A bagel shop owner emailed Billy Howell about sponsoring his local newsletter.
Billy missed the email.
His AI sales agent didn't. It flagged the thread, priced the ad slot, and told him exactly what to pitch. This morning Billy and his wife were at the bagel shop eating free bagels and talking sponsorship terms.
That story is from Greg Isenberg's new episode with Billy Howell, and it's the best walkthrough I've seen of running a real business on Grok Bot agent teams. Not a demo. An actual local newsletter with a research agent, a sales agent, a social agent, and a chief of staff routing all of it.
Greg's framing: these are businesses doing $2,000, $5,000, $10,000, $25,000 a month. Small ideas anyone non-technical can run end to end.
Here's the full episode. My breakdown below.
The business: a local newsletter with an agent payroll of zero
Billy's newsletter is simple. Three things to do every weekend, plus new restaurant news, sent every Thursday through Beehiiv.
He's run newsletters in tech and sports for clients before. His take: all you need to start is a niche and a list of where the best information lives.
Into board games? You already know the best subreddits, review sites, and message boards. That's your source list. Aggregate, editorialize, send weekly. That's a newsletter business.
The difference in Billy's version is who does the work. A research agent fills Notion cards with story candidates twice a week. A sales agent prices and pitches ad slots. A social agent makes the Instagram graphics and runs the content calendar. A chief of staff coordinates all of them.
Billy mostly talks to one bot. The chief of staff.
Start with a chief of staff, not 15 agents
This is the part most people get backwards. They spin up ten agents on day one and drown.
Billy did the opposite. He connected Gmail, then told Grok Bot: "Take stock of this business. Tell me the top three agents we should create first to drive revenue."
It audited everything and came back with three:
- A research agent for newsletter items
- A sales agent for ad slots
- A Beehiiv expert agent for platform questions
That last one is a pattern worth stealing. If your business depends on a platform (Beehiiv, Shopify, whatever), give it a dedicated expert agent so your chief of staff doesn't have to know everything.
Two rules Billy is strict about:
→ Don't let agents make your decisions. His agent spent 3 weeks unable to choose between Notion, Google Sheets, and local files for content storage. Billy finally said "we're doing Notion, no more tinkering." You're the founder. Decide.
→ Earn the right to a new agent. Have the chief of staff do a task once, review it, confirm it works. Then say "now create a dedicated bot that does what you just did." Never create an agent for a job that hasn't been done successfully at least once.
The four-week framework
Greg summarized Billy's system at the end of the episode and it's the cleanest version of this I've heard:
- Week 1: build the team. Chief of staff plus three agents. Learn to fly the plane. Use the team to buy a domain, set up the site, build your first sales list.
- Week 2: execute, no tinkering. Billy banned himself from creating new agents or touching skills. Pure execution.
- Week 3: hire and fire. The week exposed the gaps. Someone needed to man the Gmail inbox. Someone needed to run the merch shop. Those became agents.
- Week 4: automate. Billy asked his chief of staff: "Based on how we ran this week, what routines would move the ball forward while I sleep?" It wrote the plan itself.
The routines are mostly end-of-day briefs. Each agent reports five lines: what shipped, what's blocked, what needs Billy. Five lines, no more. Anything longer is burning tokens.
Then the chief of staff rolls it up into one view: what's shipped, what's stuck, what needs you.
It's a smaller, scrappier version of the setup Allie K. Miller runs with her 34-agent workforce. Same org-chart thinking, one-person-newsletter scale.
The QA loop that takes output from 50% to 90% done
This might be the single most useful tactic in the episode.
When an agent finishes a piece of work, Billy doesn't review it himself first. He says: "Create a panel of subagents and have them do an adversarial review of your work. Three rounds."
His numbers: output goes from roughly 50% done to 90% done before he ever looks at it.
And there's a second layer. After Billy gives feedback on something (say, a batch of Instagram posts), he tells the agent: "Take what we did in this thread and turn it into a skill another agent can use." Now there's a reviewer bot that critiques exactly like Billy.
Greg's been building the same muscle in his own stack (I broke down his 9-piece AI employee setup a few days ago). The pattern is converging across every serious operator: agents do the work, agents trained on your taste review the work, you review the exceptions.
A sales agent that actually finds money
Back to the bagel shop, because the setup behind it is replicable.
Billy gave his sales agent one founding task: set up and price everything we sell. It looked at the ad slots, repriced them against the market and the newsletter's growth, and produced a sales sheet ready to show any local business.
Then he added an inbox monitor. That's the agent that caught the bagel shop owner's email Billy had missed, and told him to follow up with a specific ad slot pitch.
His outbound philosophy is quality over quantity, and his cadence proves it:
→ Monday: an automation adds 5 prospects to the list.
→ Then: pick the top 3, build a custom ad package for each, draft the outreach.
→ Every draft goes to the chief of staff, which runs it by Billy before anything sends.
His words: you'll make more progress doing that than blasting 200 cold emails at a time. He skips the subdomain warming and the spray and pray entirely. Five good prospects a week, three custom pitches.
Automate the boring stuff for pennies
Here's a detail I loved because it's so unsexy.
Billy doesn't use Grok Bot to write the little two-sentence event blurbs in the newsletter. A make.com automation with a cheap OpenAI key does that. Grok Bot filters 200 candidate stories down to the best ones, then the widget formats them the same way every week: emoji, title, time, summary.
His reasoning: agents are a black box. Run the same prompt next month and the model might write differently. Once a workflow is business critical, you strip out the ambiguity and pin it down as a boring, deterministic automation. That's the whole discipline behind good business automations: build with the smart tool, then lock the repeatable part into something dumb and reliable.
His sequence: build, execute, then automate. And never automate a process you haven't first done manually with the agent. Billy had Grok Bot writing those blurbs by hand for weeks before he pinned the format.
The employee analogy lands too. You have a high performer on the team. You don't spend their hours writing two-sentence summaries of 100 items you might not use.
The $0 site stack: Astro, Porkbun, Cloudflare
One more system worth stealing. Billy keeps a "site builder" agent with a fixed stack: Astro for static sites, Porkbun for domains, Cloudflare for hosting, connected to his GitHub.
He needed a booking page and didn't want to pay for Calendly. So from the beach, on his phone, he told the agent to embed his three Google calendars on a page. It sent him mockups, spun up preview links for each style, and when he picked one it connected the domain and went live.
Total time: 2 to 3 hours. From a beach.
Where this gets interesting is directories. Billy's playbook: a research agent investigates one brand a day and writes a page in markdown, a coding agent publishes it on the Astro stack, and once traffic shows up, another agent collects referral links and does outbound for deals.
Greg's numbers on directories: solid ones become $10K, $25K, even $100K a month businesses. And they pair naturally with a newsletter in the same niche. The directory captures search, the newsletter captures the relationship, the referral links and sponsors monetize both.
Billy's ranking of lowest-barrier Grok Bot businesses: newsletter first, directory second.
What I'd steal from this episode
→ Chief of staff first. One agent that knows the whole business, before any specialists.
→ Three agents max to start. Add a fourth only after a real gap shows up in execution.
→ Five-line daily briefs. What shipped, what's stuck, what needs you.
→ Adversarial review panels, three rounds, before you ever see the work.
→ Automate only what you've already done manually. Then pin it down cheap.
Billy's parting challenge is the right one: pick one project, run it like this for a month, and pursue it aggressively. That beats researching tools for 4 weeks (I've watched a lot of founders do the second thing and call it work).
Greg's closing note: Rowan didn't think The Rundown would hit millions of subscribers when he started. You start small, you automate what you can, you keep the team lean.
FAQ
What is Grok Bot?
Grok Bot is xAI's agent product for building teams of AI agents that run tasks end to end. You create named agents with roles, connect them to your tools through plugins (Gmail, Notion, Beehiiv, Shopify), set routines that run on a schedule, and the agents message each other to coordinate work. Each agent gets its own computer it can work on, and you can run it from your phone.
How much does it cost to run a business on Grok Bot agents?
Billy pays $200 a month for a dedicated second account (he keeps his Shopify experiment on its own account to separate context). The newsletter itself runs on Beehiiv plus a make.com automation with a cheap OpenAI key for formatting blurbs. Compared to hiring even one part-time VA, the whole stack is a rounding error.
What's the easiest business to start with Grok Bot agent teams?
Billy's answer: a newsletter, then a directory. A newsletter needs a niche and a source list, and agents handle research, formatting, and sponsor sales. A directory needs a research agent writing one page a day and a coding agent publishing it. Both monetize through sponsors and referral links, and they compound together in the same niche.
Do you need to be technical to do this?
No. Billy's whole pitch is that a non-technical founder can run this. When an integration needs a command line tool (like the Shopify CLI), the agent installs it on its own computer. When it needs credentials, it drops a card in the chat where you paste an API key or click through OAuth. The hard part isn't technical. It's having the discipline to decide fast and stop tinkering.
I interview bootstrapped founders running playbooks like this every week. Real numbers, real systems, no filler.
