Making Money with Grok Bot: Billy Howell's Agent Team
Billy Howell runs a 6,000-reader newsletter on a Grok Bot agent team: chief-of-staff bot, sales agent, overnight routines — plus a 4-week starter plan and the best business models to copy.
In this episode of the Startup Ideas Podcast, Greg Isenberg sits down with Billy Howell — who has tested hundreds of tools for The Rundown and runs The Arlington Bagel, a 6,000-reader local newsletter delivered every Thursday via Beehiiv, on a team of Grok Bot agents. The day-to-day runs on a chief-of-staff bot, a research agent, a sales agent, and automated routines that keep working while he sleeps.
The starting method is concrete: one project per Grok Bot account — mixing businesses bloats context, slows everything down, and burns tokens (he pays $200/month for a second account just for Shopify experiments). Build the chief-of-staff bot first, connect Notion, Slack, and Gmail, and have it audit the business on day one and nominate the three agents most likely to drive revenue. Then run a four-week plan: week 1 build the team, week 2 pure execution, week 3 hire-and-fire against real gaps (like adding an inbox agent), week 4 add routines and automate.
The operating details are copy-ready too: every agent reports to the chief of staff in five-line briefs — shipped, stuck, needs a human — and repetitive work gets scripted or pushed to make.com so agent tokens stay reserved for high-value tasks. On quality, Billy ran an adversarial QA loop with a panel of subagents over three rounds, lifting deliverables from roughly 50% to 90% done, then turned that feedback pattern into a reusable skill. The newsletter machine is fully automated: a research bot sweeps a source list twice a week into Notion cards, make.com formats each blurb, Beehiiv sends Thursdays, and the sales agent watches Gmail, prices ad slots, and just landed a local bagel-shop sponsor.
His business-model verdict: newsletters and directories are the lowest-barrier plays and they compound each other — a research agent drafts brand pages in Markdown daily and a coding agent publishes them. The site stack is Cloudflare + Porkbun + Astro + GitHub; he shipped a booking page from the beach in two to three hours. For anyone who wants AI agents running a real small business: pick one project this month and copy the setup.
FAQ
What's the lowest-barrier business to run with Grok Bot?
Billy Howell's verdict: newsletters and directories — the two lowest-barrier models, and they compound each other (a research agent drafts brand pages in Markdown daily, a coding agent publishes). He runs The Arlington Bagel, a 6,000-reader local newsletter, on an agent team, with a second account for Shopify experiments.
How do you build an agent team from scratch?
Start with a chief-of-staff bot connected to Notion, Slack, and Gmail; on day one it audits the business and nominates the three agents most likely to drive revenue (his audit picked a research agent, a sales agent, and a Beehiiv platform expert). Then a four-week plan: build the team, execute only, hire-and-fire against real gaps, then automate with routines.
How do you keep Grok Bot token costs under control?
Three rules: one project per account (mixed businesses bloat context, slow things down, and burn tokens); script repetitive work or push it to make.com so agent tokens go to high-value tasks; and open a second account for a second business (Billy pays $200/month for his Shopify experiments account).
How do the daily briefings and routines work?
Each agent files a five-line brief to the chief of staff daily or weekly: what shipped, what's stuck, what needs a human. The short format keeps context lean, the chief of staff dispatches the rest, and routines keep the team running overnight while Billy sleeps.
How did Billy raise agent deliverable quality from ~50% to ~90%?
He ran an adversarial QA loop: a panel of subagents critiqued deliverables over three rounds of iteration, lifting completeness from about 50% to about 90%. He then turned that feedback exchange into a reusable skill so other agents can review work to the same standard on their own.