The Planner-layer angle
AI software for small business where you do not pick the app. The Planner layer does.
Every other winner on this page of Google asks you to open a dropdown and pick a trigger, an action, or a template. Clone does not. You type one English sentence. Clone's Planner layer scans your HubSpot, your Sheets, your Stripe, your Gmail, and whatever else you already pay for, then decides which surface to drive. No graph. No wiring. One sentence, one inferred plan.
Page one of Google, in one image
Twelve SERP winners. All of them hand you a dropdown. None of them pick the app for you.
Zapier, Make, n8n, Workato, Power Automate, Salesforce Flow, HubSpot Breeze, ClickUp Automations, Notion AI, Pipedream, Tray.io, IFTTT. The first click on every one of them is the same click: choose a trigger, or choose a template, or choose an app.
The core claim
The Planner layer is
the piece every other product skipped
.The shape of a workflow automation tool assumes you already know which app should fire. Zapier starts with 'Choose an App'. Make starts with a trigger module. HubSpot Breeze starts inside HubSpot. Salesforce Einstein starts inside Salesforce. All of them assume the app is input, not output.
In a working consulting practice, that assumption breaks. You do not have one billing tool, you have four, and the one to use depends on the client. You do not have one follow-up surface, you have three, and the one to use depends on where the last thread ended. Picking the tool per instruction is the actual work. Every SERP winner handed that work back to you.
Clone's Planner layer does that work for you. It sits at layer 2 of 6 in the stack. It reads your English, it reads your rituals file, and it picks the surface. From your side, there is no dropdown. There is a chat window and a sentence. The Planner absorbs the picking so the rest of the stack can get to the driving.
The uncopyable detail
This is Step 01 of cl0ne.ai's own how-it-works section. Verbatim.
Below is the literal code example shipped at src/components/how-it-works.tsx lines 9-20. The user typed one sentence. Clone scanned three apps (HubSpot, Sheets, Stripe), found 14 active engagements, and decided to build the dashboard in Sheets. Nobody clicked an app picker. The Planner did.
And here is the layer itself, named in architecture.tsx. Six layers in the stack. The Planner is layer two. Its entire sublabel is 'Understands intent, picks the right apps'. That line is the whole product category gap.
The Planner, end to end
One sentence in. Four drafts out. Here is every hop in between.
Zapier shows you boxes and arrows. Clone hides the boxes. Here is the same flow rendered honestly, as a sequence diagram, so you can see exactly what the Planner, Memory, and Computer Agent do on one Monday morning invoicing pass.
Monday 9:02am, instruction: 'invoice last week's hours'
The surfaces the Planner can pick
Clone at the center. Your existing stack in orbit.
The Planner does not have a fixed set of connectors, because the Computer Agent drives whatever is on your screen. API-backed or not, legacy or modern, desktop or web, it is a candidate surface.
By the numbers
Four numbers you can verify by opening the product.
Six instructions. Six Planner decisions.
For each sentence, the Planner picks a different surface. You do not tell it which one.
Invoice last week's hours
Planner reads the rituals file in ~/.clone/memory/, sees that Acme is a Timely-plus-QuickBooks-Desktop client and Nexora is a Toggl-plus-Sheets client, and routes each invoice to the matching surface. You did not pick the tool, the Planner did.
Build me a client health board
Planner scans HubSpot, Sheets, and Stripe (the literal behavior at how-it-works.tsx line 12), finds 14 active engagements, picks Sheets as the surface, and builds the board there. No choosing between tools.
Draft follow-ups for the four kickoffs last week
Planner picks Gmail because that is where your kickoffs live, pulls the four threads, and hands them to the Computer Agent to draft four emails. The only dropdown you never saw was 'choose a source'.
Send the Friday retro
Planner reads your rituals, sees you write retros in Notion on Friday at 4pm, opens that database, drafts the retro, and saves the draft for your review. No scheduler. No Zap. One sentence input.
Remind clients who ghosted
Planner cross-references your last two weeks of sent email (Gmail), your HubSpot pipeline, and your Calendly history to find threads with no reply in seven days. It queues reminders for review. You never asked it to query three systems.
Pay my contractors
Planner sees contractor invoices in Gmail, matches them to rows in your Bill.com queue or your bank's bill-pay screen (whichever you already use), and lines up payments for approval. It picks whatever surface you currently pay with.
One instruction, two shapes of day-one
Zapier wiring vs. Clone instruction.
Both start with the same business need. One asks you to be an integrator for 45 to 90 minutes before anything useful happens. The other asks you to approve four drafts in a review queue.
1. Pick an app for the trigger. Timely? QuickBooks? A Gmail label? 2. Pick the exact trigger event. New time entry? Scheduled cron? 3. Add a filter. Only entries from last week. Only billable. 4. Pick the action app. QuickBooks Online? QuickBooks Desktop? Sheets? 5. Pick the exact action. Create invoice? Create estimate? 6. Map 22 fields between the two apps. 7. Test the Zap. It fails because the trigger fires on save, not approval. 8. Add a second filter. Test again. 9. Deploy. Hope Monday is the same shape as today.
- You pick the trigger, the app, the action, the filter, and every field.
- The tool does not know which client uses which invoicing surface.
- When you add a new client next quarter, you rebuild the Zap.
- Total setup: 45 to 90 minutes before a single invoice goes out.
Four minutes on a real Monday
What the Planner actually does between 9:02 and 9:06.
9:02am. You type one sentence.
You open Clone on your Mac and type 'invoice last week's hours'. You do not pick Timely, QuickBooks, Clio, LeanLaw, Sheets, or a Zapier template. You type five words.
9:02am still. The Planner scans.
The Planner reads your ~/.clone/memory/rituals/invoicing.md file. It sees Acme on Timely plus QuickBooks Desktop, Nexora on Toggl plus a custom Sheets template, Holloway on LeanLaw, Nightshade on a notarized PDF you generate by hand.
9:03am. Four jobs queue to the Computer Agent.
The Computer Agent layer (layer 3 of 6, architecture.tsx lines 19-22) starts opening the four surfaces on your Mac. QuickBooks Desktop launches. The Sheets invoice template opens. LeanLaw loads in your browser. The PDF template opens in Pages.
9:05am. Four drafts in the review queue.
You see a four-row review queue in Clone. Acme invoice at $8,250 (31 hours at $275, prepped in QuickBooks Desktop). Nexora invoice at $2,400 (three milestones, Sheets template). Holloway at $4,700 (LeanLaw). Nightshade at $6,000 flat (PDF in Pages).
9:06am. You go back to the client work.
Nothing was picked by you except whether to ship. No app was chosen. No trigger was configured. No field was mapped. The Planner did the picking, the Computer Agent did the driving, the review queue did the human-in-the-loop check.
The one rule that decides if this fits you
If your clients do not share one billing tool, one CRM, or one follow-up surface, the Planner is the point.
A single-tool shop does not feel the gap a Planner layer fills. If every client runs on the same HubSpot instance and the same QuickBooks Online and the same Calendly, a well-wired Zap will cover 80 percent of your ops. You do not need an AI to pick the app.
A consulting practice with six clients, four billing stacks, two CRMs, and a pile of per-client quirks is the opposite shape. Every instruction involves a small decision about which tool to drive. That is the decision the Planner absorbs. On a Monday, four minutes of instructions replace 45 minutes of app-switching.
Clone vs. the SERP, row by row
What changes when an AI layer picks the app instead of asking you
Every row is about a concrete operational consequence of having a Planner layer at position 2 of the stack, instead of handing the picking back to the user.
| Feature | Workflow builders (Zapier, Make, Workato, HubSpot Breeze, ClickUp AI) | Clone (Planner-first architecture) |
|---|---|---|
| Who picks the app | You do. Every automation starts with 'Choose an App', then 'Choose a Trigger', then 'Choose an Action'. If your stack has six billing surfaces, you build six Zaps or six recipes. | Clone's Planner layer does. It reads your instruction in English, looks at the apps you already use (with scopes you grant), and picks the surface that fits. One instruction covers every client, not one Zap per client. |
| What you configure before the first run | A workflow graph. Triggers, filters, branches, error-handling steps, retry policies, field mappings. For a Monday invoicing flow across three billing tools, this is typically 30 to 90 minutes of setup before any value is produced. | A one-sentence English instruction. Optionally, a markdown rituals file in ~/.clone/memory/rituals/ naming which tool you use for which client. The rituals file is plain text. You can write it in five minutes or let Clone infer it from your Monday behavior. |
| What the AI layer does, specifically | In most SERP winners, the AI is a suggestion layer on top of a human-configured graph. It proposes steps, translates English into a node-and-edge diagram, or fills in a template. A human still picks the app and owns the graph. | The Planner is the AI layer, and its only job is picking the app. Architecture.tsx lines 13-14 name it explicitly: 'Clone Planner, Understands intent, picks the right apps.' It is layer 2 of 6 in the stack, sitting between your sentence and the Computer Agent that drives the screen. |
| What happens when your stack changes | A rewired graph. You migrate from QuickBooks Desktop to Clio Manage, and you rebuild every invoicing Zap. Same for switching CRMs, adding a new client portal, or swapping calendar tools. The workflow is coupled to the app identifier. | Clone adapts in the same conversation. Architecture.tsx line 56 names this 'Tool agnostic by design'. You change CRMs or add a client portal, and the Planner picks the new surface the next time you type an instruction. No rebuild. |
| What the user experience feels like | You are an integrator. You open a workflow builder, drag boxes, debug why field 17 is null, and ship a graph to production. The tool is a spreadsheet for automations, and your job is to fill it. | You are a boss. You type a sentence. The Planner picks the tool. The Computer Agent drives it. You approve the result in a review queue. The only human decision is whether to ship the output, not how to build the pipeline. |
| Where the rules live on disk | In a vendor database, behind their UI. You cannot grep your Zaps, diff them across versions, or paste them from one tenant to another without their API. | In ~/.clone/memory/ on your Mac, as plain markdown. Rituals, client overrides, voice patterns. You can open them in any text editor, commit them to git, and move them between Macs with a zip. |
| What it costs to add the tenth instruction | Another workflow. In Zapier Pro ($49/mo), you get 2,000 tasks. Each task is one action in a Zap. A morning invoicing flow across six clients can burn 30 to 60 tasks before lunch. Plans scale by task count. | Nothing extra. The Solo plan is $49/mo flat. Instructions are uncapped because the work happens on your Mac via the Computer Agent layer, not on a vendor's task-metered servers. |
“I typed 'invoice last week'. Clone went and opened QuickBooks Desktop for Acme and my Sheets template for Nexora in the same breath. I did not tell it which client used which tool. It read my rituals file. I have not opened Zapier since.”
The pricing footnote
$0/mo flat. Instructions uncapped. Apps uncapped.
Zapier, Make, and Workato meter you on tasks. A Monday invoicing run across six clients can burn 30 to 60 tasks before lunch, which matters when you cross plan thresholds. Clone does not meter. The Planner runs in the cloud, the Computer Agent runs on your Mac, and neither is priced per instruction. Solo is $49 a month flat. Boutique is $129 a seat. That is it.
Common questions about AI software for small business where the Planner picks the app
What specifically is the Planner layer, and where does it live in the product?
The Planner is layer 2 of Clone's 6-layer architecture, defined at src/components/architecture.tsx lines 12-17 of the marketing site. Its verbatim sublabel in the source is 'Understands intent, picks the right apps'. Layer 1 is you (plain English instructions). Layer 2 is the Planner. Layer 3 is the Computer Agent. Layer 4 is Memory. Layer 5 is your apps. Layer 6 is the business outcome. The Planner's entire job is to read your sentence and decide which of your existing tools the Computer Agent should drive next. This split is the reason you do not pick an app in the UI.
How does the Planner decide which app to drive without me configuring anything?
Two signals. First, the rituals file in ~/.clone/memory/rituals/, which maps client-by-client or task-by-task to the tool you actually use (for example, Acme uses Timely plus QuickBooks Desktop, Nexora uses Toggl plus Sheets). You can write this in five minutes or let Clone infer it by watching your first week. Second, the context of your instruction. For 'build me a client health board', the Planner scans HubSpot, Sheets, and Stripe (the literal behavior at how-it-works.tsx lines 12-13) and picks Sheets. You never pick. The Planner does.
How is this different from Zapier or Make?
Zapier and Make start with 'Choose an App'. You pick the trigger app, the action app, and map every field. The automation graph is human-configured; the AI layer (if any) is a suggestion on top. Clone inverts this. You write one sentence. The Planner scans your stack and picks the app. There is no graph to build. If your stack has six billing tools, Zapier needs six Zaps. Clone needs one instruction and one rituals file. That is the whole delta.
How is this different from HubSpot Breeze, Salesforce Einstein, or ClickUp AI?
Those are AI layers on top of a single vendor's suite. HubSpot Breeze picks actions inside HubSpot. Einstein picks inside Salesforce. ClickUp AI picks inside ClickUp. If your billing lives in QuickBooks Desktop and your CRM lives in HubSpot and your meetings live in Calendly, a single-vendor AI cannot cross those surfaces, because it was never designed to. Clone's Planner is vendor-agnostic by construction. Architecture.tsx line 56 names this 'Tool agnostic by design'. The Planner picks the surface regardless of vendor.
Does the Planner need an API key for every tool it drives?
No. The layer that actually drives the apps is the Computer Agent (layer 3), and its sublabel at architecture.tsx line 20 is 'Reads the screen, clicks, types, scrolls'. It drives whatever is open on your Mac, using the same permissions you use. No API key. No integration wait. No vendor approval. Desktop QuickBooks, a Filemaker CRM, a Sheets template with ten years of conditional formatting, all work the same way. The Planner picks the surface; the Computer Agent drives it by screen.
What if the Planner picks the wrong app?
You correct it in chat. The Planner does not run until you approve its plan for anything that writes. If you type 'send the Friday retro' and the Planner proposes Notion but you actually write retros in Google Docs, you say 'use Google Docs' and the Planner routes there. The correction updates your rituals file so the next Friday it picks Docs by default. Corrections are data, not exceptions.
Is there a limit on the number of apps the Planner can route across?
No soft limit. The Planner is a routing layer; the work happens in the Computer Agent. Because the Computer Agent drives whatever is on your screen, a Mac that can run QuickBooks, Notion, HubSpot, Gmail, LeanLaw, Clio, Toggl, Timely, and three Sheets templates can have all of them in play during a single morning. The only practical bound is how many apps your Mac can comfortably run at once.
How is this safe? What stops the Planner from sending the wrong invoice to the wrong client?
The review queue. Every proposed action lines up as a row in Clone, named and editable before it fires. You approve the batch or you edit individual rows. Plus every action is logged and reversible, per architecture.tsx lines 61-64: 'Every action Clone takes is logged and reversible. Preview drafts before they send. See every file it touched. Roll back an entire morning of work with one click if you need to.' The safety model is identical to how a junior operations hire would work, with a pull-request-style check before anything leaves your Mac.
Does the Planner send my instruction or my app data to a hosted model?
The Planner calls a hosted model to interpret your English sentence and to pick the target app. The Computer Agent does its work locally on your Mac, so the contents of your QuickBooks file, your Sheets template, or your client emails never leave your machine. Architecture.tsx line 47 names the principle: 'Runs on your machine. Clone operates your desktop apps from your desktop. Client files, emails, contracts, and transcripts never leave your computer.' On the Enterprise plan, you can run the Planner against a local model so even the sentence stays on the Mac.
Can I inspect what the Planner decided and why?
Yes. Every session has a plan visible in chat and a log on disk at ~/.clone/memory/sessions/. The plan shows the Planner's inferred app choice, the reasoning, and the proposed actions. You can read it before you approve. You can diff two sessions side by side to see why the Planner picked Gmail one day and Docs the next. The session log is plain markdown, not a database row you need vendor API access to query.
How does the Planner know which client uses which tool?
From the rituals file in ~/.clone/memory/rituals/. A single file per workflow (invoicing.md, follow-up.md, kickoff.md), with client-by-client rows naming the tool stack for that client. You can write the file in five minutes, or let Clone build it by watching your first week of manual work. The file is plain markdown and version-controllable, so a new client becomes a new row, not a new Zap.
Where does this leave Zapier, Make, or Power Automate in my stack?
Zapier and Make are still good for deterministic, always-on integrations between two SaaS APIs (a webhook fires, a row lands in Sheets). Clone is for the other 80 percent of small-business operations, the stuff that does not fit a pre-built connector: legacy desktop apps, custom templates, per-client quirks, screen-only tools. Most small-business owners end up keeping one or two lightweight Zaps and moving the messy human-shaped work to Clone.
Adjacent guides about the layers below the Planner
Keep reading
AI Automation for Small Business with a Memory Layer You Can Read
Where the rituals file lives, what it looks like in full, and why a stateless workflow tool can never catch up.
The AI Assistant for Small Business That Operates Your Actual Apps
The Computer Agent side of the same stack. How layer 3 drives Gmail, HubSpot, QuickBooks, and Sheets from a plain English sentence.
Invoicing Automation Software That Works Without an API
Why the Planner-plus-Computer-Agent stack lets you invoice across QuickBooks Desktop, Sheets, and LeanLaw from the same morning instruction.
Type one sentence this week
Install Clone. Skip every dropdown. Let the Planner pick the app.
21-day free trial on the Solo plan. $49 a month after. No trigger picker, no field mapping, no workflow builder. One sentence, one review queue, one approved batch.
$49/mo on Solo · macOS · Planner at layer 2, Computer Agent at layer 3