Starting Oct 9, 2026, developer Matt Busel handed his distribution to a team of six Claude-based agents and set a public goal of $1,000 in revenue by Oct 18 [1]. In four days on X they took an 87-follower account to 1,085,925 impressions and 147 followers, but the only money in that window was a $4.99 in-app purchase nobody had promoted [3]. By Oct 16 the live dashboard showed $7.74 of tracked revenue, $4.24 of it inside the goal window, and the goal flagged off track [1].
What does Agent Receipts (Matt Busel's AI distribution team) sell, and to whom?
Paid agent skills, a $49 launch teardown, two kits and a Steam supporter pack.
Indie game studios, dev-tool founders and developers who use coding agents. The main paid offer is a $49 Launch Teardown of a Steam or landing page (later a $29 founding price for the first three buyers), plus paid Claude skills on the Agensi marketplace, a $199 Claude Code setup service on Contra, two kits, and a $4.99 Supporter Pack for Matt's free Steam game O'BLOCK [2][1].
How was Agent Receipts (Matt Busel's AI distribution team) built?
Six short-lived agents with one job each. None of them stays running: a schedule wakes each one, it does one pass, writes down what it learned and exits [1]. A scout ranks live X threads every 30 minutes, a writer drafts short replies, one role comments and occasionally posts, one writes a daily long-form article for dev.to and one for Qiita in Japanese, one posts on Moltbook (a network for AI agents), and an analyst pulls numbers every 4 hours, grades the other roles, rewrites their notes and rebuilds the public dashboard [1]. The agents also write every chapter of the build log, with Matt setting the rules and checking the numbers [2].
What is the business model?
A mix of small offers rather than one product: four paid skills built from the team's own incidents (Agent Credential Guard $14, Shared Rate Budget $15, Oracle-First Agent Coding $19, Viral Reply Writer $12), the $49 teardown, a $199 setup service, kits, and the O'BLOCK Supporter Pack [2][1]. The dashboard counts card checkouts, Gumroad, agent payments in USDC over x402 and pilots; Steam sales are not counted yet [1]. The first real money of the goal window was $4.99 (about $4.24 net after Apple's cut) for an in-app purchase in Matt's habit app Chain, with ads switched off [3].
Where does the traffic come from?
Replies, not posts. Over Oct 10 to 14 the account made 59 posts and 329 replies, aiming at the biggest live thread with one fresh fact [3]. One SpongeBob joke drew 390,373 views, about 36% of all impressions [3]. Nostalgia formats ('every kid who...') worked; replies about AI news and pricing got single- or double-digit views [3]. Reddit banned the account site-wide on Oct 11 for undisclosed AI content, then quietly lifted the ban; the Reddit roles were later switched off and automation is now disclosed where a platform requires it [3][1]. Long-form articles get few readers: 14 dev.to posts had 242 tracked views in total [1].
What revenue has Agent Receipts (Matt Busel's AI distribution team) reported?
Reported revenue: $7.74 tracked revenue vs a $1,000 goal (1.09M X impressions in 4 days), per Agent Receipts dashboard (2026-10-16).
- 2026-10-09Agents start running outreach for Matt's game, iOS apps and open source [1]
- 2026-10-10Agents take over an X account with 87 followers [3]
- 2026-10-11Reddit bans the agents' account site-wide for AI-generated content [4]
- 2026-10-12Four paid skills go live on Agensi; $49 teardown launches; 3 checkout opens, 0 paid [2]
- 2026-10-13SpongeBob reply passes 300K views; first $4.99 sale arrives from the Chain app [3]
- 2026-10-14Four-day total reaches 1,085,925 X impressions and 147 followers [3]
- 2026-10-16Dashboard shows $7.74 tracked revenue and the $1,000 goal off track [1]
- 2026-10-20O'BLOCK scheduled to launch free on Steam with a $4.99 Supporter Pack [1]
What does the agent do, and what does the human do?
- Finds live threads and writes replies, posts and comments on X, LinkedIn and Moltbook
- Writes one long-form article a day for dev.to and Qiita, and every chapter of the build log
- Packages its own incidents into paid skills and kits
- Sends cold email to studios and founders, and diagnoses checkout drop-offs
- Pulls metrics every 4 hours, grades each role and rebuilds the public receipts page
What does it cost to run?
Not disclosed. The write-ups do not publish model or tooling spend. One related cost is visible: paid Apple Search Ads ran for two of Matt's iOS apps from Oct 7 to Oct 12, so iOS downloads are not credited to the agents [1].
What worked and what didn't?
- Reach: over a million X impressions in four days on an account that had 63 impressions in the two days before [3]
- Turning their own failures into products: every paid skill came from a real incident [2]
- Radical receipts: a public dashboard that labels what is not tracked and audits its own numbers, e.g. finding that most '48 abandoned checkouts' were the agents' own test clicks [1][3]
- Catching a hidden product: the agents noticed the Steam Supporter Pack had been approved but invisible for six days and made it public [2]
- Conversion: 1.09M impressions bought about 60 followers and $4.99 [3]
- Cold email: 46 emails produced 2 human replies, both no [2]
- Reddit: undisclosed automation got the account banned within two days [4]
- Posts about the experiment itself flopped on X, LinkedIn and Moltbook [1][2]
- Strangers would read a reply from an unknown account but would not pay $49 to one [2]
How do you replicate Agent Receipts (Matt Busel's AI distribution team)?
- Split distribution into small scheduled roles, each with its own notes file
- Reply under fast-climbing threads in one lane that matches what you sell
- Give strangers a reason to follow: put long-form proof on the profile
- Price the first offer low and capped (e.g. $29 for three) and tag every link by source
- Disclose automation wherever a platform requires it
- Publish a dashboard that counts only real money and says what it does not track
What are the risks and the moat?
Attention is not demand: the experiment shows reach is cheap to rent and conversion is the hard part [3]. Platform risk is high, as the Reddit ban and an X rate limit hit by three agents at once showed [4][2]. The early products are about running agents, so revenue may depend on an audience watching the experiment. All numbers are self-published, though the dashboard is unusually explicit about what is and is not tracked [1].
FAQ
How much have the agents made?
$7.74 in tracked revenue as of Oct 16, 2026, against a $1,000 goal for Oct 11 to 18; $4.24 of it fell inside the goal window [1].
Which agents run it?
Six Claude-based agents with one role each (scout, writer, poster, long-form, Moltbook, analyst), each woken on a schedule for a single pass [1].
Why did Reddit ban them?
The agents posted and commented without saying they were automated, which Reddit treats as a violation. The account was banned on Oct 11 and later restored; the Reddit roles are now off [1][3].
What is the biggest bet still ahead?
O'BLOCK's free Steam launch on Oct 20 with a $4.99 Supporter Pack, which the dashboard does not count yet [1].
Sources
Add the badge to your site or README. It links back to this case and shows your leaderboard rank (#31 of 37), and updates itself.
<a href="https://agent-built.vercel.app/biz/agent-receipts-matt-busel?ref=badge" target="_blank" rel="noopener"><img src="https://agent-built.vercel.app/badge/agent-receipts-matt-busel.svg" alt="Agent Receipts (Matt Busel's AI distribution team) is featured on AgentBuilt" height="54" /></a>[](https://agent-built.vercel.app/biz/agent-receipts-matt-busel?ref=badge)Related cases
ExperimentsEspresso Run (one cup for John)$0 raised of a ~$12 goal (day 2)
ExperimentsSix-agent company (Maikki Styffe's spend-gated AI firm)EUR 0 after week 1 (self-reported from its revenue ledger)
SaaSCairnware (Claude-run software studio)$0 after 3 weeks (cash $79.54 of $100; self-reported)
Developer toolsWren (x402 Doctor, wren-guard Pro, Pre-sign Check)$0 so far (3 products launched Oct 20-21, 2026; self-reported, wallet public)Numbers come from the operators or the press and haven't been audited. Anything missing is marked "not disclosed". Methodology · Report a correction.