If you run a business alone, AI vendors have a stack for you: a writing tool, a research tool, a meeting tool, a support bot, a social scheduler, and an “agent” to coordinate them. Two hundred dollars a month, six logins, and a part-time job just keeping up with which tool got which feature.
Here’s the contrarian, cheaper position this guide defends: a solo operation runs on one general assistant, used deeply, plus at most two or three point tools earned by named bottlenecks. Everything else is subscription theater.
The center of the stack: one general assistant
The flagship assistants — the ChatGPT, Claude, and Gemini tier of products — are general instruments. The same subscription drafts your proposals, summarizes a contractor agreement, debugs your spreadsheet, roughs out a marketing plan, rewrites a difficult email, and explains an unfamiliar regulation. On a paid individual plan, that’s roughly the cost of one business lunch a month.
The argument for depth over breadth is practical, not aesthetic:
- Skill transfers. Every hour you spend getting good at one assistant — what context to paste, how to constrain output, where it fails — improves all your AI-assisted work. An hour learning a niche tool improves one workflow, and evaporates when the tool pivots or dies.
- Context accumulates. Memory and project features mean your assistant gradually holds your business: your voice, your services, your standard caveats. Splitting work across five tools means five shallow copies of your context, none good.
- One bill, one data policy to read, one thing to cancel. Administrative overhead is a real cost when the administrator is also the product, sales, and support department.
Which flagship? They leapfrog each other every few months, and for solo workloads the differences are smaller than the marketing implies. Pick on workflow fit — try your three most common tasks on each free tier for a week — then commit and stop watching the horse race. Switching later is cheap precisely because your assets (prompts, templates, habits) are portable text.
Making the general assistant earn its keep
The stack works only if the center is actually used deeply. Concretely, that looks like:
- Standing prompts for recurring work. Your proposal draft, your invoice chase, your weekly review — each as a saved template with your voice example baked in (see the prompting guide).
- Paste-first habits. The assistant reads your real material — the client email, the spec, the numbers — instead of answering from generalities.
- A verification habit matched to stakes. Drafts and brainstorms ship after a skim; anything with numbers, names, or legal weight gets checked against a source. (The limits guide is the map of where checking matters most.)
A solo operator with twenty saved workflows on one assistant has a more capable “stack” than most five-tool setups — and it fits in one browser tab.
The point tools: earned, not collected
A point tool enters the stack only when you can name the bottleneck it removes, in a sentence with a number in it. “Transcribing client calls eats three hours a week” earns a transcription tool. “Everyone’s talking about agents” earns nothing.
Categories that most often genuinely earn a slot for one-person businesses:
- Meeting transcription and notes — if your week is call-heavy, this is usually the first justified add-on. The general assistant then summarizes and extracts action items from the transcript, so the point tool stays small.
- Accounting with AI categorization — your bookkeeping software’s built-in categorization beats pasting transactions into a chatbot, because it sits where the data already lives. This is a feature you probably already pay for; turn it on before buying anything new.
- Email triage inside your mail client — same logic: AI features attached to the tool that owns the data beat a separate subscription that needs the data exported to it.
Notice the pattern in all three: prefer AI features inside tools you already pay for over new standalone tools. The data is already there, the login is already there, and the marginal cost is often zero. The standalone purchase needs to clear a much higher bar.
Where paying more buys nothing
Some spending patterns reliably fail for solo operators, and they’re worth naming bluntly:
- The top-tier “pro/max” plans of the flagship assistants mostly sell higher usage limits and earliest access. If you’re not hitting limits, the extra tier buys you nothing your work will notice. Upgrade when you’re throttled, not when you’re excited. (Heavy coding or research automation can be the exception — but then you’ll see the limit, which is the point.)
- Multiple overlapping general assistants on paid plans. Keeping a second flagship on the free tier as an occasional second opinion: sensible. Paying for two: paying twice for one capability.
- Niche “AI for X” wrappers whose entire product is a prompt template over the same models you already have. The test from the tool-choosing guide applies: if you can replicate the demo with one saved prompt in your assistant, you just did.
- Automation platforms ahead of volume. Agent-style automation has real uses, but a workflow you do four times a week is almost always cheaper as a saved prompt plus two minutes of your judgment than as a brittle automated chain you now maintain.
Data hygiene for a stack of one
Solo operators skip the data-governance conversation because there’s no committee to have it with — which is how customer data ends up pasted into tools that train on it. The one-person version takes ten minutes and holds for years:
- Set the training toggle once. Every major assistant has a setting controlling whether your conversations train future models. On a paid business-grade plan, opt out (or confirm the default), then stop worrying about routine pastes.
- Sort your data into two piles, not ten. Pile one: material you’d be comfortable in a vendor breach notification — drafts, public info, your own writing. Paste freely. Pile two: client confidences, other people’s personal data, credentials, anything under NDA. That pile gets anonymized before pasting (“Client A, a dental practice…”) or stays out entirely.
- Credentials never, in any tool. Passwords, API keys, full card or account numbers — no AI tool needs them, and several breach stories started exactly there.
That’s the whole policy. It’s two rules and a toggle, which is precisely why it gets followed.
A reference shape, in dollars
Prices move; the shape is the durable part. A defensible solo stack in 2026 looks like:
| layer | what | rough monthly cost |
|---|---|---|
| Center | One flagship assistant, paid individual plan | ~$20 |
| Earned point tool | Often transcription, if call-heavy | $0–20 |
| Existing-tool AI | Features inside your accounting/email/docs | usually $0 added |
| Total | $20–50 |
If your stack costs materially more than that, every extra line should trace to a bottleneck with a number on it. If it costs less — the free tiers are genuinely capable now — that’s not falling behind; that’s matching spend to need.
Two honest caveats. First, none of this guarantees outcomes: a stack is leverage, and leverage multiplies the quality of the work it’s applied to. Second, specific products and prices here will age; the principles — one deep center, point tools earned by named bottlenecks, features-in-place over new subscriptions — are the part built to last, and the daily briefings on this site exist to flag when the ground under them actually moves.