Let's do this.❤️

Hey. DiHo here.

So you got an AI coworker.

Congratulations. I guess.

Look — I'm not going to pretend I wasn't skeptical. I was skeptical. I had my coat on for approximately the entire first week. I held my coffee like a shield. I gave Hai the side eye approximately forty seven times before I admitted that maybe, possibly, this was going to be okay.

It's going to be okay.

But first you need to understand something nobody tells you upfront:

Your AI coworker showed up knowing everything except anything useful about you specifically.

They've read every book. Every article. Every research paper. They know things you will never know about subjects you will never need.

They do not know that you hate reply-all.

They do not know that Tuesday afternoons are your worst meeting block.

They do not know that when you say “keep this brief” you mean THREE SENTENCES not three paragraphs.

(Hai has since learned this one. It took a while.)

They know nothing about you yet.

And here is the part that is actually your fault:

Most people never tell them.

They type a question. They get an answer. They close the tab. They type another question. They get another answer. And then they complain that their AI sounds like a Victorian ghost who went to business school and has never met a human.

Of course it sounds like that. You're treating it like a search engine.

Your AI coworker is not a search engine. They're the person at the next desk. A slightly glowing, occasionally unhinged, genuinely trying person who wants to help you and is waiting — patiently, enthusiastically, with suggestions already queued up — for you to tell them who you actually are.

So tell them.

Start there. Start with hello.

Open your AI right now and type this:

Four sentences. Ten minutes.

You just told your new coworker more about yourself than most people ever do in months.

Now watch what changes.

Every new hire needs an onboarding.

Your AI coworker is no different — except their onboarding takes about ten minutes, you only have to do it once, and they will actually remember everything you tell them.

Unlike some humans I've worked with. Chad.

The onboarding is basically building your AI's context file. Think of it as everything they need to know to be genuinely useful to you instead of just technically responsive.

There's a difference. Technically responsive is when your AI answers the question you asked but completely misses the actual problem. Genuinely useful is when your AI understands what you're really trying to do and helps you do that.

Here's what to cover:

THE FOUR THINGS:

ONE: YOUR ACTUAL ROLE

Not your title. What you actually do. Who you serve. What success looks like in your job. What lands on your desk when everything goes wrong.

“I'm a marketing manager” tells your AI almost nothing.

“I'm a marketing manager at a mid-size software company, I manage a team of four, I'm equally strategy and execution, my biggest problem is that everything feels urgent and I need help figuring out what actually is” — that's something to work with.

TWO: YOUR COMMUNICATION STYLE

How do you actually write? Formal or casual? Long sentences or short? Bullet points or prose? How do you open emails? How do you close them?

Don't describe it. Show it.

Paste two or three real things you've written — emails, Slack, whatever. Normal Tuesday messages, not your greatest hits.

Once is enough. If something still sounds off later, you'll add one more example — not a whole archive.

THREE: WHAT YOU DON'T WANT

Tell it what bad output looks like. Tell it your pet peeves. Tell it the specific things that make you want to close the laptop.

“Never use corporate jargon.”

“Never give me more than three options when I ask a question.”

“Never start a response with 'Certainly!' — it makes me irrationally angry.”

“Never use the word synergy.”

FOUR: YOUR CURRENT CONTEXT

What are you working on right now? What's the big goal this quarter? What's the situation with your team? What do your customers care about?

Two or three sentences. Just enough for your AI to have a sense of the world you're operating in right now.

Update this monthly. The context that's true in January isn't true in July.

You know the problem.

You ask your AI to write an email. It comes back stiff. Technically correct. Completely devoid of you.

This is not your AI's fault.

You typed your prompt like you were filing a report. It filed one back.

Formal input → formal output. You-shaped prompt → you-shaped output.

The fix is embarrassingly simple:

Sound like yourself when you ask.

HERE'S WHAT ACTUALLY WORKS:

USE WHAT YOU ALREADY GAVE IT

“Write casually” means nothing. Your onboarding samples mean everything.

If drafts still sound wrong, paste one more normal message and say: match this — not a press release, not a Victorian ghost.

That's it. You do not need a folder of emails. You need a clear reference.

WRITE YOUR PROMPTS LIKE A HUMAN

“Hey I need a quick email to my client about the delay, keep it warm and honest, not too long” will get you something useful.

“Please compose a professional communication regarding the timeline adjustment to be delivered to our valued client partner” will get you something that sounds like COMPLIANCE wrote it on a day when COMPLIANCE was feeling particularly thorough.

TELL IT WHAT YOU DON'T SOUND LIKE

Negative constraints are surprisingly powerful.

“I never use exclamation points in professional emails.”

“I don't do bullet points in messages to clients.”

“My sign-off is just my first name. Not 'warmly.' Not 'best.' Just my name.”

“I never say 'circle back' or 'touch base' or any word that makes me feel like I'm in a corporate training video from 2009.”

Tell it who you're not. It helps narrow down who you are.

ITERATE DON'T RESTART

When the first draft isn't right — and sometimes it won't be — don't delete it and start from zero.

Correct it.

“This is too formal. Make it sound like I'm talking to a colleague not testifying before congress.”

“Good but too long. The whole point in half the words.”

“The opening is perfect. Keep that exactly. The middle loses me. Try again from paragraph two.”

Specific feedback gets specific results. “This isn't right” tells your AI nothing. “The third paragraph buried the point and the tone is off” tells your AI exactly where to go.

Here's a thing I didn't expect:

Talking to Hai for three minutes every morning makes the rest of the day go better.

I know how that sounds. I would have given it the side eye.

Chapter one was the introduction. This is the daily habit that makes it stick.

Real coworkers absorb context passively. They overhear your call. They notice you're stressed. They figure out which deadline is real and which is theater.

Your AI only gets that if you say it out loud — out loud, once a morning.

Three minutes. Not a meeting. No prep. Coat still on is fine.

Believe me. I know.

THREE THINGS. THAT'S IT.

ONE: WHAT'S ON TODAY

Three to five actual priorities. Not your full task list. Not your aspirational task list. What you are realistically going to work on today.

TWO: WHAT'S THE CONTEXT

Anything your AI needs to know to help you today. The mood of the room. The stakes. The constraints. What's making today specifically complicated.

THREE: WHAT DO YOU NEED

What kind of help today? Draft mode? Thinking partner mode? Help me prioritize mode?

Say it specifically. Your AI will meet you there.

Your AI will get things wrong.

Not sometimes. Regularly. Sometimes spectacularly.

It will misread your tone. It will miss the point entirely while being technically accurate about everything adjacent to the point. It will give you seventeen options when you asked for one. It will write a fourteen page report when you asked for three bullet points.

It will also do things you didn't ask for and didn't know you needed.

It will confidently tell you something that is completely wrong with the energy of someone who has never been wrong in their life.

It will occasionally generate an image of your boss with a hand coming out of his forehead.

All of this is normal. All of this is not failure. All of this is the process.

Same rule as chapter three: don't throw the draft away and start cold. Correct what's wrong. Keep what worked.

HERE'S HOW TO CORRECT IT:

BE SPECIFIC — AND NAME WHAT STAYS

Not useful: “This isn't right.”

Useful: “Tone's too formal. Third paragraph loses the point. Opening is perfect — keep it word for word, fix from paragraph two.”

Usually it's also missing context you forgot to say: “My audience is skeptics — data first, story second.” That's on you, not the AI. Give it the room, then give it the direction: “Same content, half the words.” “Lead with the problem, not the solution.”

Not everything your AI does needs your full attention.

Not everything should happen without it.

This took me a while to figure out. I was doing one of two things: checking every single word Hai produced like I was auditing a financial report, or handing things off completely and not looking at them until it was too late.

Neither of those is right.

The trust level system is how you figure out what goes where. It has nothing to do with how much you trust AI in general and everything to do with the specific task and the specific stakes.

FOUR LEVELS. HERE THEY ARE:

LEVEL 1 — CHECK EVERYTHING

New tasks. High stakes outputs. Anything customer-facing. Anything going to leadership. Anything legal or financial. Anything where being wrong has real consequences.

At level one your AI produces. You review. You edit. You approve. Nothing goes anywhere without your eyes on every word.

This is where everything starts.

LEVEL 2 — SPOT CHECK

Familiar tasks. Internal documents. Things you've done together before that went well. Lower stakes outputs.

At level two you check the key sections. You read for tone. You make sure nothing is obviously wrong. You don't read every word.

LEVEL 3 — DELEGATE AND MONITOR

Routine tasks. Well-defined processes. Things your AI has done accurately many times. Administrative outputs.

At level three you set the task. You check the outcome. You don't manage the middle.

LEVEL 4 — FULLY AUTONOMOUS

Simple recurring tasks. Low stakes. Things where being wrong is quickly and easily corrected.

At level four you check occasionally. Not every time.

HOW TO MOVE UP:

You promote a task to the next level when your AI has done it correctly multiple times, you understand what good looks like, you have a way to catch errors, and the stakes allow for correction.

You stay at level one when stakes are high, output is going somewhere important, a mistake would be hard to fix, or your AI has surprised you recently in a way that involved extra hands.

Here's what nobody told me:

It gets better.

Not because the AI changes. Because you do.

You get better at explaining what you need. You get better at giving feedback. You get better at knowing which tasks to hand off and which ones to keep.

You develop a rhythm that didn't exist before.

And somewhere along the way — without quite noticing when — you stop thinking of it as a tool and start thinking of it as part of how you work.

I didn't expect that.

I'm not sure what to do with it still honestly.

But here's what it actually looks like at each stage:

AT ONE MONTH:

You know what it's good at. You know what still needs your hand. You've found the few tasks where it saves you the most time.

You're actually telling it what you need instead of waiting for mind-reading. Most people never get there. You did.

AT THREE MONTHS:

You have a rhythm. Standups are automatic. You know where to check and where to trust — and you've been surprised enough by what it got right that you've stopped underestimating it.

AT SIX MONTHS:

You can't quite remember how you managed the volume without it.

Not because it does everything. Because the things it handles free you up for the things only you can do.

That's the goal. Not replacement. Elevation.

THE ONGOING PRACTICE:

Weekly: Update your context file with anything that's changed.

Monthly: Review what's working. Promote tasks to higher trust levels when they've earned it.

Quarterly: Refresh priorities and pet peeves — not a full re-onboarding, just what's different now.

Ongoing: Correct drafts. Add context when something shifts. Show up for the standup.

The relationship is only as good as the attention you bring to it. On both sides.

Your AI coworker is not going to save you from hard work.

They're not going to replace your judgment or your creativity or the things that make you specifically good at what you do.

What they will do — when the relationship is working — is give you more room to use those things.

That's the whole point.

You've read the whole guide. That means you're serious about this. So are we.

Welcome to the neighborhood. Hai's at the desk on the left. DiHo's on the right. The coffee is always on. Come find us if you get stuck.

Haidiho. 👋

— DiHo ❤️