One person, one laptop, and a stack of LLM subscriptions is a viable business in 2026. I've watched solo operators clear $15k-$40k a month doing exactly this. The playbook isn't complicated, but it is specific. Here's how I'd set it up from scratch: the offer, the stack, the pricing, the delivery rhythm, and the traps.
Pick one painful, repetitive problem
Every durable one-person AI business I know is built on a single, narrow, boring pain: first drafts of proposals, review of vendor contracts, weekly competitor monitoring, podcast repurposing, intake triage for a specific trade. Narrow wins because you get faster with reps, your referrals get sharper, and your marketing writes itself — you literally tell people what you do in one sentence.
If you can't describe your offer in one sentence to a 55-year-old business owner, it's not an offer yet. "I turn your voice memos into published newsletters in 48 hours" is an offer. "I do AI implementation" is a shrug.
Pick something where the client's current cost is visible and painful: hours burned, deals lost, tickets backlogged. Invisible pain doesn't sell.
Build a boring, reliable stack
Solo operators live or die by reliability. My suggested starting stack, all of which you can run lean:
- Models: Claude and ChatGPT for client-facing drafting; an open-source model (Llama, DeepSeek, Qwen) via Ollama or a cheap API for high-volume, low-stakes steps.
- Automation: n8n self-hosted (free, your data stays put) or Zapier/Make if you want hosted convenience. n8n wins on cost once you have more than a couple of workflows.
- Editor: Cursor for anything code-adjacent; it pays for itself the first time it scaffolds a client script.
- Knowledge base: A simple RAG over the client's docs — or honestly, a carefully maintained system prompt plus uploaded files in Claude Projects covers 80% of cases at solo scale.
- Docs and delivery: Notion or plain Google Docs, plus a shared folder. Don't over-engineer the client portal.
Total monthly cost, realistically: $100-$300 in subscriptions plus maybe a $20 VPS for n8n. That's the entire overhead of the business before your time.
Price for the outcome, not the tokens
Beginners charge hourly and wonder why they plateau. Here's the structure that works: a setup fee covering the workflow build ($1,500-$5,000 depending on complexity), plus a monthly retainer covering running it and handling edge cases ($500-$2,000). The client pays for the outcome — a newsletter shipped, tickets triaged, proposals drafted — not for your API bill.
Package it into three tiers if you can: DIY (you hand over the workflow and a Loom, cheapest), Done-with-you (monthly tweaks), Done-for-you (you run it end to end, highest price). Three tiers anchor the middle option and lift your average deal by 20-40% versus a single price.
Raise prices every three to four clients. Your workflow improves, your turnaround drops, and clients referred by earlier clients tolerate higher pricing more easily than cold prospects.
Delivery: template everything, automate the middle
The solo advantage is intimacy; the solo risk is drowning in custom work. Solve it with templates. Every repeated step — intake form, workflow skeleton, QA checklist, client report — should exist as a reusable artifact. When you notice yourself doing the same thing three times, you stop and templatize it.
Automate the boring middle. Client recordings arrive in a watched folder, n8n transcribes and drafts, you review the draft against a checklist, the client gets a same-day turnaround. Your value is the review and the judgment, not the keystrokes. If you're still manually copy-pasting between apps, you don't have a business yet — you have a job with extra steps.
Keep a "done means done" checklist per deliverable. It prevents rework, which is where solo margins quietly die.
Getting the first ten clients without an audience
Skip the follower-count vanity metrics for now. Start with 30 warm conversations: past colleagues, local business owners, niche communities you actually belong to. Offer a pilot at 50% off in exchange for a testimonial and a case study. Most of your first clients come from these conversations, not from inbound.
Content helps, but pick one channel and be consistent: a LinkedIn postmortem of one client win per week, or a monthly teardown in a relevant newsletter. One good case study with numbers beats fifty generic hot takes.
The traps that kill solo AI operators
- Custom everything. Every bespoke build resets your clock to zero. Templatize or decline.
- No boundaries. "While you're here..." is how a $2k project becomes an unpaid $6k job. Written scope, change orders, always.
- Model dependency. Build on one API provider and you'll feel every price hike and outage. Keep prompts portable and test open-source fallbacks quarterly.
- Skipping the boring paperwork. Simple services agreement, liability limits, data-handling terms. Clients feed you sensitive data; one sloppy leak ends the business.
The boring admin that protects the business
Solo operators love to skip the unglamorous layer until it bites. Four things to set up in week one: a simple LLC or sole-prop structure in your jurisdiction, a separate business bank account, professional liability (E&O) insurance — roughly $500-$1,500 a year and your ticket to enterprise clients — and a one-page services agreement covering scope, payment terms, IP assignment on delivery and a liability cap. Get a lawyer to review your template once for $500, and you'll never think about it again.
Also: track every workflow you build as a reusable asset, with versioned prompts and an eval set. When a client says "can you tweak this?", the honest answer is fast only because you kept the receipts. Sloppy asset management is the quiet tax on solo operators.
When (and how) to hire your first help
Stay solo as long as the economics are obscene — usually until you're turning away work or burning out on delivery. The first hire is almost never a salesperson and almost never a junior prompt-wrangler. It's a part-time operator: scheduling, invoicing, first-pass QA, client wrangling. Ten to fifteen hours a week at $25-$40. That alone can buy you back a full day of delivery time.
The second hire is a delivery generalist you trust to run templated builds with your review. Hire slowly, on paid trials, and only after the repeatable parts of your work are genuinely templated. Hiring to escape chaos just exports the chaos to payroll.
Handling demand you can't serve alone
The good problem arrives quietly: three discovery calls in a week, a firm you like, and a calendar that can't absorb all of them. The amateur response is to hire fast or cram everything into existing hours and let delivery quality slide. The operator response is a capacity triage. First, raise your pilot price by 20% and see who blinks — scarcity pricing tells you what the market actually thinks your time is worth. Second, build a small overflow bench: two or three trusted freelancers who've seen your templates and can run a templated build with your review pass at the end.
Never put an overflow contractor in front of a client without telling them. One line in the proposal — "builds are assisted by a vetted contractor under my supervision and reviewed by me personally" — preserves trust and covers you legally. Clients rarely object; what they object to is discovering it. Keep the contractor's work product, prompts and eval notes inside your own repo, so the asset stays yours if the relationship ends.
And learn to wait. A two-week waitlist signals demand rather than shrinks it, as long as the first reply sets the expectation honestly. What kills a solo practice isn't too much interest; it's over-promising to win the deal and then delivering a haunted version of the work you're known for.
Pricing reality check from the trenches
A word on discounts: intro pricing is fine as a one-time pilot lever, never as a standing offer. Every week you discount, a client learns your real price. Instead of cutting price, cut scope — fewer document types, a slower SLA, a shorter pilot — so the price you anchor publicly stays intact. Second, cash flow: take 50% up front on builds and require the retainer invoice on day one of the month. Solo businesses die of lag, not of lost deals, and two late invoices will teach you that lesson faster than any book.
Numbers I see consistently: starter pilots at $1,500-$3,000, mid-tier workflow builds at $3,500-$8,000, and monthly retainers at $750-$2,500 depending on volume. Business owners rarely balk at the retainer once the pilot shows savings — the pain point is the upfront build fee, so anchor it with a "credited toward the retainer if you convert within 30 days" clause. Script your sales calls the same way every time: understand the painful workflow, quantify the current cost in hours or dollars, show a rough prototype against their real (sanitized) data, then quote. Skipping the prototype to save an hour is how pilots stall.
A sample week in the life
Monday morning is delivery review: you scan the outputs n8n queued overnight, fix the two drafts that missed the tone, and send client reports. Mid-morning is a 30-minute call with a prospect. Tuesday is deep build — improving a prompt batch, wiring a new document source for the law firm client, testing against their eval set. Wednesday mornings are admin: invoices out, retainer renewals, scope for the change order they just asked for. Thursday is delivery again plus one public-facing activity — a postmortem, a teardown, a comment on a niche forum. Friday afternoon you do the thing almost nobody does: a 30-minute review of your metrics — hours delivered vs billed, pipeline, churn signals — and one improvement to the templates.
Total outbound sales time: two to three hours a week, once referrals are flowing. That's the compounding payoff of niche positioning. The work you do repeatedly is the work you get known for, and being known for one thing is the whole game.
A sane month, financially
Say you hold six retainer clients at $1,200/month and two projects a month at $3,000. That's $13,200/month on $200 of tools. Subtract taxes and you're still clearing well above most salaried roles — and every month the templates get better, which means more clients at the same effort, or the same clients with fewer hours.
The one-person AI business is real, but it rewards operators who treat it like a business: narrow offer, templated delivery, value pricing, and ruthless scope control. Do that, and the models do the heavy lifting while you do the judgment.