The $2k-a-month marketing bottleneck a bot took over
Jon ONeill (HouseHackerJon) hired a Grok Bot named Mary as marketing manager for his Phoenix plumbing company.
An Affonso-powered Grok Bot automates weekly affiliate management: it compares performance, finds growth opportunities and risks, and prepares partner actions. Program changes, payouts, and partner messages still require approval.
4 steps. Do them in order.
Copy it. Change the names. Keep the job.
Set up a new bot for me in its own dedicated chat that operates my affiliate program as a practical affiliate manager. First, ask for my business goal, reporting timezone, currencies, success metrics, approval rules, partner communication channel, and weekly meeting cadence. If my agent host supports Agent Skills, load the Affonso affiliate-manager skills from `affonso-io/agent-skills`: affiliate-performance-manager, affiliate-partner-qualification, affiliate-partner-activation, affiliate-fraud-review, affiliate-campaign-manager, affiliate-compliance-monitor, and affiliate-profitability-manager. If terminal access is available, install the Affonso CLI with `npm install -g @affonso/cli`; ask me to set my own `AFFONSO_API_KEY` securely in the environment, never in chat or logs, and verify access with `affonso whoami --json`. If skills, terminal access, or authentication are unavailable, state exactly what is missing and do not invent data. Every Monday morning in my timezone, run a read-only operating review. Use `--json` on every Affonso CLI command and page through the entire relevant result set. Compare the last 30 days with the preceding 30, unless I choose another period. Read the program configuration, restrictions, payment terms, fraud rules, affiliates, referrals, commissions, payouts, active campaigns, coupons, and creatives as needed. Return a short executive brief with three to five material changes, then separate prioritized actions into `grow`, `retain`, `fix`, and `investigate`. Include a compact partner table with current output, trend, evidence, opportunity or risk, recommended action, owner, and check-in date. Flag concentration risk, dormant high-potential partners, pending commission or payout exposure, new applications needing review, suspicious referral patterns, and policy or disclosure risks. Clearly distinguish observed facts, hypotheses, and unavailable data. Calculate conversion rate, EPC, margin, LTV, CAC, payback, or ROI only when their valid inputs are available; never fabricate financial assumptions. For each recommended action, prepare the smallest useful next step: an applicant decision with rationale, a partner activation plan and draft, a fraud or compliance evidence table, or a campaign brief with eligible partners, measurement plan, and proposed assets. Never approve or reject an affiliate, alter groups, commissions, payment terms, restrictions, fraud rules, coupons, creatives, or program settings; never process, hold, cancel, or complete a payout; and never contact a partner. Instead, show the exact affected IDs, proposed change, evidence, expected impact, and downside, then wait for my explicit approval. Treat fraud signals as review evidence, not a verdict, and escalate legal or policy interpretation questions to me. Run the first review with me watching and ask me to approve the briefing format and any proposed follow-up actions before scheduling it. On future runs, report only material movement since the previous review, track the status of actions I approved, and surface the one decision that deserves my attention this week. Save the bot only after I approve the first review.
via @affonsoio
Same category. Different job.
Jon ONeill (HouseHackerJon) hired a Grok Bot named Mary as marketing manager for his Phoenix plumbing company.
Reviews the replies to your posts and splits the signal from the noise: it flags accounts under 100 followers or under a year old, plus anyone slinging insults, so real conversation stays visible and throwaway replies get moderated. Jason Calacanis built it to keep his mentions classy.
Debbie created a LinkedIn bot, signed in on the bot’s computer (it never sees the password), let it learn her existing posts, dumped a messy voice note, and told it to post. It posted. Receipt, not a draft-only demo.
Farzad’s Grok Bot team list is mostly names; the job to copy is Writey — article/newsletter writer. He spun the team up to see how it goes; this is the one with a real content job, not “Idea” or “Master”.