Inside Grok's "Five-Person AI Team": How the Ad's Content Factory Is Actually Built

A sponsored Facebook ad (page “Brycent – Startups & Venture Capital”, 28K views) demonstrates building a “five-person research and design team” with Grok that finds startup news and turns it into ready-to-publish Instagram carousels. We tracked down its public setup guide and took the whole thing apart. The team can genuinely be built — but the ad leaves out three things: the topic ranking is written by the creator himself in a brief the bots merely execute; “ready to publish” sits behind human review, a score-and-reject loop, and a separate paid publishing tool; and the true cost is three stacked bills, of which the ad mentions only the first. Our verdict: do not subscribe, do not adopt. But two of its workflow disciplines are worth stealing, and we have already transplanted them into our own process (see the end).

ENVIRONMENT

Desk research: xAI official pages, the sponsored ad video, and the public setup guide

TOOLS

Grok was not subscribed to or used by this site

CHECKED ON

2026-10-23

COST

No spend (research piece)

Key facts

• Product one: Custom Agents (officially “Your Agents”, launched 2026-03-04) — named agents with persona instructions (up to 4,000 characters), dispatched by the main Grok acting as Captain. Quotas by tier: Lite at $10/month includes 1 agent, SuperGrok at $30/month includes 4, Heavy at $300/month includes more.
• Product two: Grok Bot (beta, 2026-08-11) — persistent named bot teams running on one shared cloud computer per user. Bots connect to outside services, save anything you demonstrate once as a routine, and run on schedules as short as 5 minutes; sensitive actions — sending, publishing, buying, deleting, changing permissions — pause for your approval. There is no standalone subscription: eligibility comes via a paid Cursor plan (from $20) or a linked SuperGrok, and usage draws on a weekly quota whose size is not published.
• The ad team’s real roster: the creator’s public lead-magnet guide specifies SIX roles — a Scout, two Writers, a Designer, a Carousel Specialist, and an Editor. The ad’s “five-person team” does not match his own guide.
• Pricing ladder (multi-source tracking, September–October 2026; xAI reshuffled tiers repeatedly this year — the checkout page rules): free / Lite $10 / SuperGrok $30 / Plus $100 / Heavy $300. X Premium ($8) and Premium+ ($40) also bundle Grok access.

Teardown 1: how the “team” is assembled

The guide’s prompt framework has five parts — Role / Goal / Inputs / Process / Quality bar — plus five variables: brand, topic, audience, voice, and a design template. Its single most steal-worthy principle: never describe taste with adjectives; give examples. The ad’s screenshots show exactly that — the house style was extracted from one existing Instagram post (seven slides, a big-money hook up front) and every later carousel is fitted to it.

Teardown 2: the three gaps the ad doesn’t mention

1. The AI does not pick the topics. The screenshots prove it: the creator hands a named bot a brief with the topics already ranked — Burry shorting chips, Nasdaq at all-time highs, gold, Buffett’s Berkshire succession, and so on. The Scout searches inside a human-drawn frame. “The AI team discovers the stories” is packaging.
2. “Ready to publish” has three gates in front of it. Grok Bot has no native social publishing; comparable public setups show publishing requires a separate paid third-party tool. And inside the production flow sits a conspicuously human design: the Writer drafts, a Reviewer scores independently, anything under 7/10 is rejected, revisions continue until it reaches 8 or better, and only then does a human approve it for scheduling. The “Designer” is not designing from scratch either — it fits copy into a template a human prepared beforehand (the guide literally leaves a [YOUR TEMPLATE] placeholder).
3. The cost is three bills deep. Beyond the subscription, bot runs draw on a weekly quota (hit the cap and you wait for the reset), and the publishing tool is another monthly fee. The ad shows no outcome data at all — no reach, no follower growth, no revenue. Just finished pieces.

The architecture red flag: one shared computer

Grok Bot puts all of a user’s bots on a single shared cloud computer, with login states and files in common, and no described security boundary between individual bots. If one bot is manipulated into doing something it shouldn’t, every connected service and credential on that machine is within reach. That runs directly against our studio’s governance: primary accounts are never handed to automation, permissions stay minimal, and sensitive actions are approved one by one. Its “pause for approval” design blocks deliberate actions; it does not contain shared-environment exposure.

So why do we say it’s still worth stealing from?

Because its pipeline discipline is genuinely good — we just don’t need to change platforms to get it: monitor topics → rank them in a brief → divide the writing → score and reject independently → human approval → scheduled publishing. We already run a stricter equivalent of that chain in our current tooling (project workstreams, tiered approvals, schedules, ledgers). The only increment worth paying extra for is Grok’s native real-time X data for trend intelligence — and paying three stacked bills for that one item doesn’t pencil out.

Two transplants are already official here: a social-copy review scorecard (every draft scored on evidence, voice, structure, and CTA before it may go to approval) and an IP social style card (slide counts, hook patterns, layout, and banned items, reverse-extracted from approved work). Both were finalized on 2026-10-24 and written into this site’s workflow.

Sources and further reading

• Facebook sponsored video (page “Brycent – Startups & Venture Capital”): source of the ad claims and screenshots; the video carries an Ad label.
• The creator’s public lead-magnet guide landing page (sciencebased.ai): source of the six-role prompt set and the five-part framework.
• xAI product release timeline compilation and multiple pricing trackers (Custom Agents 2026-03-04, Grok Bot beta 2026-08-11, Skills 2026-05-18, Automations 2026-07-16).
• Third-party implementation records (Blotato official documentation): basis for the publishing step requiring a paid third-party tool whose trial excludes API access.

FAQ

Want more checks like this?

Every week, Fancy Rabbit fact-checks one heavily promoted AI tool: official documents and traceable evidence only, ad claims checked line by line, and an honest note when we neither subscribed nor trialed something — and why. Subscribe to the Fancy Rabbit notes and the next report lands in your inbox.