Codie Sanchez tests five AI bots on turning $100 into $1,000

Key ideas

• The agent jobs that paid were the boring ones: outreach to a clean lead list, reselling free items and following up with people who almost bought.
• The flashy job, trading stocks on social media buzz, lost money, because an agent running a bad strategy just loses faster.
• Split a job into one agent per step with hand-offs, and put a cost check early enough to drop a bad deal before any money moves.
• Set a hard price ceiling and have a person approve anything that spends money or signs a contract.
• Try an agent on your own work before you sell it, since one that doesn't save you time won't sell.
• Your existing list, of past leads, dead quotes or abandoned carts, is the cheapest place to point an agent.

Codie Sanchez gave five AI agents ten days and $100 to make $1,000, and the result runs against the hype: the boring jobs paid and the flashy one lost. Cold outreach to local business owners produced the most value, reselling free items cleared a small profit, and the bot trading stocks on social media buzz lost money. These are my notes on how each bot was set up, what it made by her count, and the setup rules worth reusing. She used Grok Bot, but the setups would work with any agent that can browse the web and work inside your apps, and that’s how I’ve written them up.

The setup and rules

She set three rules. Once a bot is set up, she does none of the work. She spends no more than $100. And money only counts if it’s cash in hand or a real saving in her business. Each bot gets a name and a plain-language brief, and a “spawn bot” builds the others by asking whatever it needs to know about the goal.

Resale bots

Three agents share the job of flipping free items from Facebook Marketplace. The first finds free listings that should resell for a profit. The second gets a pickup quote, and if delivery costs more than the expected profit, the item is dropped right there. If it clears, that agent books the courier. The third relists the item at a competitive price and answers buyers.

The agents passed on a sectional that wouldn’t fit and were wary of a fridge they couldn’t test. The first find that cleared every check was a bowling ball, which she says took 31 minutes from sourcing to arriving at her office and sold for about $20 profit. She had to take the product photos herself. Across that and items from around her office, she gives the total as $130 in revenue against $41.98 in costs, about $1 of it agent usage, for $88.02 net.

Her point is that finding free stuff is easy, and the money is in knowing whether the pickup is worth it. She then points the same three-agent shape at a bigger number: sourcing parts for a business. You set the target price and the line the agent never crosses without you. One agent sends your parts list to every nearby distributor for a quote. The second compares quotes including the hidden costs, such as freight and order minimums. The third takes the best number back to the others and asks them to beat it, until it reaches your target or runs out of room. You approve the final deal. Her example: a business spending $200,000 a year on parts that gets 8% off saves $16,000 a year.

Clipping bot

Clipping is cutting long videos into short clips for social media. She set an agent up to find clients, edit the footage and deliver it. It got stuck on Fiverr’s “are you human” check, so she moved it to Upwork, which needed a $21 plan to bid. After four days it found one lead, five clips for $40 in total, and pitched and chased it, but the client went elsewhere. A new account with no reviews couldn’t compete with freelancers with years of ratings. Revenue from outside clients was $0, and she puts the costs, including a voice tool and the Upwork plan, at about $32.

The clips themselves held up. She posted one the agent edited from her own long-form video to her main account, she judged its views against her own average and kept the bot, and no commenter flagged it as AI-made. Her editor now uses the agent as an assistant: it scans raw footage for the best moments and gives him a first cut to polish. She says she pays editors $50 to $250 per short, so even one short a week at $150 saved is about $600 a month. Her lesson: try an agent on your own work before you sell it to anyone.

Re-engagement campaign

Here the agent sold her new book to people who’d come to her launch event but hadn’t bought it. That’s a list every business has in some form: people who asked for a quote, signed up, or bought once. She told it to build a landing page in Shopify, write the follow-up emails, and send them to the first 3,000 people on the list.

The first landing page was generic and she sent it back with notes: it needed to make the reader want something, feel something and act now. After a few rounds it was good, and it had pulled real reader reviews from Amazon onto the page without being asked. The agent loaded the emails into Klaviyo, her email tool, and sent her the drafts to approve before anything went out. Two hours later the list had bought two books, from a list she says had already been emailed heavily. Her figures are $55.48 in revenue against $41 in costs, for $14.48 net.

She then generalizes it to a service business with dead quotes, using a roofing company with 500 quotes that never closed. The agent sorts them by why the customer said no: price, timing, chose someone else, or went quiet. Each group gets its own email. The price group gets a financing option, the timing group is asked whether now works, the ones who went elsewhere are asked how that job went, and the quiet ones get a two-line “do you still need a roof?” The agent marks the “not interested” replies and leaves them alone, answers questions, and books calls with anyone who says yes, so the team only has to close. If 2% come back, that’s 10 jobs.

Cold-lead bot

The last agent filled seats for a workshop her company runs in Austin, from cold leads only. A lead-finding agent searched Google, LinkedIn, Apollo and Reddit for owners of businesses doing at least $1 million in revenue, with their email or LinkedIn, into a spreadsheet. Before anyone was contacted, the agent checked her HubSpot customer records, so no one her team was already talking to got a second pitch.

She had it write three versions of the email to see which got replies, each worded for where the reader is: “here in Austin” for Austin, “here in Texas” for elsewhere in the state. Sending started slowly, 50 emails a day from one inbox. Every address went through Hunter first, an email-checking tool, to keep bounces down. Only once the list was clean did it move to four inboxes and 200 emails a day. Replies went to her sales reps, who took over by email, phone and text.

After a few days, 10 owners said they were interested and two booked calls. She values each reply at $100 and each booked call at $500, which she says is what those cost her through Meta ads, and on that basis calls the result about $2,000, against $57.26 in costs. That figure is her valuation of the leads, not cash received.

Trading bot

The trading agent scanned Reddit, X and the news for stocks with buzz and traded them in a Robinhood account. It started at a loss, was a few dollars up by day four, and ended the ten days down. She couldn’t tell what strategy it was following: it bought at a high, then again after a drop, then again on the way back up. Telling it to make money didn’t fix that.

Her read is that the bot did exactly what she asked, and the problem was the strategy. Trading on social media hype is a bad process, and an agent only runs a bad process faster. She fired it.

What I’d reuse

Across the bots that worked, the same few rules carried. Point the agent at a list or a process that already works. Give each agent one step and a clear hand-off. Put the check that kills a bad deal before any money moves. Set the limits up front, a price ceiling, a sending rate, a person approving spend, so the agent can’t talk itself into a bad decision. And measure in cash or real savings, saying plainly when a number is an estimate instead.

Key terms

Spawn bot

Her name for an agent that sets up the other agents: it asks the questions it needs about the goal, then creates the new bot.

Hand-off

One agent passing its result to the next, such as a sourcing bot handing a listing to a bot that prices the pickup.

Re-engagement funnel

A landing page and follow-up emails aimed at people who showed interest but didn't buy.

Dead quotes

Quotes a business sent that never closed, sorted by the reason the customer said no.

Bounce

An email that can't be delivered. Enough of them tell mail providers your domain may be spam, so even good emails stop arriving.

Email verification

Checking each address is real before sending, which keeps the bounce rate down. She uses Hunter for it.

Quotes from the talk