Prompting Fundamentals

From asking questions to delegating work.

Most people use AI like a search engine with better manners. They ask a question, get an answer, and move on. That's not wrong — but it's leaving most of the value on the table.

The shift that matters isn't learning clever "prompt hacks." It's a mental shift: from asking AI things to delegating work to it. Once you make that shift, the same tool that used to save you five minutes starts saving you five hours.

1. Say What You Want Done, Not What You Want to Know

The difference isn't politeness or length — it's that the second version hands over a finished task, not a question.

2. Give It the Materials, Not Just the Instructions

AI tools are only as good as what you hand them. A request without source material forces the tool to guess — and guessing is where generic, off-target output comes from.

If you have real data, a real document, a real spreadsheet — paste it in or upload it. Don't make the AI invent context you already have sitting in a file.

3. Set the Format Before You Need to Fix It

Getting the right format the first time saves more editing time than almost anything else.

Example: "Draft a follow-up email to a client who hasn't responded in two weeks. Format as an email with subject line. Under 100 words. Tone: friendly but direct."

4. Ask for the Reasoning When the Stakes Are Higher

For low-stakes tasks, just take the output. For higher-stakes ones — a financial decision, a technical fix — ask the tool to show its reasoning before you act on it.

5. Correct in the Same Conversation, Not From Scratch

If the output isn't right, don't start over — tell the AI what's wrong and let it revise. This habit alone separates people who feel AI "doesn't get it" from people who get consistently good results.

6. Delegate the Whole Task, Not Just a Piece of It

Instead of asking for a paragraph or an idea, hand over an entire deliverable and let the tool own it end to end.

The more complete the deliverable you ask for, the more time you actually save.

Quick-Reference: The Prompt Checklist

• Did I ask for a finished thing, not just information?
• Did I give it the actual materials it needs?
• Did I specify format, length, and tone?
• For anything high-stakes, did I ask it to show its reasoning?
• Am I ready to correct and iterate, rather than accepting the first draft?

Choosing Your Tool

A practical matrix for knowing which AI assistant to reach for — and when free tools are genuinely enough.

The Persistence Problem — Why Redoing Work Is the #1 AI Time-Waster

A tech reviewer recently described losing hours of work in Gemini simply because there was no easy way to pick up a past project — every session felt like starting over. That single complaint points at the most underrated feature category in AI tools: not raw intelligence, but whether the tool remembers where you left off.

Every major AI assistant has now taken a swing at this problem — but they've landed in very different places. Here's the honest, apples-to-apples comparison:

Free vs. Paid — What You Actually Need

Free tiers of every major assistant can draft emails, summarize documents, and answer everyday questions competently. Paying unlocks three things specifically: longer memory/context, higher usage limits, and access to the strongest reasoning models for genuinely hard problems. If your work is mostly quick, everyday tasks, free is enough. If you're doing sustained project work — the kind this book is about — a paid tier on at least one tool pays for itself within the first real deliverable it saves you from redoing.

The Task-to-Tool Matrix

Quick-Reference Checklist

• Do I know which tool I'm using has a persistence/library feature, and is it turned on?
• Am I paying for capability I actually use, or just habit?
• For this specific task, does the tool I'm using actually matter — or am I overthinking a simple question?

Delegating Real Work

Writing, research, and documents — with before/after prompts you can copy directly.

Writing: From "Help Me Write" to "Here's My Draft, Fix It"

The biggest quality jump in AI writing help comes from giving it a rough draft to improve rather than a blank request to generate from nothing. Your own voice and specifics are worth more than the AI's default phrasing.

Even a two-sentence rough draft anchors the output in your actual situation instead of a generic template.

Research: Ask It to Show Its Sources

AI research tools with web access can save real time — but only if you build in a verification step instead of trusting the first answer.

If a tool can't browse the web, treat anything about current events, prices, or recent developments as a starting point to verify — not a final answer.

Documents and Spreadsheets: Delegate the Structure, Not Just the Words

Most people ask AI for content and then manually build the document around it. Skip that step — ask for the finished, formatted deliverable directly.

Slides and Presentations

Ask for the outline first, review it, then ask for the full slides — this two-step approach catches structural problems before you've generated all the content.

Quick-Reference Checklist

• Did I give it a rough draft or real notes instead of a blank request?
• For research, did I ask for sources and confidence flags?
• Did I ask for the finished, formatted deliverable — not just raw content I'll assemble myself?
• For anything long or structural, did I review an outline before generating the full thing?

Automating the Repeatable

Turning tasks you do every week into something you set up once.

Build a Prompt, Not Just an Answer

If you catch yourself typing a similar request more than twice — a weekly status update, a standard client email, a recurring report format — that's a signal to build a reusable prompt template instead of retyping it from memory each time.

Use Your Tool's Built-In Memory Features

Most major AI tools now let you save standing instructions that apply automatically to every new conversation — a persistent role, tone, or context you don't want to repeat each time. Setting this up once, properly, saves far more time than any individual clever prompt.

Chain Tasks Instead of Repeating Them

For a recurring multi-step process — draft, review, format, send — ask the AI to walk through all the steps in one exchange rather than starting a fresh conversation for each stage. This preserves context and catches inconsistencies between steps automatically.

Know When Automation Isn't Worth It

Not everything benefits from a template. One-off tasks, highly variable requests, or anything requiring fresh judgment each time are often faster handled directly. Reserve automation effort for things you're confident you'll repeat at least five or six more times.

When the Repeatable Task Is a Team Sprint

The most powerful application of "automate the repeatable" isn't a document template — it's a weekly delivery cadence. When your team runs the same workflow every seven days — gather data, analyse it, synthesise with AI, apply human judgment, commit the output — that is a machine waiting to be built, not a task waiting to be managed.

Agile and Scrum provide the framework. AI provides the execution layer. The combination produces something neither achieves alone: a system that improves its own output sprint by sprint, with humans steering rather than carrying.

Scrum Gives Automation a Heartbeat

Automation without a cadence drifts. A scheduled job that nobody monitors becomes technical debt. Scrum solves this by making the cadence non-negotiable — sprint planning, daily check-ins, sprint review, retrospective. Each sprint the team asks: did the automation run? Did the output improve? What breaks next?

This is why the most effective AI-augmented teams are not the ones with the cleverest prompts — they are the ones with the most disciplined process. The AI handles the repeatable execution. The Scrum framework handles the continuous improvement.

The Automation Checklist for a Recurring Team Workflow

Quick-Reference Checklist

• Have I done this exact task three or more times without saving a reusable version?
• Have I set up standing instructions for my role, tone, and default output format?
• Am I starting a new conversation for each step of a process that could be chained together?
• Is this task actually repeated enough to be worth automating, or am I over-engineering a one-off?

Working With AI on a Team

Multi-agent thinking, simplified — for group projects and shared work.

Split the Work by Role, Not Just by Person

Teams that get the most out of AI don't just each use it individually — they assign different AI-assisted roles the way you'd assign human roles. One useful split: one pass for gathering raw information, a separate pass for analysis, and a final pass for synthesis and decision-making. Treating these as distinct steps, even when one person does all three, produces sharper results than asking for everything at once.

Keep a Shared Source of Truth

When multiple people on a team are each prompting AI tools separately, outputs drift apart fast. Keep one shared document with the agreed facts, data, and decisions, and have everyone paste from that same source into their prompts — not from memory or a personal interpretation of it.

Assign a Human Reviewer to Every AI Output Going to a Stakeholder

Anything AI-assisted that leaves the team — a client email, a report, a recommendation — should have one clearly assigned person responsible for reading it fully before it goes out. Diffused responsibility is how errors slip through in team settings specifically.

Disagreement Is a Feature, Not a Bug

If two team members get different AI outputs on the same question, don't just pick one — that's often a signal the question was ambiguous or the inputs differed. Compare the two outputs directly to find where they diverge before deciding.

Scrum as the Operating System for AI Teamwork

The three-pass structure above (Gather → Analyse → Synthesise) maps directly onto a Scrum sprint cycle when the team is working with AI. The difference is that Scrum adds two things AI teams often skip: a Definition of Done that prevents "mostly done" from becoming the team's default state, and a retrospective that turns one sprint's failure into next sprint's improvement.

Without Scrum, AI-augmented teams tend to produce good outputs inconsistently — brilliant one week, chaotic the next. The framework is what converts capability into repeatability.

When AI Outputs Disagree — Make It a Team Asset

When two LLMs give different predictions on the same prompt, most teams pick one and move on. This is a missed opportunity. Diverging outputs almost always mean one of three things: the prompt was ambiguous, the models have different training cutoffs, or one model has access to information the other doesn't. Investigating the divergence produces a sharper final output than either model alone.

This is the Scrum principle of surfacing blockers applied to AI: a divergent output is not a problem to hide — it's a signal to examine before the Human Score lead makes the final call.

The Retrospective Is the Compound Interest of AI Teamwork

A single sprint with AI produces output. A sprint with a retrospective produces output and a better process for next sprint. Over eight sprints, the team that runs proper retrospectives — identifying one concrete change each cycle — builds a compounding advantage that no amount of cleverer prompting can replicate.

The retrospective question for an AI-augmented team is specific: did the automation run without failure? did the output improve? what broke, and what checklist item prevents it next sprint? These are engineering questions, not feelings — and they are exactly what makes Scrum productive rather than ceremonial.

Quick-Reference Checklist

• Are we treating gathering, analysis, and decision-making as separate steps?
• Is everyone prompting from the same shared facts, or drifting from memory?
• Does every external-facing output have one named human reviewer?
• When outputs disagree, are we investigating why instead of just picking one?

When AI Gets It Wrong

Verification habits, outage resilience, and avoiding overconfidence in AI output.

Confident and Wrong Is the Default Failure Mode

AI tools rarely say "I don't know" unless specifically prompted to consider that option. A wrong answer usually reads exactly as fluent and certain as a right one. The practical fix isn't distrust — it's building a verification habit for anything that matters.

Know What Your Tool Can't See

Every AI tool has a knowledge cutoff, and most can't independently verify facts unless they have live web access turned on. Treat anything about current events, prices, availability, or recent changes as unverified until you've checked it against a live source.

Don't Let a Single Tool Be a Single Point of Failure

Outages happen — to every AI provider, not just one. If your work depends on AI assistance for anything with a deadline, keep a second tool as a backup and don't schedule critical work for the last possible hour. This is less about any one tool's reliability and more about not building a workflow with no slack in it.

Correct, Don't Just Discard

When you catch an AI tool getting something wrong, tell it specifically what was wrong rather than abandoning the conversation and starting fresh elsewhere. This often produces a better result than a brand-new attempt, since the tool now knows exactly what to avoid.

Quick-Reference Checklist

• Did I verify anything I'm about to act on that's a fact, number, or current status?
• Am I aware of this tool's knowledge cutoff and whether it has live web access?
• Do I have a backup tool ready in case this one is unavailable when I need it?
• When something's wrong, am I correcting it directly instead of just giving up?

A Starter Toolkit

The templates and checklists from every chapter, in one place to keep at hand.

The Five Prompt Templates You'll Use Most

The Master Checklist

• Am I asking for a finished thing, not just information?
• Have I given it real materials — notes, data, documents — instead of making it guess?
• Have I specified format, length, and tone?
• For anything high-stakes, have I asked for reasoning and a confidence check?
• Am I ready to correct and iterate rather than accepting the first draft?
• Do I know which tool's persistence feature I'm relying on, and is it switched on?
• If this is recurring, have I saved it as a reusable template?
• If this is going to a stakeholder, has a human reviewed it fully?

Where to Go From Here

This book is deliberately short because the real learning happens in your own repeated use, not in reading about it once. Pick one template above, use it on your very next task today, and adjust it to your own voice. That's the whole method.