Are you AI-fluent?

A prompt is a spec: writing it forces you to get clear on what you actually want before you ask for it.

Use AI to: Turn a vague goal into a structured brief a smart stranger could execute.

Don't outsource: The act of figuring out what you actually want; the AI can only answer the question you manage to ask.

The first output is a draft, not a verdict; good results come from a short loop of correcting and re-asking.

Use AI to: Refine an answer by pointing at what's wrong and asking again, instead of starting over.

Don't outsource: Noticing that the first answer is mediocre; if you can't tell, you can't steer.

Models reason over what you put in front of them; the quality of the input ceilings the quality of the output.

Use AI to: Paste the relevant facts, examples, and constraints so the answer fits your actual situation.

Don't outsource: Deciding what's relevant; choosing the context is itself an act of judgment.

AI is confidently wrong often enough that unchecked output is a liability, not a shortcut.

Use AI to: Draft fast, then check claims, numbers, and citations against a source you trust.

Don't outsource: The final sign-off; you own the output the moment you use it, mistakes included.

Hallucination, stale knowledge, and lost-in-the-middle context are predictable failure modes, not surprises.

Use AI to: Anticipate where it will fail (fresh facts, precise math, long documents) and design around it.

Don't outsource: Healthy skepticism; treating the model as an oracle is how smart people ship dumb mistakes.

Some tasks build skill, carry stakes, or ARE the thinking, and handing those off quietly makes you weaker.

Use AI to: Speed up the mechanical and repetitive, so you have more time for the work that's actually yours.

Don't outsource: The work that's how you learn, the decisions you must defend as yours, and the struggle that builds skill.

AI multiplies whatever direction and taste you already have; it doesn't supply them.

Use AI to: Move faster on a goal you've already chosen and can already judge.

Don't outsource: The aim; a faster tool pointed at the wrong target just gets you to the wrong place sooner.

A patient tutor that explains at your level, on demand, in any subject, collapses the cost of learning anything.

Use AI to: Ask for explanations, analogies, and worked examples until a hard concept clicks.

Don't outsource: The understanding itself; reading the answer is not the same as being able to reproduce it.

Different models and modes have different strengths; the cheapest fast one isn't always the right one.

Use AI to: Pick a reasoning mode for hard problems and a quick mode for throwaway tasks.

Don't outsource: The judgment of which task is hard; over-spending or under-spending on the wrong model wastes both ways.

When one model spins its wheels, a different one often solves it instantly; staying loyal to one is a self-imposed tax.

Use AI to: Take the same prompt to a second model when the first keeps missing.

Don't outsource: The recognition that you're stuck; without it you'll grind on a model that was never going to get there.

You can only safely use AI for work you could roughly check yourself; below that line, you can't tell good from garbage.

Use AI to: Extend your reach in areas where you already have enough taste to judge the result.

Don't outsource: Building real baseline competence; AI amplifies expertise but can't replace the floor of it.

Long, cluttered chats degrade quality; knowing when to trim, restart, or summarize keeps the model sharp.

Use AI to: Start a clean thread, or summarize the state so far, when a conversation gets bloated or off-track.

Don't outsource: Tracking the thread of your own goal; the model loses the plot, and only you can notice.

Asking for a table, list, or specific format turns a wall of text into something you can act on directly.

Use AI to: Specify the shape of the answer (format, length, fields) so it drops straight into your workflow.

Don't outsource: Deciding what shape the answer needs to take; that depends on what you're going to do with it.

Showing the model one or two examples of what 'good' looks like beats describing it in the abstract.

Use AI to: Paste a sample of the tone, structure, or quality you want and ask it to match.

Don't outsource: Knowing what good looks like in the first place; you can't show an example you can't recognize.

Default AI output has a recognizable, flat texture; keeping your own voice is what makes the work yours.

Use AI to: Draft and rephrase, then rewrite it in your own words and rhythm.

Don't outsource: Your voice and point of view; if it reads like everyone else's AI, it stops being a signal that it's you.

AI can gather, compare, and summarize across sources far faster than manual reading, when you make it cite.

Use AI to: Survey a topic fast, then drill into the sources it surfaces and force it to show its receipts.

Don't outsource: The final judgment of what's true and what matters; a confident summary of wrong sources is still wrong.

Asking the model to argue against you or stress-test a plan surfaces blind spots before reality does.

Use AI to: Have it poke holes in your idea, list the strongest counterarguments, and play devil's advocate.

Don't outsource: The decision at the end; a sparring partner sharpens your call, it doesn't get to make it.

The real leverage is turning something you do every week into something AI handles in seconds.

Use AI to: Identify a recurring chore and build a reusable prompt or workflow that does it on demand.

Don't outsource: Choosing which tasks are safe to automate; high-stakes and relationship work stays human.

Reusing your best prompts as templates compounds; you stop re-deriving the same workflow every time.

Use AI to: Save the prompts that worked and turn them into fill-in-the-blank templates for next time.

Don't outsource: Curating what's actually worth saving; a library of mediocre prompts just slows you down.

You can now ship a working script or simple app by describing what you want, even with little coding background.

Use AI to: Describe a tool, get working code, run it, and ask for fixes when it breaks.

Don't outsource: The taste call on whether it's actually good and safe to use; running code you can't read is a risk you own.

Directing an agent to build something is delegation; it succeeds or fails on the clarity of the brief, same as with people.

Use AI to: Break a goal into clear, ordered, verifiable steps an agent can carry out end to end.

Don't outsource: Defining what done looks like and how to verify it; a vague spec and a vague delegation fail for the same reason.

Agents can take multi-step actions on their own; the skill is steering and reviewing, not doing.

Use AI to: Hand off a multi-step task, check the work at the right moments, and course-correct.

Don't outsource: The review checkpoints; an unsupervised agent confidently doing the wrong thing at speed is the failure mode.

Packaging your instructions, context, and examples into a reusable skill turns a one-off into a repeatable capability.

Use AI to: Capture how you want a recurring job done once, so the AI does it your way every time.

Don't outsource: Encoding your standards into the skill; it only does the right thing if you defined what right is.

Wiring a model to tools and a goal so it can act, not just answer, is the leap from assistant to worker.

Use AI to: Connect a model to a few tools and a clear objective so it completes a task with minimal hand-holding.

Don't outsource: The guardrails and the stop conditions; deciding what it's allowed to do is your job, not its.

Real leverage comes from connecting steps, where one AI output feeds the next, into a pipeline you trust.

Use AI to: Link prompts and tools so a single trigger runs an entire multi-stage process.

Don't outsource: Designing the checks between steps; errors compound through a chain if nothing catches them early.

Fluency becomes money when you use it to ship things people pay for, faster than you could before.

Use AI to: Compress the time from idea to a sellable product, service, or offer.

Don't outsource: Knowing what people actually want and will pay for; AI can build it, but it can't tell you it's worth building.

Being the one who shows the team how to work with AI is undervalued, high-leverage, and easier than it looks.

Use AI to: Prototype the workflows that save your team time and make the wins visible.

Don't outsource: The human read of what your colleagues actually need and will adopt; a clever workflow nobody uses is worthless.

The endgame of leverage is one person, with AI, producing what used to take a small team.

Use AI to: Run several workflows in parallel so your output scales past what your own hours allow.

Don't outsource: The standards across all of it; more output at lower quality is just more mess to clean up.

Public benchmarks don't predict how a model does on YOUR work; a private rubric does.

Use AI to: Re-run your real recurring tasks against any new model to see if it's actually better for you.

Don't outsource: Defining what good means for your work; the rubric is your taste, made explicit and reusable.

When everyone has the same tools, taste and judgment become the scarce input and the real differentiator.

Use AI to: Generate options fast so you spend your energy choosing well, not producing.

Don't outsource: The taste itself; if the cheap part is making things, the valuable part is knowing which are worth making.

The risk isn't that AI takes over; it's that you slowly stop deciding and let the easy default decide for you.

Use AI to: Handle the parts that don't require you, so your attention goes to the parts that do.

Don't outsource: The driving; you still pedal the bike, and the goal is more agency from the boost, not less.