Graded · 2026-08-22 · Greg Isenberg (channel)
Running a business with Grockbot agent teams
We graded the Grockbot agent-teams episode. The operating framework is real. The business is never shown. Receipts inside.
Mixed Bag
What's actually being done
Look, the setup is legitimately cool. Billy Howell runs a 6,000-subscriber local newsletter called the Arlington Bagel, and he's got a little team of cloud agents doing the work: a chief-of-staff bot routing everything, a research bot filling Notion cards twice a week, a sales bot watching the Gmail inbox, a social bot making Instagram graphics, and a dedicated bot that just answers Beehiiv questions. The blurb writing got shipped off to a make.com automation to save tokens. He's also paying for a whole second account for a Shopify experiment, and he's got a Cloudflare plus Porkbun plus Astro stack that spits out landing pages while he sits on a beach. Little robots everywhere, working while he eats free bagels. Part of me genuinely respects it.
His framework is the best part: week one build a small team, week two execute and touch nothing, week three add or cut agents based on what actually broke, week four automate. Only create an agent when you've earned it.
The Good
Here's the thing. The great operators all started with one boring thing done obsessively, and Billy actually has that gene. This stuff is worth stealing:
- Earn the agent. The chief of staff does the job manually first. Only after the output is proven do you spin up a dedicated agent for it. Best idea in the episode, and it's how real companies promote people too.
- Five-line end-of-day briefs. What shipped, what's stuck, what needs you. Every agent files one, the chief of staff rolls them up. Costs nothing, and the human stays the boss.
- Automate the stable stuff OUT of the AI. Once the format is locked, a dumb script beats a premium agent, because a script does the same thing every time and doesn't get moody when the model updates. That's a real operator's instinct.
- You make the platform calls, not the agent. His bot dithered three weeks on where the content should live until he finally said Notion, done. Agents will shop for the perfect answer forever if you let them. So will people, by the way.
- Narrow outbound. Five prospects Monday, three custom pitches, a human reads everything before it sends. Quality over spray. The spray guys all end up in spam folders wondering what happened.
- Adversarial review panels, three rounds, before a human ever looks at the work. He says output goes from half done to ninety percent done. I believe him.
- Newsletter plus niche directory as one machine. The directory catches search traffic, the referral clicks turn into warm sponsor leads for the newsletter. Quietly the most concrete business idea in the whole episode.
The Bullshit
Anyone non-technical can run a business end to end. Look. Your own demo needs the Shopify CLI, make.com, GitHub, Astro, and Cloudflare. That's not a non-technical guy getting saved by a tool, that's a technical guy doing a magic trick and crediting the wand.
The business is never shown. This is the one that gets me. Forty minutes on a show whose whole premise is dollar figures, and the case study has the financial disclosure of a lemonade stand. Even the lemonade stand has a jar you can count. And the big sales win?
There was something I missed like last week with a local actually bagel shop owner reached out and he actually invited me and my wife and we went over there this morning and he gave us free bagels and we're talking about him sponsoring and that's because Grockbot found it.The guy emailed HIM. That's an inbox filter, not a sales team.
He doesn't know how the product works. And he says so, out loud, mid-demo:
I like pretty sure on that. I actually hadn't thought.Pretty sure. On the podcast where you're the expert.
The Shopify lane is vapor. "I'm still setting those up" Later: "I'm not like an ecom expert. This is an experiment." An experiment is fine. Rockefeller ran experiments too. He just didn't sell tickets to them.
The opener wrote a check the episode never cashed.
Most of the content I see around Grockbot is wrong. They totally don't understand why this tool is powerful. And I'm going to just transmute all that information into people's brains because I'm so excited about this and it's all I think about.Followed by a demo with zero numbers in it.
The Contradictions
- Pitches the tool as effortless for normal people, then demos command-line tools, a static-site framework, and third-party automation glue. Pick a lane.
- Preaches "run it like a real business," shows a case study with no revenue, no costs, no margin. Real businesses are made of exactly those three things.
- When Greg asked the hardest question of the night, how do you keep agents from looping into slop, the answer was "I don't, Greg, really." And look, I respect the honesty. But that's also the ceiling of the whole pitch, said out loud by the guy pitching it.
The Grades
Greg Isenberg, host: 7.7. Greg did the job. He asked how a $200-a-month tool pays for itself, he made the guest walk the sales flow, and he asked the slop-loop question that cracked the whole thing open. Twice he explained the guest's own framework better than the guest did, which is a skill. But here's the miss, and it's the one that matters: on a show whose premise is dollar figures, he never once said "what did the newsletter make last month." One question. The whole episode changes. That, and some unearned "absolutely brilliant" co-signing at the end, keeps a good night from being a great one.
Billy Howell, guest: 4.7. Dead middle of the board, and that's exactly right. The discipline is real. The cadence is real. I'd steal half his playbook and I basically told you to. But the great ones bring the ledger, and Billy brought vibes with excellent formatting. A hobby newsletter with zero revenue disclosed, a second business that's half built, and a guess about how the product works, delivered on camera. Competent tool operator. Not yet a businessman with something to show. Come back with a number, Billy. I'll be here.
Scores are opinions based on the published episode, backed by direct quotes. We grade the performance, not the person.