready prompts for debugging, code review, refactors, architecture reviews, and test design, with the constraints that stop a small fix becoming a rewrite | Developers who want a reviewable diff instead of a rebuilt file |
Inputs: transcript, rough notes, attendee list, meeting purpose, known deadlines, and any decisions you already trust. If the transcript is long, label sections by topic or timestamp before prompting.
Prompt: "Turn these meeting notes into an action plan. Context: The meeting purpose was [purpose]. Attendees were [names]. Notes/transcript: [paste]. Constraints: Do not invent owners or dates. Mark unclear items as needs confirmation. Separate decisions from discussion. Format: summary, decisions, action items table, risks, follow-up message."
Output: a five-part brief with a short executive summary, confirmed decisions, action table with owner/date/status, unresolved questions, and a ready-to-send follow-up email.
QA: Check every action item against the transcript. Remove anything the AI inferred too strongly. Confirm dates and owners before sending. If the output is too broad, rerun the prompt with one extra constraint: "Only include actions that someone explicitly agreed to do."
Workflow 2: email triage to reply queue
This workflow helps you process a crowded inbox without letting the AI speak for you unchecked. The useful output is a priority queue, not a batch of robotic replies.
Inputs -> prompt -> output -> QA
Inputs: sender, subject line, email body, relationship context, urgency, and your preferred response style. For sensitive threads, remove private data before pasting and ask the model to draft rather than send.
Prompt: "Triage these emails for today. Context: My role is [role], my priorities are [priorities], and my tone should be [tone]. Emails: [paste]. Constraints: Do not answer questions that require private account data. Flag anything that needs human judgment. Format: priority table, recommended action, draft reply for each email that can be safely answered."
Output: a table with priority, reason, next action, suggested reply, and risk flag. Billing issues, account access, and promises should be flagged for manual review.
QA: Read the priority reason first. If the model cannot explain why an email is urgent, demote it. Check all names, dates, promises, prices, and attachments. Keep a personal rule: AI can draft replies, but humans approve commitments.
Workflow 3: competitor scan to research brief
Use this as an AI workflow for research when you need a fast but traceable view of a market without drowning in tabs.
Inputs -> prompt -> output -> QA
Inputs: competitor names, website notes, pricing snippets, product pages, review excerpts, target customer, and the decision you are trying to make. You can also start from Whizi's Founder Research Stack when you want a more complete template.
Prompt: "Create a competitor scan for [market]. Context: I am deciding [decision]. Competitors and notes: [paste]. Constraints: Use only the supplied notes. Separate facts from inference. Do not invent pricing, customers, or features. Format: comparison table, positioning patterns, gaps, risks, and next research questions."
Output: a comparison table with competitor, audience, promise, features, pricing notes, proof points, weaknesses, and inferred positioning. The best version also includes a "what to verify next" column so the research does not pretend to be final.
QA: Trace every claim back to a note or source excerpt. Highlight unsupported inferences. For important decisions, run the same prompt in another model and compare which output separates evidence from opinion more cleanly.
Workflow 4: PDF summary to verified extraction
Generic PDF summaries are risky because they can miss caveats or overstate conclusions. This workflow asks for extraction and verification before any summary.
Inputs -> prompt -> output -> QA
Inputs: the PDF text or uploaded document, document type, your purpose, sections that matter most, and the exact output you need. If the document is long, ask for section-level summaries before asking for synthesis.
Prompt: "Analyze this document for [purpose]. Context: The document type is [type] and I care most about [topics]. Constraints: Quote or cite section names when possible. Mark uncertain items. Do not summarize sections you cannot see. Format: one-page summary, key facts table, risks/caveats, extracted dates/numbers/names, and verification checklist."
Output: a compact summary plus a structured extraction table. A research paper table might include claim, evidence, method, limitation, and confidence. A contract table might include obligation, party, deadline, section, and risk.
QA: Search the original PDF for every number, deadline, named entity, and quote. Ask a second pass: "List five ways this summary could be misleading." If the answer affects legal, medical, financial, or employment decisions, treat the AI output as a reading aid, not advice.
Workflow 5: code review to safer change list
This AI workflow for coding is useful when you have a diff, pull request, or file you want reviewed. The point is to reduce risk, not outsource judgment.
Inputs -> prompt -> output -> QA
Inputs: code diff, surrounding function or module, expected behavior, test output, error messages, constraints, and what should not change. Never paste secrets, production credentials, or private customer data.
Prompt: "Review this code like a senior engineer. Context: Expected behavior is [behavior]. Change goal is [goal]. Code/diff: [paste]. Constraints: Prioritize bugs, regressions, security issues, edge cases, and missing tests. Avoid style-only feedback unless it hides a bug. Format: findings table with severity, file/function, issue, why it matters, suggested fix, and test to add."
Output: a ranked review list, proposed tests, and a minimal change plan. For refactors, ask for the smallest readability improvement that preserves behavior, then list tests that prove it.
QA: Reproduce every bug before fixing when possible. Do not accept a suggested patch without reading it. Run tests, lint, and type checks locally. If two models disagree, use that disagreement as a review checklist rather than a vote.
Workflow 6: daily planning to focused schedule
A daily AI routine should help you choose what not to do. This workflow turns a messy list into a focused plan with realistic blocks, tradeoffs, and a shutdown checkpoint.
Inputs -> prompt -> output -> QA
Inputs: task list, calendar constraints, deadlines, energy level, meetings, must-do items, optional items, and one strategic goal. The more honest the inputs, the better the plan.
Prompt: "Plan my day around outcomes, not busyness. Context: Today's available work blocks are [blocks]. Must-dos are [must-dos]. Optional tasks are [optional]. Energy level is [energy]. Strategic goal is [goal]. Constraints: Protect deep work, include buffers, avoid overcommitting, and explain tradeoffs. Format: schedule, top 3 outcomes, defer list, risk list, shutdown checklist."
Output: a time-blocked schedule, three outcomes, a deferred list, and a final 10-minute shutdown routine. The defer list matters because it prevents the AI from making an impossible day look neat.
QA: Check whether the plan fits your actual calendar. Cut 20 percent if it feels too full. If a task has no success definition, ask the model to define "done" before you start.
Workflow 7: content repurposing to publishing kit
This AI workflow for writing turns one useful source into multiple publishable assets without making everything sound the same.
Inputs -> prompt -> output -> QA
Inputs: source content, audience, channel list, brand voice sample, claims that must stay accurate, claims to avoid, and the desired formats. Include examples of your best existing content if you want less generic output.
Prompt: "Repurpose this source into a publishing kit. Context: Audience is [audience]. Channels are [channels]. Voice sample: [paste]. Source: [paste]. Constraints: Preserve facts, avoid hype, do not add unsupported statistics, and make each channel native to its format. Format: core message, 5 post ideas, email draft, short social posts, long social post, newsletter section, and QA table."
Output: a content kit with reusable angles, channel-specific drafts, and a QA table listing original claim, reused claim, source location, and risk.
QA: Compare each draft to the source. Remove claims that were added for drama. Check whether the email, post, and article intro have different jobs instead of repeating the same sentence. Save the prompt as a repeatable ai prompt template only after it works on a real piece of content.
- Start each AI workflow with real inputs, not a vague request.
- Define the output before running the prompt: table, checklist, memo, draft, plan, or extraction.
- Add constraints for evidence, privacy, tone, length, and uncertainty.
- Ask for a QA step as part of the output, not after the fact.
- Save repeatable prompts by workflow, not by model: a Whizi project holds up to 10 pinned files and 32,000 characters of standing instructions.
- Run important workflows across more than one model and compare accuracy, structure, and cleanup time.
- Update each template after using it on a real task.
Frequently asked questions
What is an AI workflow template?
An AI workflow template is a repeatable process that defines the input, prompt, expected output, and quality check for a specific task such as summarizing PDFs, reviewing code, or triaging email.
How do I build an AI workflow for work?
Start with one recurring task, list the inputs it needs, write a prompt with constraints and output format, then add a QA checklist that verifies facts, decisions, names, numbers, and next steps.
Should I use the same AI model for every workflow?
No. Different models can perform better on different tasks. For important workflows, run the same prompt across models and keep the version that gives the most accurate, usable output.