Hostinger
CLOSED - $14k/video - video already recorded, awaiting Gabriele's review
Self-hosting the Hermes Agent on a Hostinger VPS: an AI agent that never sleeps, on a server that costs less than a coffee.
HOST-A — generated
Brand purple correct, real VPS dashboard, agent breaking the screen plane, silver plaque on the shelf. Ready to use.
Recommended title
I Run a Self-Improving AI Agent 24/7 on a $5 VPS (Hostinger + Hermes)
In-image text $5/mo
Pattern dollar anchor + 2 named tools
$ figure n=13, 1.43x (1.86x excl. AntiGravity); 2-tool naming n=38, 1.28x
The price IS the story. Specific and falsifiable, which is what the dollar-anchor pattern rewards.
Alternate titles & in-image text — thumbnails not yet generated
| Pick | Title | In-image text | Pattern & evidence |
|---|---|---|---|
| HOST-B | Hermes + Hostinger = An AI Agent That Never Sleeps Trades the price hook for the emotional promise. Right-hand side is concrete-ish but softer than an object. |
NEVER SLEEPS | equation A + B = Outcome n=18, 1.43x - most robust structure on the channel, survives every cut |
| HOST-C | Replace Your $200/mo AI Subscriptions With One $5 VPS Strongest pattern available, and it fires correctly here because the audience genuinely pays for the thing being replaced. |
$200 → $5 | cost collapse n=7, 3.19x - highest ratio in the corpus, but thin sample |
| HOST-D | I Deployed a 24/7 AI Agent in 11 Minutes ($5/mo) Effort anchor instead of value anchor. Only use if the real deploy time is close to the claim. |
11 MINS | time-box + dollar anchor time-box n=8, 1.80x; both together n=3, median 106,610 |
| HOST-E | Claude Code + a $5 VPS = Infinite Uptime Names the specific limit (uptime) rather than a vague 'superpowers', which is the difference between 1.38x and 0.68x. |
INFINITE UPTIME | equation + ceiling-removal ceiling-removal cluster n=17, median 75,909, 1.38x |
Supabase
IN DISCUSSION - contact unverified, do NOT send to kate@plug.dev
Receipts cold-open (30 videos, never paid), then a full-stack build where an Agent Skill does the setup Jack normally does by hand.
SUPA-A — generated
Brand green correct. The on-screen 'RLS: Configured / Tokens: Connected' badges are exactly the manual steps the Agent Skill concept automates. Note the stacked-slab glyph is a generic database mark, not Supabase's real lightning-bolt logo.
Recommended title
Claude Code + Supabase Agent Skills = Zero-Config Full-Stack Apps (NEW Skill)
In-image text 30 VIDEOS. $0 PAID.
Pattern equation + (NEW Skill) tag
equation n=18, 1.43x; (NEW ...) tag n=8, 1.89x - caveat: 7/8 cluster in the AntiGravity window
Thumbnail text carries the credibility claim, title carries the mechanic - H17 complementarity done properly.
Alternate titles & in-image text — thumbnails not yet generated
| Pick | Title | In-image text | Pattern & evidence |
|---|---|---|---|
| SUPA-B | Supabase Agent Skills Killed the Setup Step Names the exact pain (setup) rather than a generic benefit. Short, punchy, no number to defend. |
ZERO CONFIG | ceiling-removal, named limit ceiling-removal n=17, 1.38x |
| SUPA-C | I Built a $12,000 SaaS Backend in 9 Minutes (Supabase Skills) Dollar figure sits at the top of the evidence-backed $8k-$12k band. Highest-ceiling option here. |
$12,000 | dollar + time-box both together n=3, median 106,610; $8k-$12k band n=9, 1.93x |
| SUPA-D | Claude Code + Supabase Agent Skills = No More RLS Errors The sharpest option for the actual developer audience - RLS is a real, specific, felt pain. Narrower reach, higher intent. |
NO MORE RLS | equation + named specific limit equation n=18, 1.43x |
| SUPA-E | Every Level of Supabase Agent Skills Explained Cheap third test of a promising but unproven format. Flag the thin evidence if pitching this one. |
LEVEL 1 → 5 | explainer n=2, median 103,117, 1.87x - INSUFFICIENT SAMPLE, both instances cleared 95K |
Base44
CLOSED - mid-contract, video #2 - per-video rate unconfirmed
A real client brief built end-to-end in Base44 against a live clock, with the model-picker as an on-camera decision.
B44-A — generated
Strong composition and the model-picker UI reads clearly — but the model names it invented on screen (Sonnet 3.7, GPT-4o, Gemini 2.5, Llama 4) are outdated and not Base44's real model list. Fix before publishing, or the sponsor will catch it.
Recommended title
Base44 Built a $9,000 App in 24 Minutes (I Picked the Model)
In-image text $9,000
Pattern dollar + time-box + parenthetical hook
$ n=13, 1.43x; time-box n=8, 1.80x; $8k-$12k band 1.93x
Satisfies 'Base44 front and centre' literally - the brand name leads the title.
Alternate titles & in-image text — thumbnails not yet generated
| Pick | Title | In-image text | Pattern & evidence |
|---|---|---|---|
| B44-B | Base44 + Claude = A Real Client App in One Sitting Pairs the sponsor with the tool Jack actually uses - honest, and rides the established-tool search intent. |
ONE SITTING | equation A + B = Outcome n=18, 1.43x |
| B44-C | I Stopped Paying a Developer $9,000 (Base44) Highest-ratio pattern. Fires correctly - the audience genuinely pays devs. Verify the $9,000 claim is defensible. |
$9,000 SAVED | cost collapse + dollar anchor cost collapse n=7, 3.19x |
| B44-D | I Tested Base44 Against My Own Stack The honest option: directly addresses that Jack has said on camera he defaults to Claude Code. Weakest by pattern data, strongest by authenticity. Needs a fair, non-rigged framing or it reads as a paid hit piece. |
vs CLAUDE CODE | head-to-head comparison rivalry framing n=11, 0.85x - UNDERPERFORMS on this channel |
| B44-E | Base44 Lets You Pick the Model — Here's Which One Wins Built on the one genuinely brand-specific feature found in the emails. Confirm with Tal which models are actually selectable before committing. |
WHICH MODEL? | curiosity gap on a real product mechanic curiosity gap n=10, 1.04x (flat) - carried by the mechanic, not the pattern |
Genspark
CLOSED - $15k via Eezycollab - script approval required BEFORE shooting
Hermes produces an answer; hard pivot; Genspark Design ships a finished brand kit and landing page. Fable 5 firewalled out of the Genspark segment.
Brand rule: Brand prefers the title stay generic with the tool revealed mid-video. Options are flagged for compliance.
GEN-A — generated
Weakest of the five — it rendered the big number well but dropped the actual concept: there's no 'flat answer vs finished product' contrast, no brand kit and no landing page. Recommend a re-generation with a revised prompt before this one goes to Eezycollab.
Recommended title
I Built a $10,000 Brand Kit With One Prompt
In-image text $10K
Pattern dollar anchor + how-I frame
$8k-$12k band n=9, median 106,610, 1.93x
Names no tool - zero attribution risk under the brand's naming rule.
Alternate titles & in-image text — thumbnails not yet generated
| Pick | Title | In-image text | Pattern & evidence |
|---|---|---|---|
| GEN-B | $10,000 of Design Work for One Prompt Highest-ratio pattern, still tool-agnostic. Sharpest cost framing of the set. |
ONE PROMPT | cost collapse + dollar anchor cost collapse n=7, 3.19x brand-rule safe |
| GEN-C | A Full Brand Identity in 8 Minutes Effort anchor. Cleanest option if the real build time is genuinely short. |
8 MINUTES | time-box n=8, 1.80x brand-rule safe |
| GEN-D | Most AI Gives You Answers. This One Ships. Lowest script-rejection risk of any option - it is literally the brand's own positioning line. |
ANSWERS vs SHIPPED | curiosity gap + brand's own differentiator verbatim curiosity gap n=10, 1.04x (flat) - value is in using the brand's own approved framing brand-rule safe |
| GEN-E | Hermes + One Prompt = A Finished Brand Strongest structural pattern, but names Hermes in the title - check with Eezycollab whether that breaches the naming rule before using. |
FINISHED | equation A + B = Outcome n=18, 1.43x - most robust structure names a tool — check brand rule |
Gamma
NEGOTIATING - $5k internal figure - Fiona is waiting on a reply
Sales call ends, transcript walks through Fireflies into Gamma's API, out comes a branded client-ready proposal deck.
GAM-A — generated
Verified Ultramarine correct, deck fan breaking out of the meeting-transcript screen, Gamma wordmark bottom-left. The client logo on the deck is a placeholder brand.
Recommended title
I Built a $10,000 Client Proposal in 6 Minutes (Fireflies + Gamma)
In-image text $10,000
Pattern dollar + time-box + 2 named tools
both together n=3, median 106,610; $8k-$12k band 1.93x; 2-tool n=38, 1.28x
Stacks three validated cues from different families - the corpus rule that most reliably beats piling cues from one family.
Alternate titles & in-image text — thumbnails not yet generated
| Pick | Title | In-image text | Pattern & evidence |
|---|---|---|---|
| GAM-B | Fireflies + Gamma = Client Proposals While You Sleep Most robust structure. Right-hand side is a concrete deliverable, which is what separates the 646K version from the 11.9K version. |
WHILE YOU SLEEP | equation A + B = Outcome n=18, 1.43x |
| GAM-C | Stop Paying a Designer $2,000 Per Pitch Deck Highest-ratio pattern and it fires correctly - agencies really do pay for deck design. Note $2,000 sits below the $8k-$12k sweet spot; it is defensible per-deck but a weaker anchor. |
$2,000/DECK | cost collapse n=7, 3.19x |
| GAM-D | 10 Meetings → 10 Client Proposals in 6 Minutes Directly dramatises the batch capability Jack already demoed on camera. Thin pattern evidence but the strongest product truth. |
10 DECKS | compression / batch framing compression sub-form n=2, median 114,420, 2.08x - INSUFFICIENT SAMPLE |
| GAM-E | The Proposal System That Closes AI Clients Narrowest reach, highest intent. Slots into an existing content pillar rather than competing with the build-video format. |
CLOSES CLIENTS | outcome framing, plugs the client-acquisition pillar his 4 client-acquisition videos run 2.4k-5.3k views - a smaller but highly-qualified audience |