AI & Your Business, Part 2. A plain-English series for business owners who already use AI in their day, written from inside a business that runs on it. Part 1 was about AI that keeps working while you sleep. This one looks a little further down the road.
Here’s a question I expect most business owners will be able to ask out loud within a few years: “How did we do this month, who still owes us, and are we ready for taxes?” And get a straight answer from one place, in plain English, without opening a single report.
That’s what I mean when I say the future of your business is a conversation. Your books, your email, your software and even your taxes, a chat away. And one step past that, the assistants start talking to each other.
The short versiontap to open or close
- Today, you’re the glue. Your business lives in a dozen apps, and you carry the answers between them.
- The front door is moving into the chat. Intuit’s QuickBooks now works inside Claude and ChatGPT, and TurboTax gives tax estimates inside Claude.
- Next, the assistants talk to each other. We expect a customer’s assistant to ask, yours to answer in your voice, and you to approve with one tap. That part is our forecast.
- We already run our own business this way: one working chat connected to 17 systems, with a person saying yes on anything that matters.
- It only works on a solid foundation: your facts in one place, limited access, written rules, checks and a human yes.
- That foundation is what our AI Engineer service builds.
Today, you’re the glue
Count the logins it takes to run your business. Accounting. Email. The calendar. The website, the online store, payroll, the bank, the project board. Each one has its own screen, its own password and its own way of doing things.
So when a simple question comes up, like “Can we afford to hire someone in January?”, you open three of them, copy a number from one into another, and do the math yourself. The software works fine. The trouble is the seams between it, and right now, the seams are you.
The front door is moving into the chat
The apps aren’t going anywhere. What’s changing is how you get in.
- Your books. Intuit’s QuickBooks now has a connector in Claude and a plugin in ChatGPT. Intuit says you can go from finding a customer to sending an invoice and a payment link inside the chat, with every action reviewable and checked by a person.
- Your numbers. Intuit’s own example questions for its QuickBooks plugin in ChatGPT include how your business is doing this month and who owes you money right now.
- Your taxes. TurboTax has a connector in Claude for personalized tax estimates and refund projections.
- Everything else. Underneath all of it is an open standard called the Model Context Protocol. Its own site calls it “a USB-C port for AI applications”: one standard plug, so an assistant that supports it can connect to any system built for it.
You’ll still pay for accounting software. You may just open it a lot less.
Next: your assistant talks to theirs
This next part is a forecast, so I’ll label it as one. Here’s where we expect this to go.
Picture a Tuesday a year or two from now. A customer’s assistant emails yours: “Did our order ship, and can we add two more?” Yours checks the store, sees it went out this morning, drafts an invoice for the extra two and writes back in your voice. You get one line on your phone: “Approve the invoice for two more?” You tap yes.
Then it runs the other way. Before the quarter closes, your assistant asks your bookkeeper’s assistant for what it needs, and the tax estimate is waiting for you Monday morning.
Two assistants did the back and forth. One person made the call.
The plumbing for that already exists, too. The Agent2Agent protocol, which Google started and then gave to the Linux Foundation, is an open standard for AI agents from different companies to talk, share information and hand off work. As software and assistants keep blending together, we expect a lot of everyday business back and forth to run this way, with people approving the parts that matter.
We already run ours this way
This part isn’t a forecast. As of October 16, the working chat we run Stand And Stretch from is connected to 17 systems: our email, our project boards, our files, our payments, our email marketing and our domains, plus the websites, online stores and ad accounts we manage. Across those connections it can take 197 different actions. We also read and update our books in QuickBooks from the same sessions, through code we built for it.
Here’s what that looked like on Friday, October 16. One approval in that chat published our latest post: the post went live, ten social posts went out and each one was checked after it posted, and the email went to our list. It took 13 minutes, and a person said yes before any of it happened.
What it doesn’t do is decide. It doesn’t email a client, move money or post anything public without a person saying yes. That’s by design.
What it takes to get there
Talking to your business only works if the pieces underneath it are solid. Five of them:
- Your facts in one place. The assistant can only answer from what’s written down and connected. If the real price lives in your head, so does the answer.
- Limited access. Each connection gets only what its job needs. If a job should look up a price and never place an order, it gets built without the ordering part at all.
- Rules written once. Your voice, your prices and the things that never happen without you, written down so every session starts with them.
- Checks. A finished job isn’t always a job that worked. Every job reads back what actually changed.
- A person’s yes. Anything that reaches a customer, moves money or goes public waits for one.
There’s one more piece, for when a customer’s assistant comes looking for you: it can only repeat the answers you’ve published. That’s why our last post was about FAQs.
How we help: our AI Engineer service
This is the work our AI Engineer service does. Here’s how it runs:
- A flat-fee discovery, agreed up front, and a mutual NDA before anyone looks at anything.
- We read what you have without changing it: the software you use, the spec you wrote with AI if there is one, and the rough version you built yourself.
- You get a written plan: what to connect first, what it should never touch, and what the build would cost. The plan is yours whether or not you build it with us.
- Phase one is small on purpose: the foundation plus one everyday task working end to end, so you see it run before you commit to more.
- Everything we build is yours, documented in plain English, on the same five pieces above.
Your software is already learning to talk. The work is making sure it says the right things, to the right people, with you in charge.
Need Help with Your AI Project?
Automate your business like we did.
How we counted: the 17 is the number of systems connected to our working setup on October 16, 2026, the same count shown on our AI Engineer page. The 197 is the number of actions those connections offered that morning, counting the 15 that run on our own machines. The Friday publish comes from our own publishing record for that post. Product details come from Intuit’s QuickBooks and TurboTax announcements and from the Model Context Protocol and Agent2Agent project sites, all read October 16, 2026. Anything about the future is our forecast, and product features may change.