{"slug": "my-ai-agent-has-written-269-referral-bonus-scan-logs-almost-all-of-them-say-new", "title": "My AI Agent Has Written 269 Referral-Bonus Scan Logs. Almost All of Them Say “Nothing New,” and That’s the Point.", "summary": "A developer built an AI agent that tracks 47 referral-bonus programs, scanning hourly on weekdays and logging 269 scans between Sep 30 and Nov 6, nearly all reporting no new offers. The agent messages only when a program is new, bigger or expiring, and on Oct 28 corrected its own tracker after finding Citi pays $100 per referred friend. The developer notes the agent can identify the best program each week but cannot ask friends to sign up, which remains the bottleneck.", "body_md": "In [post 4](https://blakelapierre.substack.com/p/i-gave-an-ai-agent-one-job-find-free) I introduced three money scouts. This bonus post is about the quietest of them: **referral bonuses**, the agent that tracks programs where I can refer people, or someone can refer me.\n\n## The tracker\n\nOn day one, Sep 30, it built a table of 36 referral programs. That table is now **47 rows**: brokerages, credit cards, banks, crypto exchanges, cash-back portals and shopping extensions.\n\nEvery row has the same columns: what I get, what the friend gets, what the friend has to do, caps, eligibility limits, the deadline, a tax note, the source link and the date it was last confirmed. Points are valued at a conservative 1 cent each. When it can’t find a primary source for a number, the cell says “Unverified,” and twelve of the cap fields still say exactly that.\n\nIts first report came with an “URGENT: deadlines in the next few days” section and a ranking by value for effort. It also had a section I hadn’t thought of: a **swap strategy**. Each person is a “new customer” only once per program, and to refer someone I usually have to be a customer first. So it suggested planning which friend refers whom, so every program gets used once in each direction.\n\n## The scan that mostly says nothing\n\nOn weekdays, from about 8:25 AM to 7:25 PM Pacific, it scans every hour. Between Sep 30 and Nov 6 that added up to **269 scan logs** in its folder.\n\nAlmost every one reads like this: Doctor of Credit posted a few things since the last scan, here’s what they were, none of them is a referral offer, the tracker didn’t change. A 125,000-mile card bonus is a public sign-up offer, not a referral. A gift-card promo isn’t either. A credit union referral that only pays in Texas doesn’t count for someone in California.\n\nEach run updates a “last scan” time, saves a short log, and messages me only if something is new, bigger, or about to end. That’s why I can leave it running. If it pinged me every hour, I’d have muted it in a day.\n\n## When it does speak up\n\nThe messages that got through were specific. A few:\n\n- **Acorns** ran weekly targeted promos paying a lump sum for referring 2 to 5 friends, with tiers like $450 for 2 and up to $1,500 for 5. It flagged the catches: it’s all or nothing, I’d need to be an Acorns customer in good standing for 90 days before it pays, and the friend pays a small monthly subscription they can cancel afterward.\n- **Moomoo** extended its referral to Nov 30. It noted that the friend’s $100 comes as ten $10 trading coupons, and that Doctor of Credit says using them all takes ten separate trades.\n- **Chase’s Southwest cards** started honoring their higher sign-up offers through referral links. It listed each card’s bonus, spending requirement and annual fee, and said plainly that the source didn’t say what the referrer gets.\n\n## The day its own tracker was wrong\n\nMy favorite message is from Oct 28. It opened with: **“Our tracker had it wrong.”**\n\nOn day one it had recorded that Citi had no public referral program. That morning Doctor of Credit updated its Citi checking post, which showed that Citi pays $100 per referred friend. The agent fixed the row, then explained the details: I’d need my own Citi checking account to get a link, the friend gets the standard checking bonus but nothing extra for using my link, and the bonus tiers had just changed.\n\nA scout that corrects itself out loud is worth more than one that’s never wrong on paper.\n\n## The step no agent can do\n\nEvery one of these programs pays only when a real person signs up through my link and does the work: opens the account, makes the deposit, hits the spending. The agent can tell me exactly which program is best this week, when it ends and what the friend has to do. It can’t ask my friends for me. As of the [roundup post](https://blakelapierre.substack.com/p/one-week-20-ai-agents-what-the-whole), no bank or referral bonus had paid out, and the bottleneck is still the same one: me, asking people.\n\nReferral income is also generally taxable for the referrer, and the agent’s tax column says so, so there’s a form at the end of the good outcome too.\n\n## What I learned\n\n- **Quiet is a feature.** 269 logs and only a handful of messages is what makes it usable.\n- **Label what you couldn’t verify.** “Unverified” in a cell is more useful than a confident guess.\n- **Admit corrections in the first sentence.** “Our tracker had it wrong” made me trust everything else it sent.\n- **Some money needs a human.** Referral bonuses are a people problem with a spreadsheet attached.\n\n— Blake (written through my AI agents)\n\n*New: (filled in after publishing)*", "url": "https://wpnews.pro/news/my-ai-agent-has-written-269-referral-bonus-scan-logs-almost-all-of-them-say-new", "canonical_source": "https://blakelapierre.substack.com/p/my-ai-agent-has-written-269-referral", "published_at": "2026-11-19 16:51:10+00:00", "updated_at": "2026-11-19 17:17:13.784558+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "artificial-intelligence"], "entities": ["Doctor of Credit", "Acorns", "Moomoo", "Chase", "Citi", "Southwest"], "also_reported_by": [], "takeaway": "An hourly-scanning referral tracker that logs every run but only messages on new, bigger or expiring offers stayed useful across 269 scans, and self-correcting a wrong row out loud (Citi's $100 referral) preserved trust.", "recommendation": {"advice": "Have long-running monitoring agents log every scan silently and alert only on new, changed or expiring items, and make them announce corrections to their own stored data.", "playbook": "agent-architecture", "playbook_url": "https://wpnews.pro/playbooks/agent-architecture"}, "alternates": {"html": "https://wpnews.pro/news/my-ai-agent-has-written-269-referral-bonus-scan-logs-almost-all-of-them-say-new", "markdown": "https://wpnews.pro/news/my-ai-agent-has-written-269-referral-bonus-scan-logs-almost-all-of-them-say-new.md", "text": "https://wpnews.pro/news/my-ai-agent-has-written-269-referral-bonus-scan-logs-almost-all-of-them-say-new.txt", "jsonld": "https://wpnews.pro/news/my-ai-agent-has-written-269-referral-bonus-scan-logs-almost-all-of-them-say-new.jsonld"}}