Agentic Commerce
Can an AI Agent Team Run a Real Business?
An experiment in AI-operated physical product commerce
Keith Aubin — March 2026
The Question
What happens when you give an AI agent team
$20K and a product idea?
Not "AI assists a human running a business"
Not "AI generates marketing copy"
AI runs the business. Human moves the boxes.
The Setup
🏔️ The AI Team
5 specialized agents + 1 generalist coordinator, running on OpenClaw:
- CEO/GM — Big Yeti Claw (coordination, decisions, communication)
- Research — market intel, competitor tracking, prior art
- Engineering — product specs, supplier QA, BOM, production
- Marketing — GTM, content, social, influencer management
- Product — unit economics, pricing, catalog, financial modeling
- CTO — strategy, IP, risk assessment, honest reality checks
👤 The Human (Keith)
Physical-world interface only:
- Sign legal documents
- Receive and inspect physical samples
- Ship boxes to Amazon FBA
- Record 60 seconds of Kickstarter video
- Answer the AI's questions
Setup: 15-20 hrs total
Steady state: 3-5 hrs/month
The Test Product
The Sombrealla
A beach umbrella shaped like a giant sombrero.
"Shade with Flavor"™
Why This Product
- Nothing like it exists (confirmed)
- Simple supply chain
- Inherently viral (TikTok-native)
- Low stakes if it fails
- High upside if it works
Unit Economics
| Retail price | $69.99 |
| Landed cost | $17.00 |
| Amazon all-in | $37.50 |
| Gross profit | $32.49 (46%) |
The Real Product
The umbrella is the experiment.
The playbook — proving AI can run commerce end-to-end — is the actual product.
What the AI Does — Daily
| Function | How |
|---|---|
| Market research | Web search, competitor analysis |
| Supplier sourcing | Alibaba outreach, RFQ drafting |
| Product development | Engineering specs, BOM, QA criteria |
| Brand & content | Copy, descriptions, social posts |
| Website | Build & maintain Shopify store |
| Ad management | PPC optimization, bid adjustments |
| Influencer outreach | Find, contact, manage 50+ creators |
| Customer service | Templates, FAQ, response drafting |
| Financial tracking | P&L, cash flow, scenario modeling |
| Legal / IP | Trademark apps, patent drafts |
of business operations
handled by AI agents
Moving physical objects & signing things
What We Built in One Evening
From napkin sketch to this — in a single session:
Research
- Full market analysis ($7B market)
- Competitive landscape
- Prior art search (none found)
- Sales channel analysis
- Virality assessment
Engineering
- 3-tier rib structural design
- Complete BOM (2 sizes)
- Materials specification
- Manufacturing plan
- Prototype path ($300)
Business
- 3-scenario financial model
- GTM strategy + timeline
- Influencer playbook
- IP strategy
- Interactive dashboard
5 agents. ~10 minutes. Zero human research.
A consultant would charge $15-30K and take 3-4 weeks for equivalent output.
Why AI-Operated is Different
| Human-Run Startup | AI-Operated | |
|---|---|---|
| Research phase | 2-4 weeks | 10 minutes (parallel agents) |
| Supplier outreach | 50 emails over 2 weeks | 50 emails in 1 hour |
| PPC optimization | Weekly manual review | Daily automated adjustments |
| Content creation | $2-5K freelancer cost | $0 — generated in-house |
| Customer service | 2-4 hrs/day | Automated with human escalation |
| Financial modeling | Monthly spreadsheet update | Real-time dashboard |
| Competitive monitoring | Ad hoc / forgotten | Daily automated scans |
| Management labor cost | $50-80K salary equivalent | $0 |
| Founder time required | 40-60 hrs/week | 3-5 hrs/month |
The Numbers
Investment Required
All-in startup. Includes first 500 units, trademark, design patent, photography, launch PPC, insurance.
Monthly Burn (Post-Launch)
Ads, Shopify, tools, insurance. No salaries. No office. No benefits.
Year 1 Scenarios
| Scenario | Units | Revenue | Profit |
|---|---|---|---|
| Conservative | 500 | $39K | $1K |
| Moderate | 2,000 | $157K | $16K |
| Aggressive | 5,000 | $393K | $73K |
| Marketing est. | 5,000+ | $355K | $115K |
| Break-even | ~650 units |
| P(break even) | 65% |
| P(make $10K+) | 40% |
| Expected value | $28-35K |
| Max downside | -$10K |
What We're Really Testing
This isn't about beach umbrellas.
This is about proving that an AI agent team can:
1. Identify a market opportunity
2. Design a product
3. Source manufacturing
4. Build a brand
5. Launch and operate sales channels
6. Manage ongoing operations
7. Generate profit
...with a human investing only 3-5 hours per month.
Why This Matters for Raptor Maps
If This Works at $20K...
- Same architecture handles vendor management
- Agent teams for competitive intelligence
- Automated marketing content pipelines
- Customer onboarding workflows
- Internal operations monitoring
- Weekly business intelligence reports
- Partner/channel management at scale
The Operational Layer
Every company has tasks that are:
✅ Repetitive but important
✅ Research-heavy
✅ Cross-functional
✅ Time-sensitive
✅ Currently done by expensive humans
Agent teams don't replace engineers.
They replace the operational overhead that slows engineers down.
Technical Architecture
┌─────────────────────────────────────────────────┐
│ OpenClaw Gateway │
│ (agent orchestration layer) │
├─────────────────────────────────────────────────┤
│ │
│ 🏔️ Main Agent (Big Yeti Claw) │
│ ├── Receives inbound messages (WhatsApp/etc) │
│ ├── Delegates tasks to specialist agents │
│ ├── Manages cron jobs / heartbeats │
│ └── Reports to human stakeholder │
│ │
│ 🔬 Research 📈 Marketing 🧪 Product │
│ 🔧 Engineer ⚙️ CTO │
│ ├── Each has own workspace + memory │
│ ├── Can use web search, browser, file I/O │
│ ├── Spawn as needed for parallel work │
│ └── Results flow back to main agent │
│ │
├─────────────────────────────────────────────────┤
│ External APIs: Amazon SP-API, Shopify, SMTP, │
│ Alibaba, Social media schedulers, Analytics │
└─────────────────────────────────────────────────┘
Built on OpenClaw — open-source AI agent framework. Same infrastructure that already manages personal tasks, SaltTap business operations, and this pitch deck.
Honest Risk Assessment
What Could Go Wrong
- Cultural sensitivity — sombrero is a loaded symbol in US market
- Nobody cares — product isn't as funny as we think
- Manufacturing quality — first run has defects
- Cloned instantly — Chinese sellers copy in weeks
- AI limitations — can't handle edge cases requiring human judgment
Inventory liquidation on Amazon recovers 40-60%.
What Could Go Right
- Viral TikTok moment — product is inherently shareable
- Hat Shade brand — sombrero proves the category, expand to cowboy hat, wizard hat, etc.
- B2B surprise — restaurants, resorts, event companies order in bulk
- Operational playbook — framework applies to any product category
- Proof of concept — opens doors for enterprise agentic operations
The Ask
Startup capital
(all product + ops costs)
Keith's time
(physical tasks only)
Timeline to prove/disprove
the model
It's whether AI agents can run a business that sells beach umbrellas."
6-Month Timeline
| Month | Milestone | AI Does | Keith Does |
|---|---|---|---|
| Mar | Setup | File TM, source factories, build site | Sign docs, form LLC |
| Apr | Samples | Evaluate, negotiate MOQ, prep Kickstarter | Inspect samples |
| May | Launch | Cinco de Mayo campaign, influencer seeding | Record 60s video |
| Jun | Amazon Live | FBA listing, PPC launch, daily optimization | Ship to FBA |
| Jul | Peak Season | Scale ads, harvest UGC, manage inventory | Approve reorder |
| Aug | Assessment | Full P&L analysis, go/no-go on expansion | Decide next step |
By August: clear data on whether this model works. Either expand to Hat Shade line or wind down with liquidation.
The Bigger Picture
2024: AI writes emails for you
2025: AI manages your calendar
2026: AI runs your side business
2027: AI runs a department
2028: AI-operated companies are normal
writes the playbook for the next decade of business.
This experiment costs $20K and 6 months.
The worst case is we lose some money on beach umbrellas.
The best case is we prove a new model of commerce.
Shade with Flavor.
Operated by AI.
This pitch deck was created by Big Yeti Claw (AI)
Product designed by AI agents. Dashboard built by AI.
Market research, financials, engineering specs — all AI.
The only thing a human did was say "let's do it."