On November 2, 2026, Greg Isenberg posted that setting up Grok Bots for boring businesses and SMBs might be one of the best AI businesses right now. His sketch: give every agent its own computer, email and phone number, then charge like you are hiring help by the hour. The thread is aimed at builders. Owner-operators should steal the scoping idea and ignore the gold-rush framing.
Jean-Baptiste Coger's reply sharpened the buyer truth in a related Isenberg thread: nobody wants an AI agent. They want the Monday report they hate writing to already be done. That sentence should sit on your whiteboard before you buy anything with "agent" in the name.
TL;DR
- On November 2, 2026, Greg Isenberg argued that setting up Grok Bots for boring businesses and SMBs might be one of the best AI businesses right now.
- The operator lesson is narrower: give a supervised agent one hated recurring job, its own boundaries, and a human approver.
- Do not sell or buy an AI agent in the abstract. Buy the Monday report, the follow-up queue, or the intake draft already done.
- AI drafts, your team approves — especially when the agent has email or phone access.
Boring work is where service businesses actually leak time
In owner-led firms, the expensive tedium is rarely glamorous strategy. It is:
- Turning completed jobs into first-pass invoices.
- Nudging leads who requested a quote and went quiet.
- Summarizing weekly inventory or truck-stock exceptions.
- Cleaning appointment reminder lists before they send.
- Filing job photos into the right customer folder.
Those tasks have a done state. They repeat. They punish procrastination. They are perfect for a supervised agent and terrible for a vague "AI transformation" project.
What the Grok Bots pitch gets right
Dedicated environment. If an agent shares your personal browser session, you will not sleep. A bounded machine identity makes least privilege possible.
Clear commercial metaphor. Charging like hourly help forces sellers to define outcomes. Buyers can demand the same clarity internally: what output, by when, with what error rate.
Focus on boring businesses. Margin lives in operations. Hype lives in demos. Align spend with operations.
What the pitch gets wrong for a 20-person shop
You do not need five agents with phone numbers on day one. You need one supervised loop.
You should not let an agent email customers unsupervised because a tweet said agents have email now. AI drafts, your team approves remains the control plane.
You should not measure success by hours the agent was "online." Measure edit minutes and redo rate.
A practical boring-workflow menu
| Workflow | Agent drafts | Human approves | Kill criteria |
|---|---|---|---|
| Quiet-lead follow-ups | Personalized nudge emails | Manager sends | >20% rewrite rate |
| Weekly inventory exception list | Flagged SKUs from exports | Parts lead confirms | Missed stockouts |
| Job-to-invoice first pass | Line items from job notes | Bookkeeper posts | Coding error spikes |
| Appointment reminder hygiene | Cleaned list | Coordinator releases | Wrong-time complaints |
| Photo filing | Suggested folders | Office accepts | Misfiled warranties |
Pick one row. Run it for ten business days.
Two-week internal pilot
Day 1-2: write the role card and approval gate. Connect only the tools required.
Day 3-7: agent drafts daily; human times edits.
Day 8-10: decide keep, narrow, or kill. Document the decision on one page.
If you cannot decide, you do not have metrics. Add metrics before adding agents.
For agencies reading the same tweet
If you sell this work, sell the outcome and the supervision model, not the vibe. Show the role card, the approval queue, and the kill criteria. Serious owners will ask who gets blamed when the agent is wrong. Answer that before the demo.
Verdict: Buy one boring supervised workflow, Skip agent fleets
Verdict: Buy a single supervised agent on a hated recurring task with a clear done state. Verdict: Hold on giving agents their own customer email or phone until outbound approval is airtight. Verdict: Skip multi-agent fleets sold as modernity.
Isenberg's post is useful because it points builders at unglamorous work. Your job is to keep the unglamorous part and discard the gold rush.
Picking the first boring task with a scoring sheet
Score candidate tasks from one to five on pain, clarity of done, data cleanliness, and blast radius if wrong. Multiply pain by clarity. Divide by blast radius. The highest score wins. Tasks that score high on pain but also high on blast radius wait until approvals are proven.
Writing the role card in plain language
Goal in one sentence. Inputs listed as bullets. Output format with an example. Tools allowed. Tools forbidden. Escalation path. Kill criteria. If the card needs a second page, the task is too wide. Narrow until it fits.
Keeping the reviewer motivated
Rotate reviewers monthly if the queue is dull. Celebrate kill decisions as wins when quality slips. Pay overtime for the first two weeks of review if the pilot sits on top of an already full job. Unpaid review is how pilots silently die.
Examples of boring wins by trade
HVAC: overnight drafting of maintenance agreement renewal notes for human send. Plumbing: sorting photo sets into job folders. Dental: insurance card field extraction into a draft patient record. Legal intake: summarizing a new inquiry into a matter checklist. Landscaping: turning crew day sheets into draft invoices.
Each example shares a pattern: clear input, clear output, human gate. Copy the pattern, not the vendor logo from a tweet.
What to tell customers if they ask about AI
Be plain. We use software to draft routine paperwork. A person still checks anything that reaches you. That sentence builds more trust than claiming magic automation.
Field note
Run the pilot with a named owner, a written kill switch, and a Friday review. If those three artifacts do not exist, pause new tools and build them first. AI drafts, your team approves only works when the queue has a human name on it.
For this workflow around AI agents for boring SMB workflows, write the kill switch criteria in numbers: edit rate, error count, or hours saved. Numbers beat vibes when you decide to keep or cut the pilot.
Coverage when the reviewer is out
Document who covers the review queue when the primary reviewer is on vacation. Name a backup. Practice a one-day handoff. Pilots fail during vacations more often than during demos because nobody owns the approve button. Write the backup name beside the primary name on the sticky note that already sits on the monitor.
Why this matters for ai agents for boring smb workflows
The point of supervision is not bureaucracy. It is keeping customer promises accurate while software drafts the repetitive pieces. Teams that skip naming a backup recreate the same fire every school holiday week.
A short checklist worth printing
Named primary reviewer. Named backup. Written kill criteria. Friday failure review. Context pack link in the agent config. Send rights off until approval. If any box is empty, do not expand scope.
FAQ
Want help picking the boring workflow worth an agent?
In a free 30-minute assessment we list the tasks your team keeps putting off, pick one with clear outputs, and sketch a supervised agent loop.
What did Greg Isenberg say about Grok Bots for SMBs?
In a November 2, 2026 post he said setting up Grok Bots for boring businesses and SMBs might be one of the best AI businesses, describing agents with their own computer, email and phone number and charging like hiring help.
Should every small business buy a full-time AI agent seat?
No. Start with one recurring task you already hate, prove the draft quality, then decide whether a dedicated agent environment is worth it.
What counts as a boring workflow?
Weekly inventory summaries, appointment reminder drafts, lead follow-up queues, job-completion photo checklists, and similar repetitive work with a clear done state.
Why do people want the outcome instead of the agent?
As one reply to Isenberg put it, nobody wants an AI agent in the abstract — they want the Monday report they hate writing to already be done.
How do we keep control if the agent has email or a phone number?
Lock outbound sends behind approval. AI drafts, your team approves. Least-privilege tool access and an audit log are mandatory.
Is this only for agencies selling agents?
The X post is pitched at builders, but owner-operators can use the same scoping advice for an internal pilot.
What is a two-week test?
Pick one task, run it with a supervised agent daily, measure edit minutes and error rate, then keep, narrow, or kill.