Guides
The Minimum AI Stack for a One-Person Business
5 min read
Most solo operators have twelve AI tools and use two.
That’s not a discipline problem. It’s a design problem: every tool was added to solve a problem that existed for about a week, and nothing was ever removed. The stack grew by accretion, which is how most stacks grow, and now it costs $180 a month and produces a low-grade guilt every time you see the billing page.
The stack that actually works for a one-person business is smaller than you think. Three layers, one tool each. Everything else is optional and most of it is noise.
The three layers
Layer 1: one model you know well
Pick one. ChatGPT, Claude, Gemini — it genuinely doesn’t matter much at this point. What matters is that you use it enough to know its habits: where it over-explains, where it hallucinates, which phrasings you have to ban in every prompt.
The cost of switching is not the subscription. It’s the accumulated knowledge of how to get good output from this one. People who switch every quarter never develop that, which is why they’re always disappointed.
What to do: commit to one for six months. Learn its failure modes. Build a saved context block you paste into every conversation.
Layer 2: one automation layer
This is where the actual leverage is, and it’s the layer most solo operators skip because it sounds technical.
An automation layer connects things: order comes in, file gets sent; form gets filled, record gets created; week ends, numbers get summarised. You build it once and it runs.
Three options, in order of how much they’ll cost you in time:
- Your platform’s built-in automations. Most ecommerce and form tools have them. Free, limited, and enough for 80% of what a solo operator needs. Start here.
- A no-code connector (Zapier, Make, n8n). Pay per task or self-host. The right choice once you have more than five workflows.
- Custom code. Only when the first two genuinely can’t do it, which is rarer than developers suggest.
What to do: build one automation this week. Not the most valuable one — the one whose failure would be least embarrassing. See if it survives two weeks. If it does, build the second.
Layer 3: one place things live
Not a second brain. Not a knowledge system. One place where your prompts, your drafts, your client notes and your numbers live, so you can find them without thinking.
The failure mode here is elaborate: people build a Notion workspace with eleven databases, spend three weekends on it, and never open it again. The system becomes the work.
What to do: pick the dumbest option that works. A folder of text files. A single Google Doc. One Notion page with headers. The only requirement is that adding something takes under ten seconds.
What doesn’t belong in the stack
Four categories that consistently fail for one-person businesses:
AI tools that need a workflow to be useful. If a tool only works when you feed it a steady stream of inputs you don’t have, it will sit unused. The tool isn’t bad; the prerequisite is.
Anything with a dashboard you have to check. A tool whose value depends on you logging in is a tool that will be abandoned in month two. Value must arrive without you.
Second copies of layer 1. Two general models, or a model plus a “writing AI” plus a “research AI.” You will use one. The others will make you slower because you’ll spend time deciding.
Tools bought for a future version of the business. The one-person business at $5k/month doesn’t need the stack for $50k/month. Buying ahead feels productive and is mostly avoidance.
The cost math nobody does
Add up your subscriptions. Then add the time.
If you spend twenty minutes a week deciding which tool to use, that’s seventeen hours a year. If you spend an hour a month maintaining an automation that saves you thirty minutes a month, you’re down six hours a year.
Most stacks are net-negative on time and nobody notices, because the cost is distributed across small moments and the benefit is imagined.
The test: for each tool, name the specific task it does that you would otherwise do manually, and how many minutes per week it saves. If you can’t answer in one sentence, the tool is decoration.
What the three layers actually produce
A concrete example, because abstract stack advice is useless.
A solo operator selling digital products with the three layers in place:
- Layer 1 writes the product descriptions, the emails, the support replies — with a saved context block so the voice is consistent.
- Layer 2 sends the files, chases the invoices, logs the refunds, and puts the weekly numbers in one message.
- Layer 3 holds the prompt library, the customer questions, and the numbers, in a place that takes ten seconds to search.
Total monthly cost: one subscription plus a connector. Total setup time: about six hours, spread over three weeks. Total ongoing maintenance: under an hour a month.
That’s the whole stack. It doesn’t feel impressive, which is the point — impressive stacks are the ones people build instead of doing the work.
When to add a fourth layer
Only when a specific, recurring task is failing consistently and you can name the failure. “I keep forgetting to follow up on quotes and it’s costing me deals” is a reason to add something. “I feel like I should be using more AI” is not.
The signal that you need more stack is a bottleneck you can point at. The signal that you have too much is a subscription you can’t explain.
The uncomfortable part
The stack is not the constraint. Almost never.
For a one-person business under $20k a month, the constraint is almost always one of three things: not enough people know you exist, your offer isn’t clear, or you don’t ask for the sale. None of those are solved by a tool.
Building the stack feels like progress because it’s concrete and completable. Talking to twenty potential customers is neither. But it’s the thing that moves the number.
So: build the three layers, spend six hours on it, then stop. The stack should be boring. If it’s interesting, you’re doing it wrong.
If you want the pre-built version
The three systems in the Complete Stack map onto the three layers: 80 prompts for layer 1, a 90-day content engine for the layer that produces demand, and 20 automation blueprints for layer 2 — each with build order, realistic build time and failure modes, so you’re not discovering them at 2am.
But the six-hour version above works without buying anything. Start there. If you find yourself hitting the same three walls repeatedly, that’s when a pre-built system saves time instead of adding a subscription.