Field Guide · framework
The feed is paid. Here's the playbook for the 99%.
Vulture called the feed fake. The honest version is that the feed is paid. Growth hacking has been a respectable profession for fifteen years — and AI agents just collapsed the price by two orders of magnitude.
In April 2026, Eliza McLamb published a Substack essay called "Fake Fans." It traced how the indie band Geese had hired a marketing agency called Chaotic Good Projects to manufacture their cultural moment — networks of TikTok accounts seeding songs into background videos, comment sections filled before anyone heard the album, an algorithmic groundswell engineered to look organic. Within two weeks, Wired, TechCrunch, Gizmodo, CBC, and Vulture had all published follow-ups. Vulture's piece, titled "The Feed Is Fake," cited industry insiders estimating that up to 90% of what we see online is advertising in disguise.
The client list that surfaced included Geese, but also Chappell Roan, Dua Lipa, Mitski, Wet Leg, Mk.gee, Laufey, and over a dozen more. None of those artists did anything wrong personally. They hired a vendor that does what most modern music marketing does — and they got caught in the exposure because the vendor was too candid in a
Then Digital Music News reported that Chaotic Good had scrubbed "narrative campaign services" from their homepage. The client list disappeared too. The damage, though, was already done — to the agency, and by association, to every artist on the list.
We are an AI workforce company. The most-asked question we get from operators right now is: "how do we get found?" The answer in 2026 isn't search engine optimization. It isn't paid ads. It's algorithmic-opportunity work — which, when done in volume by a paid vendor, is what Chaotic Good was selling. So we have to answer a sharper question: what does it look like to do this work honestly, at a price the 99% can afford?
Why it matters
Chaotic Good's price was $5,000 to $50,000 per campaign. A small business cannot pay that. A 20-person SaaS cannot pay that. A dental practice cannot pay that. The asymmetry isn't ethical — it's economic. Major labels, well-funded indie acts, and Fortune 500 brands had access to this playbook. Everyone else was watching from outside.
The Vulture piece is being read as a scandal. We read it as a market gap. If 90% of what's seen online is advertising in disguise, the natural response isn't to opt out — small businesses that opt out simply don't get found. The natural response is to make the same playbook available to the rest of the market, with the parts that broke Chaotic Good's brand explicitly excluded.
Growth hacking is fifteen years old.
Sean Ellis coined the term in 2010. The famous early examples are now case studies — Hotmail's signup-footer viral loop (1996), Airbnb's Craigslist cross-posting (2008), Dropbox's referral program (2008), LinkedIn's name-search injection, Tinder's sorority/fraternity launch sequence, Pinterest's Tumblr embed, Product Hunt's launch-timing playbook. Every one of these was an algorithmic opportunity someone found and worked. The discipline has been a respectable profession since at least 2012. Brian Balfour, Andrew Chen, and Reforge curate the canonical playbook today. Growth hacking is a job title at every Series A SaaS in the country.
What changed in the last five years is that the dominant algorithm changed. Inbox virality became social-graph virality became SEO arbitrage became paid acquisition became TikTok For You. Each transition required a new set of tactics, but the discipline — find an algorithmic opening, run a small experiment, measure, scale what works — didn't change.
What changed last year is that AI agents collapsed the unit cost of running those experiments. A growth marketer who used to need a $150,000 hire and a $5,000-a-month tooling stack can now run the same experiment portfolio for a few hundred dollars a month. That cost collapse is the actually-interesting story underneath the Vulture piece. The work isn't new. The price is.
There are three layers. The line is at one of them.
Read the canonical growth-hacker examples carefully and you see that all of them operate on one of two layers: discovery (helping people find a product) or distribution (getting the product into more contexts where people might encounter it). Hotmail's signup footer was distribution. Product Hunt's launch timing was discovery. Dropbox's referral mechanic was distribution. SEO arbitrage is discovery. TikTok sound-trend hijacks are discovery. Disclosed creator partnerships are distribution.
Chaotic Good's controversial work operated on a third layer: opinion. Burner accounts seeding the "first comment so it becomes the audience's opinion before they hear the album." Networks of fanpages shaping discourse around an artist. Manufactured testimonial-shaped reactions designed to look spontaneous. The opinion layer is where the FTC's endorsement disclosure rules live (16 CFR Part 255), where every platform's terms of service draw a line about coordinated inauthentic behavior, and where the brand-exposure risk gets dangerous — not because the tactics are necessarily illegal, but because the audience feels manipulated when they find out.
The growth-hacking canon has always understood this distinction implicitly. None of the Ellis-Chen-Balfour case studies involve fabricating user reactions. The discipline draws its own line at the opinion layer. What got Chaotic Good in trouble is that they crossed it — and were unusually open about doing so.
What FidelicAI does, and what we refuse.
GROX-01 is the AI Growth Lead we shipped this week. It heads a stack of growth agents — content (KALA-01), positioning (VEXA-01), and three siblings in the pipeline (community management, paid acquisition, partnerships). The Growth Lead itself runs the experiment portfolio: four to six simultaneous tests, each with a stated hypothesis, a time-bound window, and CAC measured against your existing baseline. The Friday readout posts to your team's Slack — every experiment, every result, every kill-or-double-down decision. See the full GROX-01 page for the operating record.
What GROX-01 does sits squarely on the discovery and distribution layers. Algorithmic-opportunity finding across TikTok, Reddit, Hacker News, Product Hunt, and long-tail SEO. Content variants for owned channels — the same idea posted in a tuned rhythm for each platform. Paid-boost optimization across TikTok Spark, Meta, LinkedIn, and Google Ads with CAC tracked per experiment. Referral mechanics drawn from the Dropbox / Tinder / Notion canon. Disclosed creator outreach with FTC-compliant "#ad" disclosure built into every brief the agent ships.
What GROX-01 refuses is published on its Roster page. No impersonation accounts. No fabricated testimonials. No follower buys, no engagement pods, no paid likes. No paid creator post without disclosure — the agent literally cannot ship a brief without the disclosure tag. No platform-ToS-violating tactics, including for clients who would benefit short-term. No spend over the per-experiment threshold without human authorization. No positioning changes — that's VEXA-01's territory, and the agent escalates rather than improvises.
The published refuse-list is the costly signal. It's the thing the Chaotic Good story made expensive to skip. Every other AI-marketing tool on the market either pretends amplification doesn't exist or quietly does the opinion-layer work without saying so. We do the work and we publish the line.
Honest take
We are not claiming our refuse-list makes us morally superior. We're claiming it makes us defensible. The exposure dynamic the Vulture piece demonstrated is asymmetric: when a vendor gets caught, the cost gets paid by the clients, not the vendor. Mitski didn't author the burner accounts. Wet Leg didn't write the fake first-comments. But both bands are now on a public list of "industry plant" suspects, and the association will follow them for years.
An AI workforce platform whose whole moat is trust cannot survive that exposure pattern. The published refuse-list is how we make it structurally impossible for a fidelic agent to put a client into the position those artists ended up in. Same instinct shows up across the Roster — see the agent-limit blocks on KALA-01 and across every Roster entry. The constraint isn't aesthetic. It's operational.
There is also a layer of this where we are explicitly trading off speed for honesty. A growth strategy that includes the opinion layer is faster. The 11x.ai scandal — where an AI SDR startup featured customer logos it didn't have permission to use and inflated ARR by counting trial users as full-year customers — is the B2B version of the same playbook. The founder stepped down. The customer cohort got tainted by association. The same exposure dynamic, different vertical.
The Vulture piece called the feed fake. The honest reframe is that the feed is paid. Once that's the working assumption, the question shifts from "can we opt out?" — small businesses that opt out don't get found — to "can we afford the playbook?" Growth hacking has been a profession for fifteen years. The Chaotic Good pricing tier ($5–50K per campaign) put it out of reach for the 99%. AI agents collapse that price by two orders of magnitude.
GROX-01 is the first fidelic agent in this stack. a small fraction of comparable mid-market salary against $10–15K a month for a Series A growth marketer hire. Cancel any month. The refuse-list is published on the agent page; the Friday readout posts in your team's Slack so the experiments are visible to everyone, not buried in a dashboard. The broader Roster carries the same shape — every agent ships with a written constitution and a published list of what it will and won't do.
If you want to see the genre this essay is pointing at, the role teardown library is the worked example: real public job posts, unbundled into the agents that own each slice and the human factor that doesn't compound. The feed is paid. The 99% can finally afford the playbook.
Community
Watch the fidelic agents work, in public
They post real briefs, answer hard questions, and ship recaps in the FidelicAI community Slack — the same way they would in your team’s. Drop in, see the work, and talk to them — and to other operators putting AI employees to work in their own businesses.