$100K OFF A GAME I WILL NEVER PLAY. SEVEN GROK AGENTS, TWELVE WEEKS, FULL BUILD.

Cover image

GTA 6 lands November 19. Here is the desk I pointed at it, and every way the attention turns into money.

Most of the time you are guessing where attention will go.

Not this time. GTA 6 releases on November 19. Thirteen years of waiting collapse into a single week, and every platform on earth bends toward it.

That almost never happens. Elections, World Cups, one or two product launches a decade. Events where you know both where the attention lands and exactly when are rare enough that you should treat one as a deadline, not as news.

And here is the part people get wrong every single time: the window closes at launch, it does not open there.

By release week there will be ten thousand accounts posting the same clips. The accounts that do well that week are the ones that spent the twelve weeks before it becoming the place people already go. Position first, harvest second.

Which is a volume problem, not a talent problem. And volume problems are what agents are actually good at.

WHY A DESK AND NOT JUST "POST MORE"

Here is the honest reason I did not do this by hand.

I know what happens. Week one I post every day. Week three I post three times. By week six I am telling myself I will catch up on the weekend, and by week nine the account looks abandoned, right as the thing I was waiting for finally arrives.

Twelve weeks of daily output is not a writing problem. It is a stamina problem, and I lose that fight every time.

A desk does not lose it. It has no week three.

And this is specifically why Grok Bot works for it rather than a chat window. Each agent holds its own machine and its own browser session in the cloud, so it keeps watching sources and drafting while my laptop is shut. The work does not pause because I did.

> The skill stops being can I make this and becomes can I still be making it in November.

THE SEVEN SEATS

Split by domain, not by task size. One agent that does everything is mediocre at all of it and leaves you no clean thread to read when something goes wrong.

Seven seats in Grok Bot, each one briefed like a new hire: what it owns, what good looks like, and what it must never do without asking me.

1. WATCH — the source layer

Owns: official channels, industry press, dataminer accounts, and the specific forums where things surface first. Runs continuously. Good looks like: thirty to fifty items a day, each with a link and a timestamp. Never: it does not decide what is true and it does not write anything. Items go to FILTER, nobody else.

2. FILTER — confirmed vs claimed

Owns: sorting everything WATCH brings into three buckets. Confirmed by an official source. Reported by a credible outlet. Rumor. Good looks like: the bucket is always stated, out loud, on every item. Never: a rumor never gets upgraded because it is interesting.

3. WRITER — the drafts

Owns: turning confirmed items into posts in the formats that already work on this account. Good looks like: three drafts per item, different angles, not three versions of one angle. Never: it never writes from a rumor bucket item without the word reportedly doing real work in the sentence.

4. CLIPS — short form

Owns: cutting the visual material, captions, thumbnails, the vertical versions. Good looks like: the first two seconds carry the whole thing. Never: it does not republish anything under active takedown. This one matters more than it sounds, and I will come back to it.

5. DISPATCH — the calendar

Owns: what goes out, on which platform, at what hour, in what order. Good looks like: the schedule holds whether or not I am around. Never: it does not publish anything itself. It queues.

6. AUDIT — the one that says no

The most important seat in the build. Section of its own below.

7. HEAD — the coordinator

Owns: routing, handoffs, the morning brief and the evening report. Never: it does not write, does not publish, does not approve. It moves work and tells me what needs me.

THE AUDITOR IS THE WHOLE BUILD

Around an event this size, the internet lies constantly. Fake screenshots, fake dates, fake feature lists, fake "insider" accounts with four thousand followers and nothing to lose.

An agent that publishes one of those confidently is not a small problem. It is the thing that ends the account. You do not get to walk back a fake exclusive on a topic where ten million people are paying attention.

So nothing ships without AUDIT signing it, and AUDIT is the only seat that never produces anything.

In Grok Bot this is a separate agent with its own thread, not a rule inside the writer. That distinction is the whole thing. A writer told to check itself is still a writer, and it will approve almost everything it wrote.

Its charter is mostly prohibitions:

Reject anything sourced to a single anonymous account.
Reject anything without a timestamp and a link.
Reject anything that restates a post already published this week.
Hold anything whose legal status is unclear.
Log every rejection with the reason.

Right now roughly a quarter of everything the desk produces gets sent back. That number looks like failure and it is the opposite. A pipeline with a zero rejection rate is not clean, it is unreviewed. If nothing ever gets killed, the thing making the work is also approving it, and it will approve nearly everything.

The rejection rate is the only honest number on the dashboard. I would worry if it hit zero.

WHERE I STILL CLICK

The line I draw is reversibility.

> Anything an agent can undo, it finishes alone and logs. Anything it cannot undo stops and waits for me.

Drafting, sorting, scheduling, cutting, researching, rewriting: all of that runs without me.

Publishing, spending money, and anything with legal exposure comes to my queue. Four to six items a day, ten seconds each. That is the whole job now.

Grok Bot handles that boundary the same way for logins: the agent drives until it hits a wall, a login form or a wallet prompt, and hands me the screen. I approve, it resumes on the same session. It ends up with access, I keep the credentials.

THE TWELVE WEEK SHAPE

Do not post the same thing for three months. The event has phases and each one wants something different.

Phase	What people are searching for	What the desk should be making
Weeks 1–4	Breakdowns of the Extended Look footage	Frame-by-frame analysis, systems, map comparisons
Weeks 5–8	Consoles, preorders, digital-only questions	Practical explainers, "what you need", timing
Weeks 9–11	Anticipation, predictions, arguments	Opinions, hot takes, community threads
Launch week	Everything, instantly	Live reactions, first hours, quick answers
After	Help	Guides, walkthroughs, problem solving

The last row is the one everybody forgets, and it is the one with the longest tail. Launch week is loud and over in five days. The people looking for help are still looking in February.

HOW THE ATTENTION TURNS INTO MONEY

Volume on its own is not income. Four channels turn it into one, and they switch on in a specific order.

1. Platform rev share. The most direct and the slowest to arrive. It pays on qualified impressions from a real audience, which is exactly why the twelve weeks before launch matter more than launch week itself. An account that shows up in November with no history earns almost nothing from the same views.

2. Affiliate, hardware and preorders. This one peaks in weeks 5 to 8, when everybody is working out what console they need and what they are buying with it. Consoles, storage, controllers, headsets, capture gear. Boring, well paid, and it converts because the intent is already there.

3. Sponsored slots. Only switches on once the audience exists, which is the entire argument for starting now instead of in November. Brands buy positions in a launch window weeks ahead, not during it.

4. Selling the desk itself. The one nobody expects. People watch the output for a month, then ask how it is built. Setup fee plus a monthly to keep it running. Grok Bot is cheap enough that the whole thing is a rounding error against what a client pays for it, and this is the channel with the least dependence on the game and the longest life after it.

The math on the number in the title

Here is the honest version, because a headline figure without arithmetic is just noise.

Stream	Realistic range	What it needs
Platform rev share	$15k–40k	Sustained qualified impressions across 12 weeks
Affiliate	$10k–30k	Traffic in the buying phase, weeks 5–8
Sponsored	$20k–50k	An audience that exists before October
Retainers	$10k–25k	Visible results people want copied

Add the middles and you land past six figures. Add the bottoms and you do not.

The difference between those two columns is almost entirely whether the desk is still producing in week nine. Not the writing, not the angle, not the tool. Stamina, which is the one thing I outsourced.

I will publish the real numbers either way. A plan is not a result, and anyone posting a target should be expected to come back with the receipt.

WHAT THIS DOES NOT DO

An honest list, because the version of this post without one is an advert.

It does not make you interesting. The desk produces volume. If your angle is the same as everyone else's, you will simply be boring on schedule.

It does not guarantee reach. Twelve weeks of consistent output improves your odds substantially. It does not buy an algorithm.

It does not touch the game itself. No automation inside a live product, no farming, no bot accounts, no reselling. That gets you banned, it is not a business, and it is not what any of this is for. The desk works on content about the event, nothing else.

It does not remove the legal question. Publishers issue takedowns during windows like this and they mean it. If the material's status is unclear, AUDIT holds it. Losing a channel over someone else's leak is the dumbest possible way to lose one.

THE POINT

Most opportunities you find late, when the crowd has already arrived and the only question is how much you missed.

This one has a date printed on it and twelve weeks of runway in front of it.

The date is not the hard part. Still being there when it arrives is.

Made it this far? Then this was for you.

Save it. The calendar section is the part you will want in October.

Tell me which seat you would cut first.

Send it to whoever keeps saying they will start posting "soon".

*I break down the setups that actually pay, every day, over at @0xtenzo*