The best AI tool for small business is the one that passes the Friday test.

The Friday test is a single operational question: can this AI tool keep delivering your business outcomes on a Friday afternoon when you are not at the desk? Most SERP-winning posts rank AI tools by feature category (best writer, best transcriber, best CRM AI). That ranking assumes a human is always driving. A solo consultant or small-team owner is the only person driving. If you have to be present for the tool to ship outcomes, it is not an AI tool for your small business, it is a faster keyboard. Clone's own source file features.tsx, line 58, is the one file on the SERP that names the outcome explicitly: "a business that actually runs when you take a Friday off." That sentence is what this guide uses as the pass-fail metric.

features.tsx, lines 54 to 60, in the website/src/components/ directory. It is the sixth feature card on the homepage, titled "Hours back every week." The description reads: "Solo consultants report reclaiming 10 to 15 hours a week within the first month: less admin, more billable focus, and a business that actually runs when you take a Friday off." The detail line pegs 30-day ROI at 11x based on reclaimed billable hours. You can verify by opening the source of cl0ne.ai and inspecting the feature cards, or by reading the file directly in the product repo.

A copilot waits for you to open it. Every top 10 SERP winner for this keyword is a copilot: ChatGPT (you type into chat.openai.com), Otter (you click into otter.ai for a summary), HubSpot Breeze (you open HubSpot's sidebar), Jasper (you open Jasper's editor). The AI is a feature inside a single product surface. An operator sits above the surfaces and drives them. Clone's architecture.tsx line 20 names layer 3 as the "Clone Computer Agent" with the sublabel "Reads the screen, clicks, types, scrolls." That one sentence is the difference. A copilot needs your hands. An operator needs your approval, asynchronously.

Clone runs as a desktop app, so the Mac has to be on and signed in. Most solo consultants leave their laptop open at home on Fridays; plug it into power, let the screen sleep, and the Computer Agent wakes to run scheduled rituals. The Planner layer (layer 2 of the architecture) interprets your outstanding instructions, and the Computer Agent (layer 3) opens the app, types, and saves a draft. Nothing ships to a client without you approving from the review queue. That is the safe-at-rest mode: work happens while you are away, but only sends when you approve from your phone or the next time you open the laptop.

None of them, by design. architecture.tsx, principle three on line 56, states: "Tool agnostic by design. Clone uses the apps you already pay for. Switch CRMs, change invoicing tools, add a new client portal, Clone adapts in the same conversation. No re-wiring required." If you already pay for ChatGPT, Otter, Canva, Jasper, Grammarly, HubSpot, QuickBooks, or Calendly, keep them. Clone opens those windows on your Mac and operates them. Some owners eventually drop one or two tools after Clone covers the same job, but the upfront rule is additive, not replacement.

It is the published price on the Solo tier, listed in pricing.tsx lines 7 to 26. Everything is included: unlimited plain-English tasks, Mac app, Gmail/Docs/Sheets/Zoom/any app drivers, CRM and invoicing integrations, voice learning, scheduled recurring tasks. No per-task metering, no connector fees. The math is built against the claim on features.tsx line 60: 30-day ROI of 11x based on reclaimed billable hours at typical consulting rates. At a $150 hourly rate, reclaiming 10 hours a week is $6,000 a month of freed capacity for a $49 subscription. Even one Friday of reclaimed admin inside the 21-day free trial covers a year.

Zapier and Make are excellent for deterministic API-to-API jobs. If tool A sends an event and tool B needs a row inserted, a Zap is the right fit. They fail the Friday test in a different shape: they break the moment a tool in your stack is not in their connector catalog (a Filemaker CRM, a local QuickBooks Desktop install, a state government portal, a niche vertical tool). Clone sits above that layer. It types into the same UI you type into, so the catalog is the set of apps on your Mac, not a curated list. Most small businesses that use both keep Zaps for the deterministic chores and let Clone handle the judgment-shaped, multi-app Monday work.

Yes. The Solo plan ships with a 21-day free trial, no card up front. The honest test is the one on day 6 of the seven-day kickoff: sign off at 4pm on a Thursday and do not touch your laptop until Monday at 9am. Come back and read the review queue. If the drafts are there (invoices, chase notes, transcripts, follow-ups), the tool passes the Friday test for your stack. If the queue is empty, Clone's Planner misread your instructions, the Computer Agent could not drive one of your apps, or your Mac was asleep. Any of those are repairable in the second week. Book a 30-minute call and we will map the three instructions most likely to land in your queue on day one.

Same mechanism, longer horizon. Clone's Memory layer (layer 4) holds client context, voice patterns, and the cadence rituals. The Planner schedules recurring rituals (weekly invoices, weekly status notes, daily inbox summaries). The review queue accumulates until you open the laptop. Most users take a one-week vacation in month two and come back to 40 to 60 queue rows that approve in 20 minutes. That is the same test, scaled. The 11x ROI from features.tsx line 60 is computed against reclaimed hours on a four-week rolling window, so a week off moves the numerator instead of reducing the denominator.

Yes. Pricing tier Boutique at $129 per seat per month adds shared client memory across the team, firm-wide playbooks, role-based permissions, usage analytics, scheduled firm-level rituals, and Slack/Teams notifications. The Friday test extends naturally: the firm's invoicing and follow-up still ship over a long weekend because the operator runs on every team member's Mac under shared rituals, and the review queue becomes a shared inbox of drafts to approve.

architecture.tsx principle one, line 47: "Runs on your machine. Clone operates your desktop apps from your desktop. Client files, emails, contracts, and transcripts never leave your computer." The Computer Agent reads pixels and types locally. The Planner layer calls a hosted model to interpret your English sentence (the sentence, not the client data). On the Enterprise tier, both layers run against a local or private model so even the English sentence stays on-device. Memory lives in ~/.clone/memory/ as plain markdown. You can grep it, commit it to git, or delete it with rm.

They are written as category menus. Best writer, best designer, best transcriber, best CRM AI. That shape is useful if you are buying the first one, two, or three AI subscriptions for your business. It stops being useful at tool number four, because the next dollar buys less than the first. Nobody writing a category list has a reason to ask whether the tool runs on its own. It is a different product shape and a different beat. That is the SERP gap this page is built to sit in.