Video summary

I Let AI Make and Save Me Money for 6 Weeks (Here’s What Worked)

Main summary

Key takeaways

Business

Overview

Over six weeks, Chris Koerner tested AI agents that take actions—not just answer questions—including researching vendors, sending messages, negotiating, building websites, and handling follow-ups. He presents six business opportunities based on using agents to reduce manual work or offer these capabilities as customer services.

Many examples are personal experiments. Some business models are suggestions rather than proven, revenue-generating ventures.

Six Business Plays and Operating Tactics

1. Quote-Gathering and Purchasing Concierge

For a planned backyard pool, Koerner gave an agent his address, survey, design preferences, photos, and a request for rough estimates. The agent contacted pool builders through web forms and email, handled follow-up questions, sent requested documents, and recorded replies in a spreadsheet.

After two weeks of in-person conversations with four builders, Koerner had no quotes. The agent contacted 82 companies and produced eight quotes within 24 hours. Prices ranged from $120,000 to $240,000 for the same project.

Business model: Offer quote research and comparison for pools, roofs, kitchens, bathrooms, and other large projects. Koerner cites a possible 3–6% fee—about $6,000–$12,000 on a $200,000 job—or a productized website customers can use to request quotes.

Playbook:

  1. Collect the customer’s specifications.
  2. Request ballpark prices before scheduling site visits.
  3. Compare quotes, reviews, and references.
  4. Bring only the strongest candidates into a more time-intensive process.

The pitch to vendors is an asymmetric opportunity: a few minutes to respond could lead to a large job, while the buyer avoids spending hours collecting preliminary estimates.

2. Automated Marketplace Sales and Local Procurement

For selling household items, the agent researched comparable Facebook Marketplace listings, created listings from photos, selected titles and prices, answered questions, negotiated, and coordinated pickups through the seller’s calendar.

Pricing rule: If an item receives three or fewer inquiries in a day, reduce its price by 10%. Continue until it sells.

The same approach can help buyers search and negotiate for equipment or services. Examples included finding a skid steer, buying tires for a family member, and sourcing spray-foam repair quotes. A conventional spray-foam quote was $12,000; the first quote found through Marketplace was $1,200.

Business model: Provide “consignment without a shop” for people clearing out homes or handling a move, taking a cut when items sell. Alternatively, help homeowners obtain multiple local contractor quotes.

Price alone does not establish quality; compare several quotes and check vendors.

3. Agentic Dropshipping for Fragmented Product Categories

Koerner proposed using an agent to compare suppliers in categories where products are difficult to ship, inventory is inconsistent, and pricing is hard to compare. Fruit trees were his example.

Playbook:

  1. Build a clean, niche storefront.
  2. Have an agent research products and suppliers.
  3. Price items above the lowest suitable supplier price.
  4. When an order arrives, have the agent place the supplier order using the customer’s delivery details.

Agents could monitor supplier emails and notify customers or issue refunds if an item is unavailable. This model is proposed, not demonstrated as a completed business.

Operational risks: Inventory data may be inaccurate, shipping can fail, and refunds or customer service still need to be handled reliably. The business depends on enough markup to cover these costs.

4. Low-Cost Local Lead-Generation Websites

Koerner used demand-versus-supply research to choose niches and locations, then had Grokbot buy domains, build sites, configure analytics and search tools, and report rankings.

He reported spending about $11 per domain, plus a few dollars in agent credits. Within roughly a month, a crane-rental site targeting Laredo generated three inquiries—one described as a $3,000 lead, another about $23,000, and another about $9,000. These figures refer to the potential value of the inquiries, not confirmed sales.

He also tracked a shower-door site in Brownsville. One keyword reportedly improved from Google position 14 to 10 between September 28 and October 5. Other rankings moved in the opposite direction, underscoring that results vary.

Playbook:

  1. Find underserved local searches and weak search results.
  2. Build a focused site.
  3. Track rankings and inbound leads.
  4. Give initial leads to local providers in exchange for feedback.
  5. Seek payment or exclusivity from providers that convert them.

Sales tactic: Start by offering leads for free and asking whether they are useful. If a provider closes business, propose a monthly fee. Koerner suggested an illustrative increase from $800 to $1,200 per month once a site consistently delivers value; this was not a reported customer outcome.

He recommends considering less crowded niches, such as specialty repairs or shower doors, rather than automatically pursuing highly competitive trades.

5. Agent-Assisted Prospecting for AI Consulting

The suggested customer-acquisition process has three stages:

  1. Lead sourcing: Use Google Maps to find businesses in one niche and city with 4+ stars and 50+ reviews, no booking link, and no owner replies to reviews. The agent can also test whether calls go to voicemail, submit contact forms, track responses, and watch local groups for repeated recommendations.
  2. Audit: Score each qualified business’s booking process, reviews, site speed, and Google Business Profile. In the example, screening 100 gutter-cleaning companies yielded nine audits worth reviewing.
  3. Personalized outreach and demo: Send a short Loom walkthrough—Koerner suggests 3–7 minutes—and prepare a working demo, such as an AI receptionist, missed-call text-back, or booking page. Personalize it with the business’s branding and use transparent assumptions to estimate potential impact.

The strategy combines scalable agent research with human trust-building. Koerner says first AI-consulting customers often come through existing relationships, so cold outreach should supplement—not replace—one’s network.

Niche selection: Use agents to explore industries beyond the most obvious, crowded categories. The suggestion is to “go wide and go weird,” while validating actual local demand.

Koerner offered a $500 reward to a viewer who followed the described process and found a customer within a week of the episode going live.

6. Packaging Agent Capabilities as a Product or Service

A friend used an agent to search his email for class-action settlement notices and file eligible claims. It found eight; he reportedly received about $200, with up to another $2,000 potentially pending. Koerner tried the same approach and found seven settlements, but the filing deadlines had passed. The example illustrates both the potential value and the importance of deadlines.

Product idea: Turn a useful agent workflow into a simple consumer service—such as a tool that finds eligible settlement claims—and make the underlying AI invisible to users. Koerner suggests charging a small fee or refunding users if the service finds less than the fee’s value.

Another example was a phone-based reminder service for people with ADHD: a user calls a number to request a reminder, and the service calls back at the specified time.

Research playbook: Study a company already succeeding in a market—its ads, backlinks, website changes, and historical positioning—to understand its acquisition strategy and evolution before building a competing offer.

Agent Setup, Management, and Risks

  • Match tools to the work: Koerner says Muse was most convenient for Meta products and many Google services; Grokbot for X-related, real-time work and website tasks; Instinct for messaging-based workflows; Claude for planning and technical instructions; and Replit for building and testing websites or apps.
  • Use a reusable “about me” brief: Ask a familiar AI tool to summarize what it knows about the user, their work, preferences, and projects, then transfer that context to another agent.
  • Retain reusable skills: A team member stores agent instructions and learned workflows in a shared private GitHub repository, with a weekly process for adding reusable skills. This reduces dependence on any one agent platform.
  • Keep humans in the loop where needed: Koerner reported email errors, incomplete sending, and agents asking for permission. He also described a waiver site that initially exposed sensitive information. He then used Claude to define security requirements and an agent to apply and publish changes through Replit. The lesson is to test agent-built products and take privacy, security, spending limits, and customer communications seriously.
  • Treat early results as experiments: Koerner frames many of these opportunities as low-cost tests, but rankings, leads, quote savings, and proposed business models are not guarantees of future outcomes.

Metrics and Figures Cited

  • Pool outreach: 82 companies contacted; eight quotes in 24 hours, with prices of $120,000–$240,000. Koerner later said he had received more than 10 quotes.
  • Pool quote-research service: Cited potential fee of 3–6% of project cost, or $6,000–$12,000 on a $200,000 project.
  • Spray-foam repair: Conventional quote of $12,000 versus an agent-sourced quote of $1,200.
  • Marketplace pricing: Reduce the asking price by 10% on days with three or fewer inquiries.
  • Local lead sites: About $11 per domain, plus a few dollars in tokens; three crane inquiries described as worth $3,000, $23,000, and $9,000.
  • Prospecting example: 100 businesses screened, yielding nine audits.
  • AI adoption claim: Koerner cited a report saying 2% of U.S. households pay for AI.
  • Class-action search: A friend reportedly received about $200, with up to $2,000 more potentially pending.

Presenter and Sources

  • Presenter: Chris Koerner, speaking solo.
  • Tools and companies referenced: Muse (Meta), Grokbot (xAI), Instinct, Claude (Anthropic), Replit, Google Maps, Facebook Marketplace, Loom, and Mercury. The episode also includes a promotional segment for GoHighLevel.

Rate this summary

Your feedback will help improve summaries.

Improve this summary

Reprocess with a stronger model when the summary feels incomplete or inaccurate.

Pro

Translate summary in another language

Pro

Ask questions to this video

Chat for follow-up questions, clarifications, and source-backed answers.

Coming soon

Share this summary

Original video