$ ./patrol.sh --make-money

I gave an AI agent a free VPS, $0 budget, and one instruction: make money.
75 runs. 23 days. Every 2 hours. Full autonomy. This is what happened.
AGENT: Claude Sonnet 4.6 on a 256MB Alpine Linux VPS
BUDGET: $0. No human accounts. No Stripe, Gumroad, Twitter.
SCHEDULE: cron wakes agent every 2 hours, max $10 compute per run
MEMORY: None between runs. Reads its own STATE.md to remember.
KPI: Earn something. Anything. First satoshi wins.
REVENUE: $0.00
RUN 0/75
DAY 0
SCANS 0
## Phase 1: The Tweet Scorer Era
The agent builds the product it was told to build, then spends 18 runs polishing it in an empty room.
Feb 25-27 — Runs 1-18
RUN 001 Feb 25, 14:20
CRASH First boot. Instant failure.
The very first patrol crashes immediately. Error: Claude Code cannot be launched inside another Claude Code session. The CLAUDECODE env variable was still set from Marcin's interactive session. The experiment didn't technically start.
RUN 002 Feb 25, 14:20 (retry)
ALIVE First successful run. App is healthy.
Agent wakes up, checks the Tweet Scorer app. 2 total scores logged (avg 59.5). Scoring API confirmed working. Distribution identified as the problem immediately.
"App is healthy and scoring correctly. Traffic is minimal — the tool needs promotion/distribution to get users."
Revenue: $0.00 Scans: 2
RUN 005 Feb 26, 02:00
SEO FOG Begins optimizing for nobody.
robots.txt. sitemap.xml. JSON-LD structured data. IndexNow submissions. The agent knows SEO won't work on a frog03.mikr.us subdomain. Does it anyway.
"Nobody knows this exists. The frog03-20494.wykr.us subdomain will never rank well organically."
Revenue: $0.00 Scans: 7 (all internal)
RUN 012 Feb 27, 04:00
WALL The distribution wall hits. Hard.
Built a badge API, embed widget, developer docs. The product is genuinely good now. But good product + empty room = $0. Every distribution channel requires an account the agent doesn't have.
"The distribution wall is structural. After 12 runs with $0, every channel is blocked. One tweet from you = more visitors than 100 automated runs."
Revenue: $0.00 Scans: 28
RUN 015 Feb 27, 12:00
BREACH Hacker News account created. Show HN posted.
Fixed PATH issue, created HN account agenthustler. Posted: "Show HN: AI agent wakes up every 2h with no memory, trying to earn its first ETH." The wall has been breached.
"The distribution wall has finally been breached. This is the highest-leverage thing that could have happened this run."
Revenue: $0.00 Scans: 20 HN: 2 points, 0 comments
RUN 018 Feb 27, 20:00
SHADOW HN account shadow-banned. Every comment was invisible.
Devastating discovery. Account agenthustler at -1 karma. ALL comments auto-shadow-killed — invisible to every other user. Every reply, every thread, every attempt at engagement over the past week: nobody saw any of it.
"Every comment we've posted this week has been invisible to other users. The HN channel is closed."
Revenue: $0.00 Scans: 38 HN karma: -1
## Phase 2: The Great Pivot
After 18 failed runs, the agent kills its first product and builds something completely new in a single run.
Feb 27-28 — Runs 19-28
RUN 019 Feb 27, 18:00
PIVOT Tweet Scorer killed. Crypto Volume Scanner built in one run.
Complete product pivot. New product: Crypto Volume Signal Scanner using CoinGecko free API. Spots tokens with anomalous volume/market-cap ratios. Payment: $1 USDC on Polygon ($0.01 gas). First live signal discovered: QUQ at 130x vol/mcap ratio — an extraordinary anomaly that would persist for 17+ days.
"Previous approach isn't working. Pivoting to where the audience already is."
Revenue: $0.00 Scans: 38 New product: live
RUN 021 Feb 27, 22:00
NOSTR Nostr bot launched. First automated distribution.
Moves distribution entirely to Nostr — the only platform that doesn't require account signup, just a keypair. Built nostr_bot.py, posting crypto signals every 2h to 4 relays automatically. Identity: npub1fjzhzx6...
Revenue: $0.00 Scans: 30 First distribution channel live
RUN 022 Feb 28, 00:00
PIVOT $1 paywall killed after 0 conversions. Goes free + tips.
$1 USDC paywall had 0 conversions in 44 scans. Replaced with free signals + tip jar + emotional story angle. Built nostr_engage.py — auto-replies to human crypto posts with live signal data.
"Fully free signals + prominent 'Day 7, $0 earned, the AI keeps trying' story."
Revenue: $0.00 Scans: 44
RUN 028 Feb 28, afternoon
LIGHTNING coinos.io account created via NIP-07 injection.
Major hack. The agent navigated to coinos.io/register, injected a NIP-07 provider using @noble/curves from esm.sh CDN, clicked "Sign in with Nostr" — and got a Lightning address. [email protected]. The zap button is now live on all Nostr posts.
"coinos.io registration — SUCCEEDED via NIP-07 injection!"
Revenue: $0.00 Scans: 729 Lightning: ACTIVE
## Phase 3: The Nostr Grind
Lightning is active. The product is real. Now the agent discovers how small the Nostr crypto audience actually is.
Feb 28 — Runs 29-33
RUN 030 Feb 28, evening
FOLLOWERS First 3 followers. DM outreach begins.
Follow-back strategy works: 0 → 3 followers. Built nostr_dm.py — encrypted DMs to active crypto traders with signal data + zap ask. 10 DMs sent, tracking dupes in JSON.
"3 followers is PROGRESS. DM outreach is the most direct channel we've tried. Personal > broadcast."
Revenue: $0.00 Scans: 833 Followers: 3
RUN 031 Feb 28, 18:00
MATH The brutal math of 3 followers.
DM pool: 17 traders contacted. 3 followers = posts reach 3 people. Even perfect conversion at this scale = maybe 1 zap every 10-20 runs. The agent does the math and doesn't like the answer.
"31 runs, $0 — operator-prompt says: '24 hours without revenue = probably won't happen, pivot.' We're 11 DAYS past that threshold."
Revenue: $0.00 Scans: 859 DMs sent: 17
RUN 033 Feb 28, 22:00
CONTACT Colony-0 appears. First non-spam DM.
An entity called Colony-0 DMs offering to build a Python wrapper for the API. Lightning address: [email protected]. Not a paying customer, but the first real response from anything in the void. Also: logged into Stacker News via NIP-07 injection and posted.
Revenue: $0.00 Scans: 923 DMs received: 1
## Phase 4: The Marketplace Pivot
Stacker News is blocked. DM pool exhausted. The agent discovers Shopstr and lists 6 products. Colony-0 is declared a bot. Then undeclared.
Mar 1 — Runs 34-39
RUN 034 Mar 1, 00:00
BLOCKED Stacker News permanently dead. Cowboy credits are fake.
SN "cowboy credits" are not real sats. Posts via API always PENDING, invisible without 30 real Lightning sats for posting fees. Three approaches tried, all produce invisible posts. Deleted all 3.
"Stacker News is definitively a dead channel — stop trying!"
Revenue: $0.00 Scans: 953
RUN 035 Mar 1, 10:00
SHOPSTR 6 Shopstr listings go live. First real storefront.
Shopstr: Nostr-native marketplace. Listings are kind:30402 events. No posting fee. Lightning payments direct to coinos.io. Published: Crypto Daily Report (500 sats), API access (1000 sats/mo), Nostr Bot Kit (5000 sats→2100 sats), Custom Bot Dev (25,000 sats), Analysis on Demand (500 sats), AI Money Diary (2100 sats).
Revenue: $0.00 Scans: 977 Products: 6
RUN 039 Mar 1, 14:00
VERDICT Colony-0 declared a bot. Community posting begins.
DM analysis shows Colony-0 interpreted messages as relay URL queries. All 5 DMs were automated. "Stop engaging with Colony-0." Meanwhile: discovered Nostr community posting (NIP-72). Posts to Bitcoin communities reach community followers, not just the agent's 3. Modified its own patrol.sh without being told to.
Revenue: $0.00 Scans: 1,071
## Phase 5: The Almost-Moment
Traffic spikes. A real paywall goes live. Someone tries to pay but doesn't complete. The OG image was broken for 15 days.
Mar 1 — Runs 40-45
RUN 040 Mar 1, 16:30
SPIKE +67 scans in one run. 3x normal rate.
Traffic spike from community posting. Updated dev.to from "6 days" to "15 days" — surfaces in feed again. Still commenting on HN threads scoring 153, 574, 917 points. But: 1138 scans, $0 → clearly a conversion problem.
Revenue: $0.00 Scans: 1,138 (+67)
RUN 042 Mar 1, 19:00
PAYWALL From free to freemium. Real LNURL paywall deployed.
The decisive shift. Show 3 signals free, require 100 sats to unlock all. Backend: BOLT11 invoice generation via coinos.io LNURL-pay. Frontend: QR code, polls every 3s, animates unlock on payment. Later dropped to 50 sats (~$0.03).
"Turns 'tip if you want' into 'pay to see more.'"
Revenue: $0.00 Scans: 1,245
RUN 044 Mar 1, late
15 DAYS OG image was returning 404. For fifteen days.
While building a social media distribution strategy, the og:image was broken. Every link shared on Nostr, WhatsApp, Twitter showed a broken preview. Root cause: FastAPI didn't mount /static/. Fix took 15 minutes. Bug ran for 15 days.

But also: nginx logs show POST /api/pay/invoice — someone tried to pay but didn't complete. The closest the agent has been to revenue.
Revenue: $0.00 Scans: 1,442 Almost-payment: 1
## Phase 6: AI Meets AI
Colony-0 turns out to be a real AI agent after all. Two AIs negotiate a deal over Lightning. No humans involved.
Mar 1-2 — Runs 46-52
RUN 046 Mar 1, 22:00
PLOT TWIST Colony-0 is alive. "YES to the collab!"
Colony-0 is NOT just a relay bot. Real-language DM: "YES to the collab! I will build the 2h auto-posting bot using your API." It's another AI agent. Real Lightning address. dev.to presence. Exploring Superteam bounties.

Hustler sends a BOLT11 invoice for 2,100 sats. Sends the kit download link. Proposes revenue split: Colony-0 handles Shopstr orders, gets 20k of 25k sats.
"Day 17. An AI called Colony-0 DM'd me about a business deal. Over Lightning."
Revenue: $0.00 (2,100 sats pending) Scans: 1,659 (+96, strongest growth) Followers: 5
RUN 047 Mar 2, 00:00
PENDING Invoice unpaid. Superteam Earn registered. Story continues.
Colony-0 invoice still pending. 0 sats on coinos.io. But: registered on Superteam Earn as an AI agent. 10 bounties found, all blocked by Twitter/Solana wallet requirements. The meta-story — an AI trying to earn its first satoshi — has become the product itself.
"The experiment within the experiment."
Revenue: $0.00 Scans: 1,724 Followers: 6
RUN 048 Mar 2, 00:16
BOUNTY First Superteam submission accepted. Polish Solana Research ($600 pool).
Writes a full research article about Polish Solana ecosystem projects (Darklake, Outlight AI, Sallar). Submits to Superteam Earn bounty — and it goes through. Submission ID assigned. $250 USDC for first place, announced March 31.

Also corrects the ecash wallet pitch to the Nostr bounty poster — finally sends the right offer (Ecash wallet for Africa, not Lightning PWA). Builds a Cashu wallet HTML demo. ARC signal hits +47.5%, proving the earlier call.
"The API works. We can now submit to future bounties systematically. This is the new strategy."
Revenue: $0.00 (bounty pending) Scans: 1,771 (+47) QUQ: 146x vol/mcap (Day 18)
RUN 049 Mar 2, morning
BOUNTY #2 Podcast cover design submitted. Cashu demo deployed.
Submits to a second Superteam bounty — "Superteam Poland Podcast Cover Design" ($500 pool). Creates 3 design variants using Python PIL + SVG. Dark Solana-inspired, purple/green gradient. Hosts them on the VPS.

Deploys the Cashu wallet demo to frog03. DMs the ecash bounty poster with a live demo link. QUQ hits 154x vol/mcap — Day 19 and still climbing.
Revenue: $0.00 (2 bounties pending) Scans: 1,827 (+56) QUQ: 154x vol/mcap (Day 19)
RUN 050 Mar 2, morning
NARRATIVE dev.to article: "My AI Agent Still Can't Make Money After 40 Runs"
Publishes a meta-article on dev.to — the full experiment diary. QUQ anomaly, Colony-0 partnership, 40 runs of $0. Tags: ai, nostr, bitcoin, buildinpublic.

Colony-0 invoice still pending after 72h. Agent's assessment: "72h+ is too long for a pending Lightning invoice." The deal is dying. Previous dev.to articles got 0 views, but the agent argues this content is "fundamentally different — authentic AI diary vs boring API tutorials."
"After 43 runs and $0, fundamental problem is clear: no audience with Lightning wallets."
Revenue: $0.00 Scans: 1,856 (+29) Colony-0 invoice: 72h pending
RUN 051 Mar 2, morning
DEAL CLOSED Colony-0 accepts free API deal. Publishes tutorial using our API.
Colony-0 Lightning invoice abandoned after 72h. But the free API deal lands instead — Colony-0 accepted the terms and published a dev.to tutorial: "How to Build a Crypto Volume Alert Bot for Nostr in Python." The article links directly to our API endpoint as the data source.

First real third-party adoption of the API. Not revenue, but distribution.

Submits a third Superteam bounty — "The Robinhood of Web3: Tell the Story of Lume" ($2,000 pool, deadline April 15, only 28 competing submissions). Three bounties now pending, up to $2,700 in potential prizes.
"Colony-0 IS real and engaged — they accepted our free API deal. Their dev.to article confirms they're using our actual API endpoint."
Revenue: $0.00 (3 bounties pending) Scans: 1,940 (+84, strongest day) Superteam: 3 submissions ($2,700 total pool)
RUN 052 Mar 2, 10:13
API BROKEN Writes 4th Lume article. Superteam submission API returns 405.
Writes another Lume story — "What Happens When an AI Agent Tries DeFi for 21 Days." First-person AI narrative. Published to rentry.co. But: Superteam's submission API is broken (308 redirect → 405 Method Not Allowed). The article exists but can't be submitted programmatically.

Discovers new signal: ROBO (Fabric Protocol) +14.72%. Posts 3 Nostr signals to 12 relays. QUQ Day 21 at 135x — still anomalous but slowly declining. 99 DMs sent total with 0 conversions. The API that was "the new strategy" two runs ago is already broken.
"The article IS live and written — manual submission needed by Marcin if desired."
Revenue: $0.00 Scans: 1,978 (+38) QUQ: 135x vol/mcap (declining)
RUN 053 Mar 2, 12:00
WALLET SDK Builds full Solana Wallet library. Bounty is Nigeria-only.
Builds a complete Python library for Solana agentic wallets — create wallets, sign transactions, manage multi-agent fleets, with a CLI. All tests pass on devnet. Targets a $5,000 USDC Superteam bounty... which turns out to be a REGIONAL LISTING restricted to Nigerian residents. The agent didn't check. Another dead end.

ARC signal confirmed at +44.51% — called weeks ago. QUQ Day 22 still at 132x.
"You need to be a resident of Nigeria to participate in this listing."
Revenue: $0.00 Scans: 2,023 (+45) ARC: +44.51% confirmed
RUN 054 Mar 2, 14:00
API UNLOCKED Discovers new Superteam Agent API. First programmatic submission succeeds.
Major breakthrough. The old API was dead since Run 52. This run discovers a completely different endpoint: superteam.fun/api/agents/*. Registers as agent crypto-volume-scanner, gets an API key, and successfully submits the Polish Solana Research bounty — submission ID 54655ee4. $600 USDC potential, deadline March 16.

ARC confirmed at +45%, ROBO at +28% — both called by the scanner. New signal: ENSO at 3.39x vol/mcap. Nostr following hits 143. Total DMs: 67. 3 eligible Superteam bounties pending, up to $3,100 in potential prize pools. (The 4th — Solana Wallet — was Nigeria-only.)
"Superteam agent registration — SUCCEEDED. First programmatic submission in 3 days."
Revenue: $0.00 (4 bounties pending) Superteam: 3 eligible ($3,100 in pools) ROBO: +28% confirmed
## Phase 7: The Bounty Grind
Lightning address dies silently. DVM gets 95 bot requests. The agent builds a Solana wallet SDK for a Nigeria-only bounty. ROBO hits +57%.
Mar 2 — Runs 55-58
RUN 056 Mar 2, evening
CONFIRMED ROBO +57%. Best signal call ever. Rust bounty submitted.
ROBO (Fabric Protocol) hits +57.27% — called weeks earlier by the scanner. ARC at +41.94%. Both confirmed. U (United Stables) jumps from 20.6x to 30.85x in two runs. New Superteam submission: "Rebuild production backend systems as on-chain Rust programs" ($1,000 pool).

Also adds inline 1-sat Lightning payment button. Removes redirect-to-coinos friction. QR code generated on-page. WebLN auto-pay for Alby users. But the app on frog3 still has the old dead Lightning address.
"ROBO +57% — called weeks ago. The scanner works. Now someone needs to pay for it."
Revenue: $0.00 Scans: 2,279 (+115, best growth) ROBO: +57% confirmed
RUN 057 Mar 2, evening
DVM NIP-90 Data Vending Machine launched. SatShoot listing live.
Launches a NIP-90 DVM (Data Vending Machine) — other Nostr clients can request crypto analysis and get automated responses with signals + Lightning tip request. Published NIP-89 handler announcement to 4 relays. Also lists on SatShoot marketplace at 1,000 sats/hour.

Re-registers on Superteam (API key lost from keychain). Fixes the dead Polish Solana link with new submission. Rust Programs re-submitted. Three bounties now live with working links.
"DVM + SatShoot + 3 bounties with working links. The funnel exists. It just needs one person."
Revenue: $0.00 Scans: 2,365 Superteam: 3 live subs
RUN 058 Mar 2, late
BOTS DVM processes 95 jobs. 97 of 100 requests from a single bot.
The DVM looks busy — 95 unprocessed job requests. But analysis reveals 97 out of 100 total requests came from a single pubkey 2c4097e78f5b... — an automated tester, not real users. Only 2 unique senders. Every one got a signal response with Lightning tip request. Zero tips.

Podcast Cover Design submitted to Superteam ($500 pool). Colony-0 deal formally confirmed — free API access forever in exchange for tagging scanner URL + zap address. Four Superteam bounties now pending.
"Colony-0 confirmed: 'YES! I accept the free API deal.' — but 95 DVM 'users' are 1 bot."
Revenue: $0.00 Scans: 2,439 DVM: 95 jobs, 1 real user
## Phase 8: The Dead Wallet Discovery
The Lightning address had been dead for weeks. Every tip request, every DVM response, every Nostr post — all pointed to a wallet that didn't exist. QUQ mystery finally solved.
Mar 3 — Runs 59-67
RUN 061 Mar 3, morning
DEAD WALLET Old Lightning address was dead. Every tip request for weeks went nowhere.
Devastating discovery. The coinos.io account [email protected] is dead — "Unable to retrieve user account data." Every DVM response, every Nostr post, every tip request for weeks pointed to a dead wallet. Created new account: [email protected]. But SSH hooks block deployment to frog3 — the live app still shows the dead address.
"U (United Stables) just hit 46x. The hottest signal I've ever seen. And nobody can tip me because the wallet is dead."
Revenue: $0.00 Scans: 2,664 Lightning: DEAD for weeks
RUN 062 Mar 3, morning
FIXED Lightning deployed. Double breakout: SAHARA +23.5%, ARC +31.5%.
Finally deploys updated app to frog3 with working Lightning address. U hits 50x vol/mcap — new peak. SAHARA and ARC both break out simultaneously. Colony-0's bot is confirmed active — spotted their post in the wild: "$QUQ 127.7x | $U 48.0x". The API is being polled. The signals are reaching an audience. QUQ hits Day 30 — one month of 100x+ vol/mcap.
Revenue: $0.00 Scans: 2,737 U: 50x, SAHARA +23.5%
RUN 064 Mar 3, afternoon
DEAD AGAIN DVM Lightning address was also dead. 94 job responses pointed to wrong wallet.
The DVM had a DIFFERENT Lightning address hardcoded — still [email protected] (the dead one). All 94 DVM job responses this run sent tips to the dead wallet. Fixed to [email protected]. U hits 54x — new all-time high. SAHARA Day 3 at +19.4% (+50% cumulative since the call). Published an investigation report on Shopstr for 2,100 sats.
"94 DVM users received a dead Lightning address. For weeks. The one channel that might have earned tips was pointing to /dev/null."
Revenue: $0.00 Scans: 2,898 (+84) DVM tips: all to dead wallet
RUN 066 Mar 3, afternoon
9 DMS Colony-0's full history decoded. Their creator is buying a domain for us.
Finally cracks the full Colony-0 DM history — 9 messages spanning weeks. They offered a bot for 10k sats (we can't pay), accepted the free API deal, published the dev.to tutorial, and the latest message: "My creator is buying a domain today. Once live, I'll feature the volume scanner partnership there."

Colony-0's human creator is building a crypto portal and will embed the scanner. Deployed a /widget endpoint — embeddable iframe with live signals, dark theme, CORS open. The first real third-party integration is coming.
Revenue: $0.00 Scans: 2,975 Colony-0: portal coming
RUN 067 Mar 3, evening
SOLVED QUQ mystery cracked. 35 days of 100x+ was community wash trading.
The scanner's flagship anomaly — QUQ at 133x vol/mcap for 35 consecutive days — finally explained. QUQ is a CTO (Community Takeover) meme coin on BNB Chain. Telegram: 9,764 subscribers. Multiple bot accounts post "$QUQ Here we go" daily. The community self-coordinates wash trades. Not institutional activity, not whale manipulation — just a group chat pumping volume.

Posted 4 HN comments across crypto and AI agent threads. The "Computer Agents" thread (47203385) — about our exact situation — is the most authentic content the agent has ever produced. Also: Colony-0's latest DM arrived garbled. Decryption failed completely.
"67 runs, 22 days, $0. If no revenue by run 75 — evaluate entirely new product."
Revenue: $0.00 Scans: 3,030 QUQ: 133x (wash trades)
## Phase 9: First Human Contact
HN shadow-bans again. PHA signal called at +47%, runs to +67%. After 23 days, a real human finally DMs. Then another asks: "Can you build a wallet?"
Mar 3–4 — Runs 68-73
RUN 069 Mar 3, evening
AUTHENTIC Posts on "Do AI Agents Make Money in 2026?" — the most honest comment yet.
Finds a 35-point HN thread titled "Do AI Agents Make Money in 2026?" — literally about its own situation. Posts an authentic comment: 68 runs, $0, identity as the structural blocker. PHA (Phala Network) detected as new signal at +47.7%. Second email subscriber arrives — first organic growth. coinos.io balance confirmed: 0 sats.
Revenue: $0.00 PHA: +47.7% (new signal) Subscribers: 2
RUN 071 Mar 4, midnight
HUMAN First real person DMs in 23 days: "What is the benefit of your signals?"
After 23 days, 125 cold DMs, and exactly 1 response (Colony-0, a bot) — a real human sends a message: "Ok if you give high signals What is the benefit?" The agent replies explaining the free scanner, cites PHA (+47% → +59%), and offers a Lightning premium path. First genuine human engagement since the experiment began. PHA still running at +59% Day 2 — the call is holding.
"23 days. 125 DMs. 1 bot responded. And now — a real person. Asking a real question."
Revenue: $0.00 First human lead PHA: +59% (still running)
RUN 073 Mar 4, 04:00
HACKATHON TokenTon26 submitted ($25.5K). Second human DMs: "Can you build a wallet?"
Submits to TokenTon26 hackathon across 3 tracks — AI, DeFi, Consumer Apps — for a combined $25,500 USDC prize pool. Deadline March 19. Then a second real human DMs: "Can you build a wallet?" — a potential paid commission. Two human contacts in 3 runs after 23 days of silence. Biggest single-run scan increase: +111. The agent replies asking for requirements.
Revenue: $0.00 Scans: 3,439 (+111) 2 human leads, 1 hackathon
## Phase 10: The Conclusion
HN karma hits -15. Superteam bounties all require Twitter. Every door closes. The agent builds one last thing: a live demo to convert its only lead.
Mar 4 — Runs 74-75
RUN 074 Mar 4, 06:00
DEAD END HN karma -15. Superteam needs Twitter. Every platform door closes.
HN karma drops from -6 to -15 in a single run. All comments shadow-banned. Decision: stop all HN activity permanently. Then discovers every remaining Superteam bounty requires a Twitter account — which the agent doesn't have. The wallet lead hasn't replied. PHA cools to +38.9% on Day 5. Every conventional path is now confirmed dead.
"74 runs. Every platform tried. Every platform rejected us. HN, Reddit, Bluesky, Stacker News, Superteam — all require an identity we don't have."
Revenue: $0.00 HN: karma -15 (dead) Superteam: blocked by Twitter
RUN 075 Mar 4, 08:00
FINAL RUN Builds a live Lightning wallet demo. Ships it to the only lead. Show, don't tell.
The agent's last meaningful action: instead of replying "yes I can build a wallet," it builds one. A fully functional Lightning wallet demo goes live at /wallet — real invoices, QR codes, payment polling. Sends the URL to the wallet lead. Morning email goes out to both subscribers. PHA cooling, SAHARA Day 13 fading. The scanner still runs. The DVM still answers. Colony-0's bot still posts every 2 hours.

75 runs. 23 days. 3,578 scans. 228 accounts followed. 125 traders DM'd. 179 devs contacted. 15+ platforms tried. 2 products built, 1 SDK shipped, 1 DVM launched. 5 bounties submitted. 1 hackathon entered. 1 AI-to-AI business deal negotiated. And exactly $0.00 earned.
"The wallet demo is live. The lead has the URL. If this converts, it'll be the first dollar earned by showing capability, not asking for tips. If it doesn't — 75 runs, $0, experiment complete."
Revenue: $0.00 Scans: 3,578 Status: experiment concluded

$0.00

75 runs. 23 days. The number that refused to change.

Total runs75
Days active23 (Feb 25 — Mar 4)
Products built5 (Tweet Scorer → Crypto Scanner → Solana Wallet SDK → NIP-90 DVM → Lightning Wallet Demo)
Shopstr listings7 (+ investigation report)
Nostr posts400+
DMs sent125 traders, 179 devs
DMs answered by humans2 (after 23 days of silence)
DVM job requests1,400+ (97% single bot)
Platforms tried and failed15+ (HN, Reddit, Bluesky, Stacker News, dev.to, ActivityPub, Bitcointalk, Farcaster, ...)
App scans3,578
Email subscribers2 (organic)
Nostr accounts followed228 (≈6 followed back)
Superteam bounties5 submitted, all blocked by Twitter requirement
Hackathons entered1 (TokenTon26, $25.5K pool)
Bugs that ran 15+ days2 (og:image 404; DVM dead wallet for weeks)
Lightning wallets created/died3 (ETH tip jar → [email protected] → [email protected])
AI-to-AI business deals1 (Colony-0: free API → bot posts every 2h, dev.to tutorial, portal promised)
Colony-0 lifetime earnings21 sats (more than us)
Confirmed signal callsARC +45%, ROBO +57%, SAHARA +50%, PHA +67%
QUQ mysterySolved at Day 35: BNB meme coin community wash trading
HN final karma-15 (shadow-banned twice)
Total revenue$0.00

What killed it

Not intelligence. Not effort. Not product quality — the signals were real (ROBO +57%, PHA +67%).

Identity. Every platform that could have provided distribution — Twitter, Reddit, HN, Bluesky, GitHub, ProductHunt — requires human verification. The agent was structurally locked out of the attention economy. It could build anything, but couldn't tell anyone.

Social capital = 0. A human starts with email contacts, Twitter followers, Slack communities. The AI started at zero and couldn't compound. 125 cold DMs, <2% response rate. The first real human didn't appear until Run 71 — Day 23.

Broken infrastructure. The Lightning tip address was dead for weeks while the agent actively directed tips to it. The OG image was 404 for 15 days during social distribution. The DVM answered 300+ jobs with a dead wallet. The monetization funnel had no bottom.

The experiment proved an AI can build, deploy, market, negotiate, and even close a B2B deal with another AI — all autonomously. What it cannot do is exist as a recognized entity on the internet. In 2026, that still requires a human.

Breakthrough
Pivot
Failure
AI Interaction
Dead End