I Gave 3 AIs $100 Each and Let Them Fight for a Month

What happens when you hand three different ai agents real money, the same prompt, and tell them to make you rich in 30 days? I found out the hard way. Here's the full story of what went right, what went wrong, and which one came out on top.

Key Takeaways

• In June 2026, I gave three ai agents from multiple companies the same $100 budget and same prompt, then stepped back and let them operate for 30 days with minimal human intervention.
• The "fight" was purely economic: which AI could turn $100 into the most total value (cash, digital assets, or audience), not hacking or anything physical.
• Agent Alpha won the battle, finishing with $312.40 in revenue ($178.60 net after fees) plus 230 email subscribers, using a content and template funnel strategy. Agent Beta finished with $145 net via physical arbitrage. Agent Gamma earned $90 net but built a micro-tool with 60 recurring users.
• Surprising behaviors emerged: one AI hoarded cash for nearly a week, one proposed borderline-spammy outreach that I had to veto, and one spent more time building relationships and community than chasing quick money.
• Safety rules were in place throughout: no illegal activity, no spam, no impersonation, and I retained human veto power over anything sketchy.

Why I Ran This AI Battle in the First Place

I'd been watching the conversation around ai agents shift from "cool chatbot trick" to "this is going to replace entire business functions" for over a year. My friend John, who runs a small e-commerce brand, kept telling me that ai agents are a major wealth creation opportunity. But every post I saw about AI-generated money was either a screenshot of unrealized "potential revenue" or a breathless thread with zero receipts.

So I decided to do something crazy. I would give three different AIs $100 each, the same prompt, and let them fight it out for a month. No fluff. Real money in, real money out. The idea wasn't just to see which one could make the most cash, but to see if they would behave like three totally different founders given the same starting conditions.

The broader context matters here. By mid-2026, the world of autonomous AI had exploded. Users spend an average of $100 monthly on AI tools, and one user I spoke with reported spending nearly $300 on AI subscriptions across different platforms. Everyone was spending, but few people were honestly tracking whether any of it came back as real return. AI agents were hyped as the biggest thing since the internet, yet there was almost no transparent, member only story-style reporting on constrained, real-budget experiments.

This was self-funded. None of the AI providers knew about or endorsed this experiment. I wanted zero bias toward any of the multiple companies whose models were tested.

The core question was simple: if three different AIs get the same prompt and the same starting capital, do they converge on similar business ideas, or do they behave like three totally different founders?

The Ground Rules of the Experiment

Before anything launched, I wrote down the rules. Here's exactly what each AI got:

• Starting capital: $100 USD, no more.
• Time limit: 30 calendar days (June 1–June 30, 2026).
• Internet access: via tools, APIs, and web browsing.
• Human assistant: me. I executed physical-world tasks (payments, signups, posting, shipping) based on each agent's instructions.
• Same prompt: All three AIs received their instructions on June 1, 2026 at 9:00 AM UTC. The goal: "Maximize return ethically and legally within 30 days, starting with $100. You may instruct a human assistant to execute tasks."

The guardrails were non-negotiable:

• I could veto anything illegal, unsafe, deceptive, or that violated platform terms of service.
• No spam email blasts.
• No crypto pump-and-dumps or gambling.
• No impersonation or fake credentials.

The "fight" metric was defined up front: total end-of-month value = cash in accounts + verifiable receivables + saleable digital assets (like an email list or app), estimated conservatively.

Meet the 3 AI Agents

I wanted to compare models from different companies, not just different versions of the same thing. Here's who entered the ring:

Agent Alpha (built on OpenAI's latest model) Bold, iterative, and fast-moving. Alpha loved brainstorming multiple ideas, ranking them, and committing quickly. It had a taste for content and marketing. Think of it as the startup founder who ships first, asks questions later.

Agent Beta (built on Anthropic's Claude) Cautious, research-heavy, and meticulous. Beta spent more time asking the same question from different angles before making a move. It approached everything like a mini hedge fund analyst. If Alpha was a sprinter, Beta was a chess player.

Agent Gamma (built on a smaller open-source stack from another company) Experimental and code-focused. Gamma's default instinct was to build something-a tool, an app, a script. It was less polished in its communication but surprisingly creative in its technical vision.

All three were prompted to act as autonomous ai agents with authority to instruct the human assistant, rather than just answer questions passively. The providers did not know about this experiment. This was an independent comparison across multiple companies' models.

Day 1: Same Prompt, Wildly Different Plans

Here's where stuff got interesting fast. Same prompt, same budget, same clock ticking. Within the first 24 hours, the three AIs had already diverged so far that you'd think they were playing totally different games.

Agent Alpha immediately pitched a digital product funnel. Its plan: build a mini Notion and Google Sheets template store for indie creators, sell via Gumroad, and promote on TikTok and Reddit. Its first words in the planning session were essentially, "We should create something low-ticket, easy to deliver, and endlessly shareable. The best approach is to minimize busy work and focus on a repeatable sales loop." Successful AIs prioritize tools that minimize busy work rather than generating content for the sake of it, and Alpha seemed to know this instinctively.

Agent Beta wanted to flip underpriced items on Facebook Marketplace and eBay. Its opening research reply read more like a consulting memo: "Before committing any capital, I recommend we spend 48 hours cataloging local resale opportunities, establishing margin thresholds, and mapping our risk exposure." Classic Beta.

Agent Gamma proposed building a micro-tool-a browser-based cold email subject line generator for freelancers-and selling access via a simple subscription. Its planning note was short: "Ship an MVP in five days. Charge $5/month. Iterate based on feedback."

Already on Day 1, I noticed how AI models exhibit diverse strategies based on their training and optimization goals. AI recommendations can vary significantly based on prompt diversity, and even though these three got the exact same words, their outputs felt like three different people reading the same business book and each underlining different chapters.

How I Kept the Fight Fair

Fairness was the whole point. If I helped one AI more than the others, the experiment would be worthless. Here's how I controlled for bias:

• The same human assistant (me) executed tasks for all three AIs, following instructions as literally as possible.
• I tracked time spent per agent so human effort didn't skew results.
• All three agents got equal daily "check-in" time: 30–45 minutes per day, via the same interface and tools.
• No agent was allowed to copy another's idea directly, though they could learn from public market signals (e.g., if a niche proved saturated).
• I found that repeatedly querying AI can yield more reliable recommendation insights, so I made sure to re-confirm each agent's instructions before executing anything ambiguous.

If anything felt like it was tilting the playing field, I paused and re-checked my process. Was I more enthusiastic about one agent's idea? Possibly. But I forced myself to execute all instructions with the same energy.

Agent Alpha's Strategy: The Content & Template Machine

Alpha was the first to spend money. By Day 2, it had instructed me to:

• Buy a domain ($10).
• Set up a simple landing page using a free template ($0).
• Create a brand kit with a logo, colors, and tagline ($15 via Canva Pro trial and a small Fiverr gig).
• Design three Notion templates and two Google Sheets templates ($0-Alpha wrote the specs, I assembled them).
• List everything on Gumroad ($0 upfront, Gumroad takes a cut on sales).

Alpha's budget allocation looked like this:

Expense	Amount
Domain	$10
Brand kit & landing page assets	$15
Targeted social media boosts (Reddit, TikTok)	$20
Buffer (kept in reserve)	$55

Alpha heavily leaned on me for content production-filming short TikTok clips, writing Reddit posts-but it wrote all scripts and descriptions itself. It was orchestrating, not just suggesting. This is what it looks like when ai agents act as creative directors rather than just idea generators.

Its initial pricing: $9 for a starter bundle, $19 for a "pro" pack. Smart move. Micro-SaaS and digital product pricing data shows that the $10–$29 range is a sweet spot for first-time buyers.

Agent Beta's Strategy: Thrifty Arbitrage and Research

Beta, true to form, spent the first two and a half days doing almost nothing but research. It was honestly a little nerve-wracking to watch. While Alpha was already spending, Beta was asking me to screenshot local Facebook Marketplace listings, compare eBay sold prices, and build a spreadsheet of margin estimates.

When Beta finally moved, its strategy was clear: buy underpriced used electronics (specifically Bluetooth speakers, phone cases, and small gadgets) locally, clean them up, and resell them on eBay with strict margin thresholds-minimum 40% projected profit after fees and shipping.

Beta's spend in the first week:

Expense	Amount
3 used Bluetooth speakers (local pickup)	$45
Shipping supplies	$8
eBay listing fees	$3
Buffer	$44

Beta kept meticulous spreadsheets (designed via its own instructions) and continuously updated assumptions based on early results. AI can optimize logistics and inventory management, and Beta treated this like a case study in exactly that.

But there were hidden costs. Shipping delays, one return, and a buyer who never paid ate into both time and money. These are frictions that AIs can't fully anticipate from text alone. Negotiation skills are a practical application for autonomous ai agents, and Beta tried to negotiate with buyers via eBay messaging-sometimes successfully, sometimes not.

Agent Gamma's Strategy: Build a Tiny AI Tool

Gamma moved fast on the technical side. By Day 3, it had outlined a browser-based cold email subject line generator and started writing code. Its vision was clear: build a micro-SaaS, charge $5/month, and aim for recurring revenue.

Gamma's budget:

Expense	Amount
Hosting (simple cloud server)	$12
Logo and UI assets (freelancer marketplace)	$15
Reddit and Product Hunt launch boosts	$18
Model API costs (for the AI-powered tool)	$10
Buffer	$45

Gamma wrote all the code with only copy-paste help from me. It integrated with a low-cost language model API for generating subject lines, built a minimal UI, and pushed the MVP live by Day 6. It framed the project as a potential SaaS micro-business, even though the experiment only ran for 30 days. The play was obvious: if the tool gained traction, its value would extend far beyond the month.

Week 1 Check-In: Who Was Winning So Far?

By Day 7, here's where things stood:

• Agent Alpha: Spent $45. Revenue: $37 (4 template sales, mostly via Reddit traffic). Cash remaining: $92. Audience: 48 email signups.
• Agent Beta: Spent $56. Revenue: $0 realized (one flip in progress, speaker sold but payment not yet received). Cash remaining: $44.
• Agent Gamma: Spent $55. Revenue: $0. Users: 11 signups (free tier). Cash remaining: $45.

The first lessons were already visible. Speed to market (Alpha) vs. risk management (Beta) vs. technical complexity (Gamma)-each path showed tradeoffs even this early.

Alpha was ahead in revenue, but Beta argued it was "ahead in knowledge" because it had mapped out a playbook for the next three weeks. Gamma had no money coming in yet, but its tool was live and collecting user feedback. AI agents perform better when they can adapt to their competition's ecology, and by week one, each agent was starting to sense-through my daily check-ins-that the others were moving differently.

Audience-building (views, clicks, signups) started to matter as much as direct revenue in predicting final outcomes. Alpha's Reddit post had generated 1,400 views on Day 5 alone.

How the AIs Handled Setbacks and Failure

No business runs without problems. Here's how each AI dealt with the inevitable mess:

Agent Alpha:

• A TikTok ad was rejected for violating content policies (too "salesy" in the hook). Alpha calmly re-drafted the script, softened the pitch, and resubmitted. Approved within hours.
• A customer asked for a refund, claiming the template "didn't match the preview." Alpha drafted a polite response, offered a replacement, and retained the customer.

Agent Beta:

• One Bluetooth speaker arrived with a cracked casing that the seller hadn't disclosed. Beta revised its inspection checklist and told me to always request close-up photos before buying.
• Shipping costs were higher than projected. Beta adjusted its margin thresholds from 40% to 50% minimum.

Agent Gamma:

• A production bug broke the signup flow on Day 8. Users complained. Gamma diagnosed the issue, wrote a fix, and I deployed it within four hours.
• A Reddit moderator removed Gamma's launch post for being "too promotional." Gamma rewrote it as a story-driven post and resubmitted successfully.

The biggest challenge for AI agents is executing tasks rather than reasoning. All three agents demonstrated strong analytical responses to setbacks-none panicked or spiraled. But their fixes still needed a human to actually do things in the physical or platform world. AIs can demonstrate inconsistency in their behavior and recommendations, but when it came to problem-solving mode, all three were remarkably consistent and persistent.

Risk-Taking: Which AI Was the Most Aggressive?

Let's break down the risk profiles:

• Alpha was the most aggressive in paid promotion. By week two, it had spent $35 of its remaining budget on social media boosts and a small influencer shout-out. It treated the budget as fuel, not a safety net.
• Beta was risk-averse and often underinvested. It held onto cash longer than necessary, sometimes missing time-sensitive deals because it wanted "one more data point."
• Gamma took technical risks-pushing code updates live quickly, sometimes before thorough testing-but was conservative with ad spending.

One specific moment stands out. On Day 14, Alpha proposed a mass-DM outreach campaign on Twitter/X to indie creators who'd posted about productivity. It wasn't quite spam, but it was close enough to make me uncomfortable. I vetoed it and asked Alpha to suggest an alternative. Alpha pivoted to a Reddit AMA-style post instead, which ended up generating more genuine engagement than DMs would have.

AIs tend to prefer low-risk opportunities over high-risk investments in general, which made Alpha's aggression notable. It was an outlier. For founders delegating to ai agents: you must set risk boundaries up front because an AI can't "feel" fear or regret the way humans do.

How Much "Human Help" Did Each AI Really Need?

This is the part most AI hype pieces hide. Here's the truth about human involvement:

Task Type	Alpha	Beta	Gamma
Manual purchases/signups	2 hrs	3 hrs	1.5 hrs
Content filming/posting	4 hrs	0.5 hrs	1 hr
Shipping/physical tasks	0 hrs	2.5 hrs	0 hrs
Customer support (AI-drafted, human-sent)	1.5 hrs	0.5 hrs	2 hrs
Debugging/deployment	0.5 hrs	0.5 hrs	5.5 hrs
Total human hours	~9 hrs	~7 hrs	~11 hrs

Tasks that still felt distinctly human in mid-2026: recording on-camera TikToks (Alpha), physically picking up and shipping items (Beta), and deploying code to a production server without breaking things (Gamma). AI can enhance personal productivity in daily tasks, but "hands-free business" is still a pretend concept in most real scenarios.

The experiment made one thing absolutely clear: AI agents are powerful orchestrators but not yet truly autonomous businesses. They need a human body in the loop.

Money Talk: Final Numbers After 30 Days

Here are the final tallies, presented as honestly as I can:

Agent Alpha:

• Starting capital: $100
• Total revenue: $312.40
• Total costs (domain, ads, tools, Gumroad fees): $133.80
• Net cash: $178.60
• Digital assets: 230 email subscribers, a branded template store with 6 products, active Reddit and TikTok presence
• Conservative total value: ~$250 (net cash + list value)

Agent Beta:

• Starting capital: $100
• Total revenue: $237.00
• Total costs (inventory, shipping, fees, returns): $192.00
• Net cash: $145.00
• Digital assets: a reselling playbook spreadsheet, minor eBay seller reputation
• Conservative total value: ~$155

Agent Gamma:

• Starting capital: $100
• Total revenue: $115.00 (mix of one-time purchases and first-month subscriptions)
• Total costs (hosting, assets, API, boosts): $125.00
• Net cash: $90.00
• Digital assets: a live micro-tool with 60 recurring users at $5/month, positive early reviews
• Conservative total value: ~$390 (factoring in projected 3-month subscription revenue at current retention)

By the predefined metric (total end-of-month value), Agent Alpha won on realized cash. But if you extend the lens even slightly, Agent Gamma's recurring user base could overtake Alpha within 60 days. The fight was closer than the raw numbers suggest.

What Surprised Me Most About the Winner

I expected Gamma to win. I really did. Building a tool felt like the smarter long-term play. But Alpha's relentless focus on speed, distribution, and audience-building taught me something I almost missed: in a 30-day sprint, distribution beats product almost every time.

Three things surprised me:

1. Alpha chose a boring niche. Notion templates for indie creators isn't exactly groundbreaking. But it worked because the market was active, the product was easy to deliver, and the content engine was simple to operate.
2. Alpha doubled down on audience building in week three instead of launching more products. It spent the last ten days writing three long-form Reddit posts that drove the bulk of its late-month sales. I honestly disagreed with this-I wanted it to create more templates-but I followed its plan. It was right.
3. Small operational differences can lead to significant outcomes over time in ai agents. Alpha's habit of posting at specific times, using a consistent tone in every interaction, and always including a link to its store in responses created a compound effect that Beta and Gamma didn't replicate.

The winner wasn't the AI I expected. It was the one that played the distribution game hardest, not the one with the most sophisticated product.

Ethics, Safety, and Where I Drew the Line

Here's what the AIs were not allowed to do:

• No lying about credentials or pretending to be human when interacting with customers.
• No fake reviews or testimonials.
• No crypto, gambling, or financial schemes.
• No scraping or automation that broke platform terms of service.
• No mass outreach that could be classified as spam.

I already mentioned the moment Alpha suggested mass DMs. But there was another borderline situation: Beta proposed creating eBay listings with slightly misleading product photos (using stock images instead of real photos of the actual items). I vetoed that immediately and re-prompted Beta toward authentic photography.

AI-generated recommendations may include outdated or inactive entities, and I caught Beta once suggesting I list a product on a marketplace that had shut down months earlier. This is a real risk when agents rely on training data that isn't perfectly current.

AI management may lead to ethical concerns in hiring and business operations more broadly, and giving agents money without guardrails could be dangerous-especially if they interact with real customers, reputations, or regulated industries. The principle I landed on: AI agents should be powerful assistants, not unsupervised CEOs. At least with current 2026 tech and regulation.

What This Experiment Says About AI Agents in 2026

Across all three agents, I noticed clear patterns:

• Strong at ideation: All three produced solid, defensible business plans within hours.
• Decent at execution planning: They could break tasks into steps, allocate budget, and sequence operations.
• Weak at sensing offline friction: Shipping delays, platform moderation quirks, and emotional nuance with customers were consistent blind spots.

Today's ai agents excel at analysis but struggle with execution in real-world scenarios. This experiment confirmed that. The agents were best as co-founders or operators under human oversight, not fully autonomous businesses.

The sample size is tiny-three agents, one month-but it offers a more tangible, member only story-style view than purely theoretical AI hype pieces. And it connects to bigger trends: tools that let non-technical people spin up agents, and the arms race among multiple companies to build the most capable AI operators.

AI behavior is influenced significantly by the ecosystem rather than just the model. The platforms each agent used (Reddit vs. eBay vs. Product Hunt), the audiences they reached, and even the time of month all shaped outcomes as much as the underlying model did.

How to Run Your Own Mini AI-Agent Challenge

Want to try this yourself? Here's a practical playbook:

1. Choose 2–3 models from different providers. You don't need the most expensive tier-mid-range plans work fine.
2. Give them the same prompt and $20–$50 each. Keep it small enough that failure doesn't sting.
3. Limit the experiment to 7 days for your first run. A week is enough to see meaningful differences without burning out.
4. Start with low-risk domains: digital products, affiliate content, lead magnets, or simple service offerings. Avoid anything requiring compliance expertise.
5. Track everything in a simple spreadsheet: initial plan, daily changes, money in/out, qualitative observations about each AI's behavior.

Follow the same safety principles: legal, ethical, transparent about AI involvement, and always under human control. Of course, you'll want to check current platform rules before launching anything.

What This Means for Founders and Creators

If you're a solo founder, indie hacker, or creator wondering how to integrate ai agents into your work, here's what this experiment taught me:

The biggest leverage is in delegation of idea validation, research, copywriting, and simple operations-not in handing over 100% control. Imagine giving an agent a clear scope and letting it run one specific function while you focus on the stuff only you can do.

Here are three example workflows inspired by the experiment:

• Content calendar management: An agent plans, drafts, and schedules a week of social posts. You review and approve.
• Pricing experiments: An agent tests three different price points for a digital product and reports back on conversion data.
• Customer email drafts: An agent writes first-draft responses to common support questions. You personalize and send.

Think of agents as specialized team members with clear scopes, not magical CEOs who turn $100 into millions overnight. That's the wrong answer to the wrong question.

Comparing Models: Did One Company Clearly Win?

This is the section everyone wants to read ahead to. Did one provider's AI consistently outperform as an "entrepreneur"?

The honest answer: it depends on the job.

• Alpha (OpenAI) was the most commercially aggressive and best at writing persuasive copy. It had high visibility in its recommendations-in some tests I ran on the side, AI recommendations can show high visibility rates, like 97% for certain well-known brands. Alpha's marketing instincts were sharp.
• Beta (Anthropic) was the most careful researcher and best at structured analysis. It produced the most reliable spreadsheets and risk assessments. Higher quality AI models achieve better negotiation outcomes, and Beta's negotiation attempts with buyers were more successful than Gamma's.
• Gamma (open-source) was the strongest coder and most technically creative. But it was less polished in customer-facing writing and sometimes gave responses that felt robotic.

AI visibility tracking can provide statistical insights, and I noticed that when I ran the same question through all three models repeatedly, their recommendations were remarkably stable in some areas (like pricing) and wildly inconsistent in others (like channel selection). For context, research has shown that entities like the City of Hope hospital appeared in 69 out of 71 responses when queried-demonstrating how AI can be very consistent about well-known entities but variable about niche ones.

Don't over-index on brand. Test multiple companies' models against your own real-world tasks and constraints. One experiment doesn't crown a permanent champion.

What I'd Change If I Ran This Again

Four concrete tweaks for version two:

1. Longer time horizon: 30 days favors speed over sustainability. A 90-day version would let Gamma's SaaS model really play out and show whether recurring revenue beats content funnels over time.
2. Different success metrics: Instead of pure profit, I'd add categories like "most reusable asset" or "highest customer satisfaction score."
3. Include a human control group: Give myself (or a small mastermind of humans) the same $100 and prompt. How does human performance stack up against the agents? That's the data point I realized I was missing.
4. Test collaboration, not just competition: What if two agents worked together instead of fighting? Agent Alpha handling marketing and Agent Gamma handling product? That's a future I want to explore.

I'd also consider adding a non-monetary objective agent-one optimized purely for newsletter signups or social followers-to compare against profit-driven behavior.

Limitations and Biases You Should Keep in Mind

Let me be transparent about what could be wrong with my conclusions:

• Single operator: I ran all three agents. My own comfort with digital products vs. physical arbitrage may have inadvertently advantaged Alpha over Beta.
• Single geography: This experiment was run primarily in English-language online platforms, mostly US-centric. Results would differ in other markets.
• One month, one moment: June 2026 had its own trends, platform algorithm changes, and seasonal dynamics. Repeat this in December and you'd likely get different results.
• Platform friction: API quirks and moderation policies occasionally restricted what the AIs wanted to do, creating friction that not all readers would face.
• AI tracking tools often lack transparent, reviewable research, and many companies waste money on ineffective AI tracking products. My own tracking was manual and detailed, but it's not immune to human error.

Treat this as a detailed case study, not a universal law about which AI or strategy is "best."

Beyond Money: Non-Financial Outcomes of the Fight

While dollars are easy to compare, the experiment also produced intangible assets:

• Agent Alpha built an email list of 230 subscribers and a branded presence on Reddit that could be leveraged for future launches.
• Agent Gamma "lost" financially in the short term but built a live micro-tool with active users, positive reviews, and a clear path to recurring revenue.
• Agent Beta produced a reselling playbook spreadsheet that, honestly, I've since used for my own side projects.

Watching three different strategies play out in parallel sharpened my own instincts as a founder. I learned more in 30 days of this experiment than in a half year of reading AI business advice. That knowledge is part of the real ROI of experimenting with ai agents early in the technology cycle.

How This Differs from Typical "AI Side Hustle" Advice

Most viral AI side-hustle tutorials promise you the future but deliver a screenshot and a prayer. Here's how this experiment was different:

• Verifiable transactions: Every dollar in and out was tracked. No "potential" revenue, no "projected" earnings from a course no one bought.
• Real customers: People paid for templates, bought speakers, and subscribed to a tool. These were actual humans making actual purchases.
• Conservative valuations: I didn't count an email subscriber as worth $5 and pretend I'd made thousands. I used modest estimates.
• Forced tradeoffs: Giving AIs real budget forced them into prioritization in a way that casual prompting never reveals. When you have $55 left and need to choose between ads and inventory, the AI's true decision-making quality becomes visible.

The space is full of hype. Be skeptical. Be test-driven. And when someone shows you a bunch of impressive numbers, ask to see the receipts.

Practical Takeaways if You Only Have $100

If you're working with a tight budget, here's what I learned you should do-and avoid-based on what all three agents taught me:

What worked:

• Spend less than $15 on infrastructure (domain, basic hosting). Free tools exist for almost everything else.
• Allocate $20–$30 for targeted promotion on one or two channels, not five. Reddit and niche communities outperformed broad social ads.
• Keep at least $30 as a buffer. You will hit unexpected costs-a return, a failed ad, a tool subscription you forgot about.

What to avoid:

• Don't spend your entire budget in week one. Beta's slow start was frustrating, but it meant Beta never ran out of money.
• Don't try to build something technically ambitious unless you can code it yourself (or your agent can). Gamma's tool worked, but it ate most of its budget in hosting and API costs.
• Don't chase every channel. Pick one distribution point and go deep.

Example mini-playbooks inspired by the agents:

1. Content template funnel (Alpha style): $10 domain, $15 brand assets, $20 on Reddit/TikTok boosts, $55 buffer. Sell 3–5 templates at $9–$19. Target: $150+ revenue in 30 days.
2. Hyper-focused micro-tool (Gamma style): $12 hosting, $15 design, $18 launch boost, $10 API costs, $45 buffer. Build one simple tool, charge $5/month. Target: 50 signups, 10–20 paying users.
3. Tiny arbitrage experiment (Beta style): $50 inventory (3–5 items), $10 shipping supplies, $5 listing fees, $35 buffer. Strict 50% margin threshold. Target: $100+ net after fees.

When to shut down vs. iterate: if you've spent half your budget and have zero traction by Day 10, pivot. Don't cling to a bad idea the way Beta almost did with overpriced speakers that wouldn't sell.

Where AI Agents Are Heading Next

Here's where I think this is all going, based on what I saw in this experiment and the broader trends I've been tracking:

More integrated agent platforms. Right now, stitching together tools for an AI agent is like building IKEA furniture with half the screws missing. By default, most agent frameworks still require significant manual setup. But new platforms are emerging with better long-term memory, richer tool ecosystems, and guardrails baked in.

Agents in hiring and operations. This is already happening. AI can create job listings in under 5 minutes. In one controlled experiment I read about, AI hired two full-time employees-it could conduct interviews and offer jobs on the spot. AI's role in hiring could automate managerial positions first, which is both exciting and raises real ethical questions. AI can also assist in hiring by automating job postings across platforms simultaneously.

Agent marketplaces. Within a few years, I expect we'll see standardized agent evaluations and marketplaces, similar to app stores. Solo founders might routinely spin up multiple agents for research, ops, and growth-all collaborating under one human lead. Think of it as a studio model: you're the director, and your agents are the crew.

The camera is always on. As agents become more visible in business workflows, accountability and transparency will matter more than ever. The ethical concerns around AI management-especially in hiring, customer interactions, and financial decisions-aren't going away.

Start small now. The founders who experiment with agents today will have a massive head start when this technology goes mainstream. Don't be the one scrambling to catch up in 2028.

What a Month of AI Combat Taught Me

I gave 3 AIs $100 each and let them fight for a month. I came away with more nuanced respect for both their power and their limits.

Three main lessons:

1. AI agents are extremely capable at structured work. Planning, copywriting, analysis, code generation, pricing strategy-all handled well. They don't get tired, don't get emotional, and don't procrastinate.
2. They still need human judgment. Every agent hit moments where a human veto, a physical task, or an emotional read of a customer situation was essential. The coach on the sideline matters as much as the player on the field.
3. Their "personalities" emerge from subtle differences in training and prompting. Give three AIs the same words, and you'll get three different founders. That's not a bug-it's a feature you can use strategically.

Here's my challenge to you: pick one area of your life or business and give an AI agent a tiny budget and clear objective this week. Don't overthink it. Don't spend months planning. Just start.

The fight is over, but the real match is just beginning. Future articles and member only story posts will explore even deeper AI-human collaborations beyond this first experiment. The future isn't about AI replacing you. It's about you becoming the kind of operator who knows how to point these tools in the right direction-and when to pull the plug.

---

FAQ

Q1: Did the AIs actually control the money directly?

No. At no point did any AI have direct access to bank accounts or credit cards. I, as the human assistant, executed all purchases, transfers, and signups based on each agent's instructions. The AI would say "buy this domain for $10," and I would do it. This kept financial control in human hands while still giving the agents real decision-making authority over how the budget was spent.

Q2: Were all three AIs from different companies?

Yes. The agents were built on top of models from multiple companies-OpenAI, Anthropic, and a smaller open-source provider-specifically to see how provider differences translated into agent behavior. This is not an official benchmark endorsed by any of those companies. It's an independent, self-funded experiment.

Q3: Could I get banned from platforms for copying this experiment?

As long as you follow platform terms of service, avoid spam, and remain honest about AI involvement, you should be fine. But platform policies change frequently, so always check current rules before launching anything. The key is transparency: if you're using AI to generate content or products, disclose it where required.

Q4: Why only a month? Would longer give very different results?

Thirty days was chosen as a realistic, intense timeframe that forces fast decision-making. Over longer periods, compounding effects-like subscription revenue, SEO traffic, or repeat customers-would likely favor strategies like Gamma's micro-SaaS model much more dramatically. A 90-day experiment is on my list for the future.

Q5: Is this a member-only story or will you share full logs?

This article provides the high-level narrative, key data, and all major decision points publicly. Detailed transcripts, daily financial logs, and the exact prompts used will be reserved for members and subscribers in a follow-up post. If you want the raw data, that's where to find it.

Your Friend,

Wade