AI Productivity · Field Guide

Getting Real Work Out of AI

One page of the things that actually matter — the reference from your training session, to come back to and keep using.

01 — Start Here

The tools all do roughly the same thing: you type what you want in plain English, they write back. You don't need a new language or special syntax. If you can describe the task to a sharp new hire, you can do it here.

There are four worth knowing. Pick one and get comfortable — they're more alike than different. Each has a free tier, plus a paid plan (around $20/month) that unlocks the smartest models and the image tools.

ChatGPT (OpenAI) — the most widely used. Great all-rounder for writing, summarizing, and quick answers.
Claude (Anthropic) — strong at longer documents, careful reasoning, and following detailed instructions. The one we tend to build on.
Gemini (Google) — lives alongside Gmail, Docs, and Drive, and is very good with images.
Copilot (Microsoft) — the same idea, built into Word, Outlook, and Excel.

The one-line takeaway: the free versions are enough to learn on. Once a tool starts saving you real time every week, the ~$20/month plan pays for itself almost immediately.

For Keystone Recognition: You're a Microsoft shop, so Copilot already sits inside Word, Outlook, and Excel — use it right there for work on your own files, where it has the context. For more general everyday AI work it can feel clunkier, so step out to ChatGPT, Claude, or Gemini. Right tool for the job.

02 — Talk To It Well

The single skill that separates a frustrating answer from a useful one isn't technical — it's how clearly you ask. A vague request gets a vague answer. The fix is to tell it three things: who it should be, what you want, and what "good" looks like.

Vague: "Write a tagline for a banner ad."
Clear: "You're a marketing copywriter for a custom-awards company. Write 5 short, punchy taglines for a web banner promoting engraved recognition awards for corporate clients. Under 8 words each, confident, not cheesy."

The four moves that fix almost everything:

Give it a role. "You're a detail-oriented proofreader…" instantly raises quality and tone.
Say what good looks like. Length, tone, format, who it's for.
Show an example. Paste one you like and say "in this style."
Then just talk to it. The first answer is a draft. "Shorter." "More formal." "Lose the third one." Keep going until it's right.

03 — Give It Context

The tool only knows what you tell it. It can't see your files, your past orders, or how your company does things — unless you put that in front of it. The more relevant material you give it, the more the answer sounds like your business.

One caveat: tools wired into your accounts — like Copilot inside Microsoft 365 — can already reach some of your files and email. Here we're mostly talking about the public chatbots, which know nothing about your business until you hand it to them.

Paste it in. Drop in the email thread, the notes, the messy list — then ask. "Here's the customer's email. Draft a friendly reply confirming the order and asking for their logo file."
Upload a file. Attach a PDF, spreadsheet, or image and ask about it. "Pull the order details out of this PO."
Keep going in the same chat. It remembers everything earlier in the conversation. Start a fresh chat only when you switch to an unrelated task.

Super-simple shorthand: Here is [the material]. I want you to [the task]. Make it [tone / length / format]. It's for [who].

04 — Reusable Assistants

If you find yourself typing the same setup over and over, save it once and reuse it forever. Every major tool lets you save a reusable setup — write the instructions, load the background, give it a name — and from then on it knows the role and context; you just hand it each new task. ChatGPT's version is a Custom GPT; both ChatGPT and Claude have Projects; Gemini has Gems.

Custom GPT vs. Project — same engine, different job.

A Custom GPT answers "how should it behave?" Built to be handed out — make it once, share a link, even on a personal paid plan. Think of it as a small app you built.
A Project answers "where does the work live?" A workspace holding your instructions and a shelf of background docs, where every chat inherits them. Sharing a Project across your team generally needs a business/team plan.

(Gemini's Gems are a third flavor — free to build and use — but they live in your own Gemini rather than being handed around.)

Good candidates to build: anything you do repeatedly with a consistent setup — drafting replies in your company's voice, turning POs into a clean checklist, writing product descriptions, proofreading to a house style. Start one for yourself; grow it into a shared tool when you know it earns its keep.

For Keystone Recognition: You don't have a shared company AI account yet — everyone's on their own login. So for now, anything you build is yours individually. If a custom assistant turns out to be worth sharing across the whole team, that's a small team-plan step you can take when the time comes — no rush. (Your build-your-own quick-start guide — included in this hub's download bundle — walks the exact steps.)

05 — Everyday Wins

You don't need a big project to get value today. These quietly give you back time — start with whichever you'd use this week.

Rewrite and polish. "Make this clearer and more professional, keep it short." Great for tricky customer replies.
Summarize the long thing. "Give me the three things I actually need to do from this."
Get unstuck / brainstorm. "Give me 10 tagline options." Beat the blank page — pick, don't accept wholesale.
Turn messy into tidy. "Turn this into a clean bulleted checklist" or "into a short table."
Explain it to me. "Explain this contract clause in plain English." A patient expert that never makes you feel dumb for asking.
Draft the first version. A proposal, a job post, an FAQ. Let it do the hard part — getting to a rough draft — then make it yours.

For Keystone Recognition: Sales — polish a tricky customer reply or draft a quote follow-up; production — turn a messy spec into a clean checklist; everyone — knock out the first draft of a vendor email or product description.

06 — Working With Images

Two very different things get lumped together as "AI images." Knowing which you need saves wasted time.

Generating from scratch. "Make me a picture of a mountain." Fun and fast — but it invents something that doesn't exist. Rarely what you need when the actual product has to be in the shot.
Editing a real photo. Start with your own product photo and change it — remove the background, drop it on clean white or into a new scene. The real product stays front and center.

Putting your product in a real scene — the most useful move for a product business: (1) start with the blank — a plain product photo on white; (2) add the branding — a logo wrapped to the surface, where it'll actually be decorated; (3) place it in context — "on a conference table in a bright office." Same product, now in a scene that sells it.

The standout tool for this is Google's Gemini — its image engine is unusually good at keeping a real product looking like itself while changing everything around it. It's not flawless — AI can nudge small details — so this is for concepting and listings, not final production art.

For Keystone Recognition: vendor blank → add the customer's logo (etch or UV) → drop it into the setting it'll be used in. It's mock-up ready, not production art (a designer still builds the real vector). Free is fine for concepting; once it's a regular part of making customer images, a small paid tier keeps output clean and your uploads private. (This is exactly what your Product Concept Studio Gem does.)

07 — Good Habits

Always read it before you send it. It writes confidently even when it's wrong. You're the editor. For facts, numbers, and names, verify.
Don't paste anything truly sensitive. Customer payment details, passwords, private personnel info — keep them out.
It's a draft partner, not the final word. The best results come from a back-and-forth, with your judgment on top.
If an answer's off, say so. "That's not quite right because…" gets you a better one. You can't break it — push back and keep going.

For Keystone Recognition: The everyday material is fair game — customer logos, art files, product specs, order details are how the work gets done. Keep the genuinely private things out: customer payment info, anything from HR or personnel, and pricing or margins you wouldn't want a competitor to see.

08 — Agentic AI

You'll hear "agentic AI" everywhere. Everything so far has been AI that answers. Agentic AI is the next step: AI that acts. You give it a goal, and it carries out the steps itself — using tools, moving between apps, working through the task and checking its own progress. You'll see it dressed up a dozen ways ("agents," "agent mode," "computer use," "Operator"); under the marketing it's all the same idea.

What you can do yourself — mostly connecting your apps and automating multi-step chores, much of it no-code. A few examples: a new order email lands → it pulls the details → adds a row to your spreadsheet → drafts the confirmation reply; or every Monday, gather the week's numbers, summarize, and email the team. The no-code tools: Zapier, Make, n8n, and Power Automate / Copilot Studio if you're a Microsoft shop.

A straight-up heads-up: even the "simple" personal stuff gets tricky — connections that won't authenticate, a step that quietly fails. This happens to people who build it for a living too. If a "quick" automation fights you, that's normal. Start small.

Start with the small wins. When it outgrows a weekend project, that's the cue — not your problem to brute-force.

For Keystone Recognition: Since you're a Microsoft shop, your natural on-ramp is Power Automate (and Copilot Studio for AI-driven agents). Good place to try a small "when an email comes in, do X" automation before anything bigger.

09 — From Here

All of this was high-level on purpose — a quick, referenceable cheat sheet. The real value comes from getting practical: what are you actually working on day to day, and how are you using AI on it right now? The way to get better is to do it live, together, so everyone picks up the nuggets they didn't know.

The cheat sheet is the map. Putting it to work is where it gets real.

Build your own: the step-by-step build-your-own-assistant quick-start guide is included in this hub's download bundle — it walks you through creating your first Custom GPT or Gem from scratch.