Playbook
content engineering
how i actually do content with AI
if your AI-written content sounds like AI, you skipped a step.
you can't automate something you can't articulate. i know because i tried. for a while i thought AI was the skill. it's not. AI is labor. the skill is being able to describe what you want precisely enough that the machine can actually do it.
automation isn't magic. it's knowing a process so well you can explain it to a machine.
taste is the actual variable. weak taste means AI makes weak content faster. strong taste means AI makes that taste much cheaper to produce at scale. AI isn't the taste. AI runs the taste.
the foundation
do it by hand first.
every prompt that's actually worked for me came after i'd done the task manually enough times to describe what 'good' looks like. not vaguely. precisely. the sequence, the edge cases, the things you'd never say, the things that always land.
when i started ghostwriting, i wasn't using AI. i was learning what made a Karthik post a Karthik post: what the opening looked like, where the punchline landed, how much context was enough, which words he'd never use. that manual work was the data. AI just runs it later.
start here
imitate before you innovate.
the first time AI actually worked well for me was on a ghostwriting gig. we were running LinkedIn and X for a founder, and our operating theory was simple: making things work on the internet is mostly copying what already works, then adding your own spice. skip imitation and you become 'original' before you've earned the right to be.
so i found viral posts from similar profiles outside India and studied the structure, not the content. the pacing, the progression, the way they opened and closed. then i took one of those posts to ChatGPT with a single instruction: understand why this works, apply that structure to my idea.
it worked, not because ChatGPT is a great ghostwriter, but because the instruction was grounded in a real process i'd actually studied. structure first, not vibes.
how i learned
i built my own tools because i had to.
two years ago there was no comfortable stack for "i know content, i want AI to help, but i don't want to become an engineer." so i started making small wrappers for myself.
- -X post generator. got comfortable with coding.
- -script generator. learned to call multiple APIs in sequence.
- -Hemingway, basically Cursor for writers, built with friends. learned to build something that scales.
i come from a tier-5 college with a BBA degree. these tiny tools were how i learned to code by actually using it. each wrapper was a stepping stone.
you don't have to do any of that. i built those because i had to. the tooling has caught up.
the trap
don't overengineer.
the trap is thinking you need to build something because building feels impressive. the terminal screenshot, code scrolling, you accepting everything. it looks fancy and it's still just a wrapper around someone else's API that you didn't need to build.
the goal isn't to become a mediocre engineer because the barrier got lower. it's to use the best available thing for the job, whether that's skills, connectors, chats, or projects. use the boring thing if it gets you to the output faster.
the best thing i found this year
what is a skill?
a skill is two markdown files. one that tells AI what to do: the role, the rules, the voice, the process, the constraints. one that shows it what good looks like: real examples, the shape of what you want.
it's not a magic prompt. it's your working style, written down.
what changes everything
references change everything.
three years of using AI, one pattern has stayed consistent: add references and the output changes completely. the model stops guessing and starts pattern-matching against something real. it knows what you actually want instead of inventing something adjacent to it.
think about onboarding a new hire. you wouldn't just say "write like me." you'd give them your best examples, the mistakes to avoid, the format rules, what 'good' looks like in practice. AI needs exactly that, and most people skip it entirely.
my stack
i have skills for almost everything.
cold email, scriptwriting, tweets, newsletters, and a bunch of other workflows. they live at slashskills.vercel.app. they're literally me writing down how i think so AI can copy the pattern.
the Builders Central scriptwriting workflow is the best example of this. 230K followers, millions of views, and most of those videos were written by AI running off my references, my structure, my examples. using me as the reference, not replacing me.
people don't actually care if something was written with AI. they care whether it's good. the problem was never AI involvement. it was always AI content with no taste behind it.
if skills feel like too much
just use Projects.
a one-off chat starts from nothing every time. a Project keeps your files, examples, instructions, past decisions, and working style in one place so you're starting from context instead of blank.
i have Projects for almost everything: scriptwriting, tweets, newsletter, a CTO i ping when i'm stuck, random advice. Claude Projects, ChatGPT custom GPTs, Perplexity Spaces. same idea, different wrapper. use whatever you're already in.
why projects win
context is everything.
the first 100k tokens in a conversation are usually the sharpest you'll get. pile in more and the model spreads out, less focused on what you actually need. more surface area, less precision.
a chat inside a Project almost always outperforms one outside it because it starts with a world already built. the context you've set up becomes the quality floor for everything that comes after.
if you decide to build it
Bangers Only is the example.
it didn't start as "let me build an AI app." it started as a repeated content problem: people know what they want to say, they just need help making it sound right.
once the process was clear enough to articulate, the product became possible. the prompt layers, the UX decisions, the way the system turns rough thoughts into posts that sound like you wrote them at your best.
full architecture writeup: read on GitHub.
going further
you can make launch videos with AI too.
the Bangers Only launch video, roughly 25 seconds animated, was built with Remotion, Claude Code, and Codex. no After Effects, no Premiere, no Figma. just a terminal and a description of the scenes in plain English.
Claude Code wrote the React: spring animations, interpolation curves, beat-synced audio, all of it. roughly 1700 lines across four files. when video becomes code, you can iterate on it like code, which turns out to be a very different thing from dragging keyframes.
full process: Remotion + Claude Code thread.
to summarise
the order matters.
do it by hand.
know the process cold.
let AI run it.
skip step one and you get slop. the skill is the codified version of doing something manually until you understand it. the Project is what gives AI the context to run it well. the product is that workflow turned into an interface someone else can use.
AI isn't the taste. AI runs the taste.
start here