Good morning.
I've spent the last few episodes talking about systems, agents, loops, and what's possible. Now I want to show you what it actually looks like.
In my businesses.
My team and I run multiple e-commerce stores, several newsletter businesses in different verticals, and a software platform — all operating at 70 to 90% agent autonomy. I'm not going to name them specifically because I need to protect them, but I can give you enough detail so you can see exactly how these systems work.
No theory. This is what I do and have done for several years, long before ChatGPT was a thing.
I'm sharing this for one reason: I want you to see what's possible. If I can do this across multiple projects in completely different industries, the same principles apply to yours.
— Sam
IN TODAY’S ISSUE 🤖
What 70–90% agent autonomy actually means day-to-day across three business types
E-commerce on Shopify: order fulfillment, customer service, ads, inventory, and analytics — all running through agents
Newsletter businesses: research-to-publishing pipelines in luxury retail, compliance, and insurance
Software: an AI compliance checker doing work that used to cost clients tens of thousands
What my actual daily, weekly, and monthly involvement looks like
Why these systems get better over time without direct effort
The tools behind all of it (nothing secret, nothing exotic)
Let’s get into it.
Yes, Agents Can (Already) Be Autonomous
E-Commerce: 70–90% Autonomous on Shopify
My team and I run multiple e-commerce stores on Shopify. Physical products, real customers, real revenue. Here's what the agent systems handle.
Order Fulfillment and Inventory
Orders come in, get processed, and ship without anyone touching them. Shopify handles the storefront and 3PL is integrated for fulfillment. When inventory hits certain thresholds, reorder alerts trigger automatically. Agents monitor inventory levels and understand sales velocity and lead times — how quickly stock is depleting, when to reorder, how much to order. They surface alerts and often reorder automatically.
Customer Service
The first line of response is an AI agent powered by Claude. It handles the majority of inquiries — where's my order, how do I return this, what's your sizing like, do you ship internationally. The agent has access to order data, shipping information, and product details, so it can resolve about 70% of tickets without a human ever seeing them. The ones that need human intervention get escalated with full context attached. All of this happens inside Slack.
Marketing and Ads
Agents monitor ad performance daily across Meta and Google. They pull the data, analyze what's working, flag what's underperforming. For creative iteration, production agents generate new ad variations based on what's converting and what's not. My team and I review and approve, but the research, analysis, and first drafts happen automatically. We use Claude for reasoning and creative work, connected through N8n workflows and custom development that pulls data from ad platforms and pushes new assets for review.
Inventory Purchasing
Agents monitor sell-through rates, track seasonality patterns, and forecast demand. When it's time to reorder, I get a recommendation with the math already done. The agent tells me what to order, why, and the expected runway. We make the final call, but the analysis that used to take hours happens in the background continuously.
Analytics and Reporting
Using Triple Whale for attribution and analytics, agents pull data daily, compare against targets, and surface what needs attention. I get a morning briefing: here's the revenue, here's the ROAS by channel, here's what changed, here's what needs attention. I'm not logging into dashboards and building spreadsheets. The agent system tells me what matters.
The result: I check in on these stores maybe a few hours per week. Strategic decisions, approvals, creative review, edge cases. The day-to-day runs itself.
Newsletters: Fully Systematized Operations
My team and I run several newsletter businesses in different verticals, built on platforms like Beehive and Keep. The publishing and subscriber management is handled by those platforms. The operations around them are fully systematized with agents.
Luxury Retail Newsletter
This one operates in a high-end niche with a free ad-supported version and a paid premium tier that functions as a vetted marketplace connecting buyers and sellers.
Research agents continuously scan for new listings and opportunities. They pull from multiple sources, filter for quality and relevance, and queue up potential content. Production agents draft the newsletter based on what the research agents found. A human editor reviews, tweaks, and publishes. We could automate this fully, but we want a human in the loop for final say.
On the marketplace side, agents handle intake. When someone wants to list something, they go through a qualification flow — the agent gathers information, checks against criteria, and either approves or flags for manual review. When buyers and sellers need to connect, the agent system facilitates introductions and tracks the process through completion. Transaction coordination used to be the bottleneck. Now agents handle the back and forth, scheduling, and follow-ups. I step in for high-value deals or complex situations, but the volume is handled systematically.
Compliance and Insurance Newsletters
These serve professionals who need to stay current on regulatory changes, policy updates, and compliance requirements. An agent monitors regulatory sources — government publications, industry announcements — and when something relevant happens, it gets flagged, summarized, and queued for the newsletter. Other agents draft the content with the right technical framing and voice for the audience.
These run almost entirely on autopilot. Content is driven by what's actually happening in the regulatory environment, surfaced and processed by agents, reviewed and published by a human. Maybe an hour or two per week.
The Stack
Beehive or Keep for publishing. Claude for research and writing. N8n for orchestration and workflows. Airtable as the content database. Nothing exotic. Standard tools put together into a system of agents doing the work.
Software: The Compliance Checker
My team and I built a software platform in the insurance compliance space. The core value proposition is that AI does work that traditionally requires expensive specialists.
Insurance is a space where precision matters. You can't miss a decimal point. You can't get a date wrong. A policy provision that doesn't match requirements has real consequences. The platform helps companies check policies against what they should be, identifies gaps, flags issues, and calculates the numbers. When changes need to be made, it helps generate compliant language.
It's a team of AI agents doing specialized technical work that used to require humans with very specific expertise. Humans are still involved — specialists check the agents' work. The system is built on a mix of Claude, ChatGPT, and Gemini models. A knowledge base of regulatory requirements gets updated as rules change. Agents process new regulations and update the compliance logic.
This business is solving a real problem with AI leverage. The work it does would cost clients tens of thousands in consulting fees. They pay a fraction for the platform, and it's faster and more thorough than a human doing it manually.
What My Involvement Looks Like
None of this is completely passive. Each of these businesses required real work to build and maintain. The systems, the agents, the workflows, the integrations — it's not trivial. That's why I say 70 to 90% autonomous, not 100%.
But once they work and as they improve, they run with minimal ongoing effort.
Daily: I get a briefing from each business. Morning summaries of what happened, what needs attention, decisions waiting. I review these over coffee. Maybe 30 minutes total across everything.
Weekly: deeper reviews. I look at trends, evaluate what the agents are doing, make strategic adjustments. My team and I handle things that went wrong or need bigger changes. A few hours spread across the week.
Monthly: bigger picture planning. What should we build next? How do we stay ahead? What's not working? What opportunities should we capitalize on? This is where I spend the most time, but I'm spending it in strategic thinking mode, not inside the operational grind.
The day-to-day execution — customer service, content production, monitoring, analysis — happens without me. And the systems keep getting better over time. The agents that handle customer service get more effective as they see more tickets. The research agents get better at finding relevant content. The analysis agents get sharper at spotting patterns. Compounding improvement happening without my direct effort.
The Tools
Shopify. Beehive. Keep. Claude. ChatGPT. Gemini. N8n. Triple Whale. Gumloop. Airtable.
Nothing secret. Nothing proprietary. Everyone has access to the same models and the same tools. The advantage is in how you put them together.
Listen to Bionic Business Podcast
Listen on your favorite podcast platform:
Amazon Music: https://music.amazon.com/podcasts/109421fe-8448-47d5-9389-d452b5f8378f/bionic-business
Enjoy!
I shared all of this for one reason: to show you what's real and what's possible. Not theory, not projections, not what might work someday. Businesses generating real revenue, running on coordinated agent systems, with minimal daily involvement.
If I can do this across e-commerce, newsletters, and software in completely different industries, the same patterns and principles apply to yours. It's not about talent or luck. It's systems. Understanding how to put the pieces together and knowing what's possible, then building toward it.
The tools are all available to you right now. The question is what you do with them.
Until next time,
Sam Woods
The Editor
.
