In Andrew Warner's tour of agent-run businesses, the strongest ideas are not "give a bot a company." They are familiar jobs with a buyer: rescue neglected leads, assemble qualified research, compare quotes, or build one tiny tool for a repetitive problem. Agents can do much of the searching, drafting and setup. They do not create demand simply by being autonomous. The clearest counterexample in the episode spent money on ads, got clicks, and sold nothing.
Pick a buyer and a costly job before picking an agent. Sell a small, inspectable outcome, then use the agent for variable work and ordinary code or automation for repeatable steps. Keep a person responsible for data access, outreach, payments, claims and final delivery. Creator-reported revenue in this episode is an example, not a typical return or an audited result.
Watch Andrew Warner's Episode
Creator: Andrew Warner, The Next New Thing. He reviews clips from other creators rather than independently auditing each business. The episode is presented by Zapier; the Zapier MCP segment at 15:27 is sponsored.
Seven Ways an Agent Could Support Revenue
Each offer needs a buyer and a measurable outcome. A product demo, a quote and money in the bank are different kinds of evidence.
| Model and buyer | Agent job | Builder note |
|---|---|---|
| 1. Niche directory sponsorship Brands wanting a focused audience | Collect and organize useful cases, update pages, draft a newsletter and sponsor placements. Alex Finn's directory is live. | Alex reports $1,100 from two sponsors; we did not audit Stripe. Distribution still needs an owner. Start with a narrow audience and disclose paid placements. |
| 2. Inbound lead recovery Businesses already receiving inquiries | Find unanswered Instagram DMs, summarize intent and draft relevant follow-ups. | Meta says Muse requires approval to send. Measure qualified replies or booked calls, not message volume. |
| 3. Local website offer Trades and local firms with visible web gaps | Research prospects, build a sample page and prepare a specific improvement proposal. | Andrew's free-site idea opens conversations; it is not proof every owner wants a redesign. Verify the business and ask before publishing or representing them. |
| 4. Event sponsor research Meetup and event organizers | Find comparable events, identify sponsors and draft a short case for fit. | Another sponsorship is a lead signal, not permission for automated bulk outreach. Confirm contacts, relevance and the organizer's offer. |
| 5. Quote-comparison concierge Buyers facing complex purchases | Gather contractor quotes, track assumptions and ask the buyer when a supplier needs a decision. | Chris Koerner reports eight pool quotes in 24 hours. The range is not a realized saving; compare scope, exclusions, qualifications and contract terms, not price alone. |
| 6. Marketplace research Resellers with local inspection capacity | Screen listings, compare sold prices, estimate fees and ask before contacting sellers. | A free sofa's asking-price comparison is not profit. Include condition, collection, shipping, returns and time. No margin until the item sells. |
| 7. Tiny paid software People with one recurring task | Find complaints, build a narrow tool, connect checkout and draft support content. | Jordan's Invoice Bot offers paid reminders; the episode does not establish retention or profit. Recurring reminders should run in code after the agent builds the system. |
What the Examples Actually Prove
The directory is a business model, not a passive-income guarantee. The live Agent Use Cases site has pages, a newsletter and a sponsor offer. Alex shows a Stripe screen and says two placements produced $1,100 in the referenced video. This is his reported result at the time, not a ledger we verified, and it does not show audience acquisition cost. His site has a human operator. The replicable lesson is the link between a useful niche audience and a sponsor with a reason to reach it.
Lead recovery has a shorter path to value. Brandon Carter demonstrates Muse surfacing old Instagram conversations and drafting replies. The buyer already has inquiries; the pilot can compare missed leads, responses and appointments with the previous process. Meta's business announcement says sending, publishing and spending require approval. Do not evade that control with another agent just to send more messages. Review consent, context and promises before each first-pilot reply.
The $100 experiment is the useful failure. Tucker gave five agents a week and a $100 budget. His own video, replayed at 22:12, reports $29.87 spent, 118 ad clicks and zero sales. The agents built and advertised a product, then stopped before consuming the full budget. That shows operational ability and a stop rule, not product-market fit. Andrew's criticism is the right one: a human should challenge the buyer problem before paying for traffic.
Less autonomy can produce a more reliable service. Chris's pool-quote comparison is a compelling personal account, but an apparent price spread does not establish actual savings. A concierge must check specifications, license and insurance, schedule and the final contract. Similarly, a $9/month invoice reminder should not depend on a general agent running every day. Use the agent to design and improve the workflow; use predictable code for scheduled reminders, with human support for disputes and exceptions.
Watch tool costs as well as token costs. Andrew's sponsored Zapier section suggests moving routine steps out of agent loops. That is an architectural test, not a promise of a 71% saving for your business. Zapier's documentation says MCP is available on its plans and successful calls currently consume tasks from the plan allowance. Compare full workflow cost and reliability before replacing an agent step.
Which Idea Is Worth Your First Pilot?
| If you have... | Try... | Evidence to collect |
|---|---|---|
| Existing inbound inquiries | A reviewed lead-recovery brief for 10-20 old conversations. | Relevant replies, booked calls and time saved against baseline. |
| A relationship with local owners | One site or process audit and a draft improvement offer. | Owner feedback and willingness to pay for the next step. |
| Knowledge of a costly purchase | A buyer-authorized quote comparison with a fixed scope. | Comparable bids, buyer time saved and decision quality. |
| A recurring software complaint | A manual prototype for one paying design partner. | A completed task and paid renewal, not clicks alone. |
Copyable Business Pilot Prompt
This is an original JQ AI SYSTEMS pilot brief, not a prompt quoted from Andrew or the featured creators. It begins with a buyer interview and stops before outreach or payment.
Test one paid outcome
Keep the first run small and reviewable.
Watch the Nine Source Videos
These are the examples Andrew quotes or reviews. Each creator's result is their own report; the companion video provides the full context behind a short clip.
Business and leads
Research and experiments
Products and small operations
Episode Chapters
| Time | Topic |
|---|---|
| 00:00 | Agentic income |
| 00:27 | Alex Finn's directory |
| 02:15 | Give an agent a CEO role |
| 03:09 | Find business ideas |
| 04:39 | Build the agent team |
| 06:45 | Muse for Instagram leads |
| 09:54 | Automated lead follow-up |
| 10:57 | Hyperagent for local businesses |
| 12:18 | Find event sponsors |
| 13:57 | Sponsor outreach |
| 15:27 | Zapier MCP sponsored segment |
| 17:06 | Contractor quote shopping |
| 20:42 | Turn the workflow into a service |
| 22:12 | Five agents start a business |
| 23:33 | The human-in-the-loop problem |
| 24:54 | Testing the product |
| 26:51 | What the experiment got right |
| 27:18 | Marketplace flipping |
| 29:33 | Automated marketplace research |
| 31:12 | Find a problem and build a tiny tool |
| 33:18 | The agentic software stack |
| 34:57 | One problem, one tool |
| 35:24 | Invoice Bot |
| 36:27 | When the agent should step aside |
| 37:30 | Where agentic businesses are going |
Source Note and Tools
Checked 7 November 2026. Andrew's video and supplied transcript establish what he discussed and when. The nine videos above are the creators' own demonstrations. Official product pages below confirm features or access, not the creators' revenue, time savings or conversion rates. We did not audit their payment accounts, repeat their tests or verify that any result is typical. This is business-model analysis, not an earnings promise.
Agent products: Grok Bot and its approval controls; Meta Muse and small-business connectors; Hyperagent; OpenAI Dots and its access limits. Dots personal-plan rollout excludes the EEA, including Portugal, as of this check.
Workflow and examples: Agent Use Cases and its sponsor offer; Zapier MCP and task billing. The Invoice Bot site mentioned in Jordan's video was not reliably reachable during this check, so the video is linked as the source for that example rather than presented as a verified live service.