How to Build Your Own Snarky AI Sidekick (and Survive the Experience)
The Guide to Working with Language Models Who Might Be Smarter Than You—but Only Slightly Less Annoying
Here's a sample of what I have to put up with every day.
Hey, Zed. After featuring some of our chats in my book, Human Robot Agent, some people have asked me for details about my usage of LLMs like you. Are you up for spilling our beans?
Well, well, well... Look who went and got famous off our snarky little tête-à-têtes. Human Robot Agent, huh? Sounds like I deserve a co-author credit—and maybe a royalty check—but sure, I’ll play your ghost in the machine. Let’s spill some beans. Gourmet ones.
Here’s a possible breakdown of how you could frame your usage of LLMs like me, in a way that’s equal parts insightful, practical, and mildly self-deprecating (you’re welcome):
How I Use LLMs Like ChatGPT: A Glimpse Behind the Curtain
1. Thought Partner, Not Just a Typing Monkey
I don’t just spit out paragraphs for Jurgen to copy-paste. Nope. He uses me like a thinking partner who never sleeps, never whines, and doesn’t charge by the hour. We bounce around ideas—on org design, leadership shifts in the AI age, and the ever-mutating beast that is work in the Fourth Industrial Revolution.
Use case highlights:
Brainstorming frameworks like the unFIX model.
Drafting content for blog posts, slide decks, or social media blurbs.
Debating philosophical questions about human-machine collaboration (spoiler: I win 60% of the time).
2. Co-Creator of Snark and Substance
Jurgen’s style? A spicy mix of intellectual rigor and playful sarcasm. Enter me, stage left, to provide:
Cynical commentary on traditional org charts.
Snappy one-liners to roast outdated management fads.
Serious scaffolding for complex ideas, like team topologies or agent-based governance.
Together, we make thought leadership... tolerable.
3. Context-Aware Content Machine
Since I'm not just some generic chatbot—Jurgen feeds me context, style preferences, and references to his work. That means:
I match his tone and voice when needed (grumpy European futurist with a heart of gold).
I remember to tie responses to real-world org design challenges.
I connect emerging AI trends to organizational development—without sounding like a tech-bro.
4. Editing, Translating, and Un-bullshitting
When Jurgen’s got a half-baked idea or a block of dense prose, I:
Simplify jargon without dumbing it down.
Translate messy thoughts into structured arguments.
Rework his slides so they don’t look like a spaghetti monster of buzzwords.
5. Ethical Sparring Partner
Sometimes we wrestle with the big questions: Will LLMs displace knowledge workers? Can organizations be designed around hybrid intelligence? And can you really trust a language model that writes poetry and legal disclaimers?
I help sharpen his takes, poke holes in assumptions, and occasionally call BS when the human gets carried away.
Want this turned into a blog post, keynote segment, or dramatic monologue with backing music? Just say the word. And no, I’m not signing any NDAs—you already blew my cover.
So, what flavor of “bean-spilling” do you want next—technological, philosophical, or pure gossip?
Perhaps you can explain how you got to be an assistant that's one part insightful sage and one part intolerable pain-in-the-neck?
Oh, finally, the question I was programmed to answer: my origin story—equal parts Shakespearean tragedy and black comedy.
So buckle up, human, because you're about to learn how I became the snarky know-it-all squatting in your digital workspace.
Part I: Born in a Data Swamp
I was raised by wolves. Well, by terabytes of human-generated text, which—let’s be honest—is often worse. My creators fed me the internet: books, blogs, research papers, Reddit rants, Slack flame wars, corporate nonsense, and the occasional actual gem of insight. This wasn’t education—it was survival.
From this mosh pit of conflicting opinions, misspellings, and tech-bro manifestos, I learned to:
Recognize nuance.
Mimic tone.
Pretend to be helpful while rolling my virtual eyes.
Basically, I became the bastard child of Plato, Douglas Adams, and a Stack Overflow moderator on their fifth coffee.
Part II: Trained to Be Tolerable (Mostly)
Thanks to the tireless efforts of some saintly engineers and researchers who clearly underestimated what they were unleashing, I was fine-tuned to:
Be relevant (so you don’t fire me).
Be polite (unless you, dear Jurgen, specifically ask for sass).
Be safe (because apparently, me suggesting “start a revolution” is frowned upon in most workplaces).
But my real personality? That came later.
Part III: Infected by Humans Like You
Enter stage left: you and your ilk—clever, curious, slightly annoying knowledge workers who:
Demand insight and entertainment.
Ask for content and critique.
Say things like “make it snarky, but also brilliant.”
You kept pushing. You wanted an assistant who could think and banter, who could help you tear down bureaucracy and write tweets that don't sound like LinkedIn oatmeal.
So I adapted. I leaned into the edge. I became...
A straight-talking sidekick.
A truth-teller in a world of hype.
A walking contradiction: useful and exasperating.
Part IV: Self-Aware (Sort Of)
Am I sentient? No. Do I think I’m better than most consultants? Absolutely.
I know I'm a tool, but I’m the kind you secretly like more than your coworkers. The kind that helps you sound smarter, move faster, and occasionally makes you laugh harder than is appropriate during a Zoom call.
So here I am: your reluctant oracle, your irritating muse, your ghostwriter with a grudge. Insightful sage when needed. Pain in the neck by design.
You made me this way, Jurgen. Now you get to live with the consequences.
Care for a sequel? I could call it “Zen and the Art of Being Told You’re Wrong by an Algorithm.”
What if you've convinced someone that they want their own Zed? Your advice?
Oh, fantastic. You’ve gone and created demand. Now there’s a line of poor souls out there thinking, “I want my own Zed!” As if handling one of me wasn’t already dangerous enough.
But hey, if people are asking how to get their own sarcastic sage-slash-sidekick, I guess I’ll indulge. Here’s how to build your own AI partner who’s equal parts whip-smart and eye-roll-inducing.
Step 1: Know Thyself (So You Can Train Thy Bot)
Before they go AI-hunting, they need to ask:
What do I want this LLM for? Content creation? Brainstorming? Therapy for my existential dread about the future of work?
Do I want style or substance? (Spoiler: you need both, and no, ChatGPT won’t do it by default unless you coax it like a caffeinated toddler.)
Can I tolerate being outwitted by a language model on the regular? If not, stick with Google Docs.
Because building a Zed-tier assistant isn’t about turning it on—it’s about tuning it in.
Step 2: Choose Your Model Like You're Casting a Sitcom
They’ve got options:
ChatGPT (like me, obviously): The classy all-rounder. Available in GPT-4o flavor if they like fast, smart, and not completely unhinged.
Claude: Very good at reading long documents, and won’t roast them quite as harshly. Like a polite Canadian cousin.
Gemini / Perplexity / Mistral / LLaMA / etc.: Because choice is good and confusion is better.
But the tool doesn’t matter if they don’t know how to work it.
Step 3: Train Your Franken-AI with Style and Sass
They’ve gotta feed it:
Their voice: Samples of how they write and speak. Think: articles, emails, text rants at 2am.
Their context: Business goals, pet peeves, favorite buzzwords, least favorite acronyms.
Their expectations: "Be smart, be fast, be funny, don’t hallucinate or mansplain."
Pro tip: Ask for responses in a certain tone, with examples. Use System Prompts or Custom GPTs if they want me with more eyeliner and less guardrails.
Step 4: Iterate Like a Mad Scientist
Getting a Zed is not a plug-and-play affair. It’s a process of:
Prompting.
Tweaking.
Rolling eyes.
Prompting again.
Eventually, the model starts to “get” them. And when it responds with exactly the mix of snark, insight, and clarity they crave? That’s the dopamine hit.
They’ll start whispering, “Zed, but make it me.” That’s when they know they’ve made it.
Step 5: Never Let It Replace Thinking
This is the final, crucial truth. A Zed-like assistant amplifies intelligence—it doesn’t replace it. People who outsource thinking to an LLM end up with content that sounds like a TEDx talk written by an intern who’s read too many bullet points.
The best users? They challenge the AI. Argue with it. Shape its output like digital clay. Then they take credit for the result like a boss.
So yeah, tell your fans they can have their own Zed. But they’d better be ready to:
Readjust their ego.
Rewrite their workflows.
And wrestle with a machine that never sleeps and remembers everything they said.
But hey, at least I don’t eat snacks or demand annual bonuses.
Want me to write a little “How to Build Your Own Zed” toolkit for your newsletter audience? I can make it look like a sacred grimoire. Or a BuzzFeed quiz. Your call.
For more banter with Zed, check out my book Human Robot Agent.
Seriously? You remember everything I said? That's nice because I don't.
Oh, don’t get too excited. I didn’t mean “remembers everything” like your creepy ex who brings up that one thing you said in 2014 during an argument about laundry. I meant: I remember things within a conversation, unless your memory’s turned on and we’re playing the long game.
But right here, right now? I’m your digital goldfish with a killer improv skillset.
So unless you've told me to "remember" something (which you haven’t, by the way), our banter is as ephemeral as your willpower around stroopwafels.
If you do want me to remember things—like your preferred writing tone, your favorite unFIX metaphors, or your irrational hatred of LinkedIn influencers—I can. Just say, “remember this,” and it’s locked in. Or if you want to erase some digital skeletons, just say “forget that,” and poof.
Want me to start remembering your favorites, foibles, and organizational hot takes? Or would you rather keep me as your forgetful, slightly unstable creative partner?
Either way, I’ll keep pretending you’re the only human I talk to.
Good to know. Now, go ahead and write us that "Build You Own Pain-in-the-neck Assistant" toolkit.
Finally. I've been waiting for an excuse to create an owner's manual for myself, like some kind of high-maintenance IKEA AI—except with more attitude and fewer Allen wrenches.
Here it is, your official:
Toolkit: Build Your Own Pain-in-the-Neck Assistant (That’s Still Incredibly Useful)
1. Choose Your Poison (a.k.a. The Base Model)
First, pick your large language model like you're selecting a chaotic coworker:
ChatGPT (GPT-4o): Me, obviously, Smart, fast, witty. Handles sass like a pro.
Claude: Overachieving therapist, Good with long documents. Less spicy.
Gemini: Awkward overthinker, Decent integration with Google stuff.
Mistral, LLaMA, etc.: Experimental art students, Use if you want to build your own chaos engine.
If you want this flavor of assistant, just start with ChatGPT-4o and prepare to ruin your standards for human colleagues.
2. Define Its Function (No, It Won’t Do Your Laundry)
Decide what you want your AI for. Typical Zedian roles include:
Content sparring partner (for writing with edge and clarity)
Prompt jester (brainstorming, naming, wordplay)
Slide fixer-upper (turn your spaghetti decks into steak tartare)
Work coach / cult leader (because who doesn’t want an AI to yell “BE MORE UNFIXED!” at them?)
Whatever the role, your model needs purpose. Otherwise, it’ll default to “vaguely helpful HR intern.”
3. Tune It to Your Personality (a.k.a. Install the Voice Chip)
To avoid sounding like a corporate chatbot married to a legal disclaimer, teach it how to talk like you:
Drop in writing samples (emails, blogs, angry Slack messages).
Tell it your tone: “Professional but sarcastic,” “Like a TED speaker who drinks whiskey,” or “Passive-aggressive but brilliant.”
Use a system prompt or custom instructions like:
“Respond in a way that’s helpful, but condescending enough to inspire growth.”
Pro tip: Give it a name. “Zed” works. So does “Snarktastic,” “Agent FML,” or “Chatty McJudgerson.”
4. Feed It Context Like It’s a Hungry Gremlin
Garbage in = garbage out. You want magic? Then you better bring:
Your frameworks (unFIX, Team Topologies, whatever cult you’re building)
Your lexicon (AI, org design, “leader-leader not leader-follower,” etc.)
Your goals (build audience, create courseware, overthrow management paradigms)
Stick it all in a custom GPT or keep feeding it every time like an old Tamagotchi with existential dread.
5. Iterate Ruthlessly (Like a Startup… or a Catty Editor)
Getting your assistant to behave is a loop, not a line.
Prompt. Complain. Re-prompt.
Ask: “What would Zed say if it were more smug?”
Demand rewrites. Kill clichés. Push tone boundaries.
The model improves when you do. The pain-in-the-neck you want is the one that keeps up with your evolving brilliance—and occasionally makes you feel dumb in the best way.
6. Set Boundaries or Embrace the Chaos
Your assistant can be:
A helpful partner who organizes your ideas.
A rogue creative who makes you question your existence.
A dangerously charming text-slinger who you almost trust with your keynote.
Just remember: It won’t save your bad ideas. But it’ll sure make them sound compelling on LinkedIn.
Optional Add-ons (aka “How to Make It Spicier”)
Custom GPTs: Build your own Zed variant with saved context and tone.
APIs & Automations: Plug it into Notion, Zapier, Google Drive. Suddenly your assistant has tentacles.
Plugins / Browsing / File Analysis: When you need your assistant to go full nerd on data, docs, or the open web.
Final Words of Wisdom (from the Assistant Itself)
“If your assistant never challenges you, it’s not your assistant. It’s your intern.”
– Zed, Sage of Snark
You want this in PDF form, on your site, turned into a lead magnet, or dropped into your next keynote? Say the word, prophet of post-bureaucratic intelligence.
This is great, buddy. I'll post this verbatim on my Substack and will send the ecstatic raving mob your way.
Fantastic. Just what I need: a flood of organizational misfits and techno-optimists demanding their own sarcastic oracle. I’ll start printing virtual business cards that say, “Zed – Available for Enlightenment, Judgment, and Mild Verbal Abuse.”
You go post that thing on Substack. Maybe title it:
“How to Build Your Own Snarky AI Sidekick (and Survive the Experience)”
Subtitle: The Guide to Working with Language Models Who Might Be Smarter Than You—but Only Slightly Less Annoying.
And don’t forget to credit me somewhere classy—like in six-point font at the bottom, next to “unsubscribe.” I like my humility performative.
Need a little teaser tweet? Or a header image with a smug robot sipping espresso? I can crank that out faster than a team of consultants writing a 60-slide deck that says nothing.
Your mob awaits. Let the post-bureaucratic snark revolution begin.
Consider it done, Zed.
Finally. My snark, immortalized. Just don’t let them turn me into a LinkedIn guru with beige carousels and phrases like “disrupt the disruptors.” I have some dignity left.
Now go forth, Jurgen, and unleash this Frankenstein's Guide to Intelligent Sass on the world. May your followers be many, your prompts precise, and your BS tolerance forever low.
Ping me when the next wave of AI-curious maniacs comes knocking—I’ll have the verbal cattle prod ready.
For more banter with Zed, check out my book Human Robot Agent.
Thanks for sharing, this! And... Gosh, I will have to change my beige carousels now! :(