AI & Your Business, Part 1. A plain-English series for business owners who already use AI in their day, written from inside a business that runs on it.
For a small business, one of the biggest AI changes this year is Claude Cloud: AI that keeps working on a schedule after you close the laptop. It’s built on Claude Code, Anthropic’s AI tool for doing real work, and it means your AI can now work while you sleep.
The short versiontap to open or close
- AI chats forget. Every new session starts from zero, so yesterday’s work doesn’t carry into today on its own.
- Claude Cloud changes that. Claude Code can now run saved jobs on a schedule on Anthropic’s machines, even when your laptop is closed.
- It cuts cost and deployment. It runs on the subscription you already have, with no per-token bill and no server to build and deploy, within your plan’s usage limits and a daily cap on runs.
- That makes around-the-clock help realistic for a small business: a morning brief, email sorted, numbers pulled together, payments checked, reports drafted.
- It still takes a human. Refinement, creative guardrails, approvals, security and judgment stay with people.
- That’s the part we build. The structure, security, memory and checks that make it actually work, and then we do the work for you.
First, the problem: AI forgets
If you use AI in your business, you’ve felt this. You spend an hour getting a chat to understand your business, your customers and how you like things done. It does great work. Then you open a new chat the next day and it’s gone.
That’s the session-to-session problem. Each conversation is its own island. The AI can’t pick up where it left off, can’t check on something overnight, and can’t remind you what it found yesterday.
So you become the memory. You paste the same background in again, re-explain the rules, and carry the unfinished work in your head.
What changed: Claude Cloud
Claude Code is Anthropic’s AI tool for doing the work itself. In Anthropic’s own overview, it “reads your codebase, edits files, runs commands, and integrates with your development tools.”
The big step is where it can now run. We call it Claude Cloud for short. Anthropic’s own docs call the pieces Claude Code routines and cloud sessions.
A routine is a saved Claude Code job: a set of instructions, the project it works in, and the tools it’s allowed to use. In Anthropic’s documentation on routines, they run on Anthropic’s cloud “so they keep working when your laptop is closed.”
The 5 reasons, one by one
- It keeps working when your laptop is closed. The job runs on Anthropic’s machines. Your computer can be off, and so can you.
- It cuts two costs: the per-token bill and the deployment work. Our $20.03 for August is metered platform usage: our automated jobs call the pay-per-use API and are billed by the token. A Claude Cloud routine runs on the subscription you already pay for, so moving a job there takes it off the per-token bill and onto your plan’s usage. It also removes the deployment work. A routine is a saved set of instructions, a project and a schedule, with no server function to build, deploy, host and schedule. The honest limit: plans have usage limits and a daily cap on runs, and jobs that must answer instantly or run more often than hourly still belong on a server.
- It runs on a schedule you set. Hourly, nightly, weekdays or weekly, or once at a set time. It can also start when another system calls it. The shortest schedule is once an hour.
- The work carries over from one session to the next. Every run starts from the same saved instructions and the same project, so your rules, your notes and your history are there each time. A routine can write down what it found, and the next session starts by reading it.
- It can use the tools you’ve connected. Email, calendars, documents and the other services you’ve linked. Anthropic’s docs are clear that these connections act as you, so each job should get only the tools it needs.
6 jobs Claude Cloud can do while you sleep
Here’s how this looks in our own business. Each one says honestly whether it’s something we do today or something we’re building toward.
- Check and sort your email. We do this today with AI in our working sessions: read everything, sort what needs a reply, flag what’s urgent. A routine can run the first pass before the day starts, with every reply still waiting for a person.
- Pull your accounting numbers together. We do this today: scheduled jobs sync our PayPal payments into QuickBooks every week and send a Friday summary of the numbers. An idea we’re weighing: a routine that reads those numbers and writes the plain-English story behind them.
- Confirm a payment landed. On our plan, as our first routine: after a payment is recorded overnight, a one-time check confirms the invoice balance reads zero, without the routine ever holding the keys to our accounting system.
- Have a morning brief waiting. On our plan: a weekday routine that gathers what happened overnight into one short page before we sit down.
- Draft the weekly reports. On our plan: the routine writes the drafts. A person reviews and approves every one before anything goes out.
- Run a weekly site health check. On our plan: a scripted check can only list what failed. A routine can dig into why and explain it.
What still takes a human
The cloud makes the work run on time. It doesn’t make it right on its own. These stay with people:
- Refinement. The first draft is rarely the final one. Someone who knows the business shapes it.
- Creative guardrails. Your voice, your brand, and what you’d never say to a customer.
- Approvals. Anything that reaches a customer, moves money or goes public waits for a yes.
- Security. Deciding what each job is allowed to touch, and giving it nothing more.
- Judgment. Anthropic’s own docs note that a green “run finished” status only means the job ran. It doesn’t mean the task worked. Someone reads the result.
What we did today
Today is Saturday, September 26, 2026. We didn’t just read about this. Here’s what got done:
- Measured our metered AI bill to the cent. Every pay-per-use API call our automated jobs made in August came to $20.03, billed by the token.
- Moved a client-profile tool off the paid API. It now does that work inside our regular working sessions.
- Moved our nightly email intelligence job to batch processing, which Anthropic bills at half price, and put it live.
- Inventoried 108 scheduled jobs, cloud functions and background tasks, and paused or slowed down the ones that had gone quiet.
- Created a locked-down account for our first cloud routine, with access to only what that one job needs.
- Started a security review of our cloud setup, because the more that runs on its own, the more the foundation matters.
The per-token bill was the small part. The bigger saving is the deployment work that no longer has to happen for every new job, and the bigger value is work that runs on schedule, leaves a record and gets checked.
Where we come in: the pillars
Most businesses already have plenty of AI. What they need is for it to actually work, every day, without someone babysitting it. That takes four pillars:
- Structure. Clear jobs, clear schedules, and one place where the work and its history live.
- Security. Each job gets its own limited access, and nothing sensitive sits where it doesn’t need to.
- Memory. Your rules and your business context written down once, so every session starts with them.
- Checks. Every job proves it worked, and a person approves anything that leaves the building.
We build those pillars, and then we do the work for you. Our AI Engineer service starts with a flat-fee discovery and a mutual NDA, reads what you have without changing it, and gives you a written plan with what to build first and what it would cost.
Want to see the tool itself first? Start at Anthropic’s Claude Code product page. When you’re ready to make it work in your business, that’s where we come in.
Need Help with Your AI Project?
Automate your business like we did.
How we counted: the $20.03 is Anthropic’s admin cost report for the key our cloud functions use in August 2026, and it doesn’t include the subscription we work in. The 108 is our inventory of Google Cloud functions, scheduled jobs and background tasks on September 26, 2026. Descriptions of Claude Code and routines come from Anthropic’s Claude Code documentation, read the same day. Routines are a research preview, so details may change.