OpenAI Opens ChatGPT to Small Business: AI That Does the Work
This week AI stopped talking and started doing the work
Hello,
The big news this week is not a smarter chatbot. It is that the major AI companies have quietly turned their tools into workers. OpenAI opened a ChatGPT for Small Businesses program built on its new GPT-5.6 workspace, aimed at owners who want the software to finish a job rather than just answer a question. Anthropic pushed Claude further into desktop work, and Google leaned into Gemini for anything that lives inside a Sheet, a Doc, or your Drive.
Underneath the announcements there is one real shift. AI agents that carry out a full task, not a single reply, now sit inside the apps you already use. For a small team that changes the maths. The repeatable, boring work can be handed off. The judgement stays with you. That is the part worth paying attention to, so let us walk through what it means, the tools worth trying, the one win you can set up this week, and the scams that got a lot more convincing.
OpenAI's new small business program is aimed at more than 5 million business users, with a workspace that runs tasks across your files and apps rather than answering one question at a time.
From answers to agents
For two years AI mostly gave you words. You asked, it replied, you did the actual work. The change this week is that the top vendors have each shipped an agent, a version of the tool that carries out a multi-step job on your behalf. Some run in the cloud and keep working while your laptop is shut. Some sit on your desktop and drive your apps directly. Others are built for code.
You do not need to track every product name. The point for a business owner is simpler. The market has moved from "should we use AI" to "where exactly should a human stay in control". That is a much better question, and it is the one I would be asking in your shoes.
Here is the honest read. This is genuinely useful, and it is also easy to get wrong. An agent that books, replies, or reconciles on your behalf is only as safe as the process you put around it. Hand it a messy workflow and you get faster mess. Hand it a clear one and you get hours back. That order matters, and most people get it backwards.
The advice from the people building these tools is the same advice I give clients: pick one repeatable task, set a 30 day test, give it a human owner, and measure the output against real work.
Four tools, not forty
Every week I get asked which AI tool to buy. The answer is almost never "another one". You do not need a stack. You need one tool per job and the confidence to actually use it. If you are starting from scratch, these four cover most of what a small business needs, and none of them require code.
A meeting notetaker
Fathom or Otter sit quietly in your calls, take the notes, and hand you the action items afterwards. You stay present in the conversation instead of scribbling and missing half of it. This is the easiest first win for most owners.
A thinking and writing partner
ChatGPT or Claude for proposals, emails, quotes, and long documents. Claude in particular handles very long files well, which helps when you are working through a detailed report or a big quote. Treat it as a sharp junior who drafts fast and still needs checking.
A scheduling assistant
Motion or Reclaim.ai builds your day around your real priorities and quietly reshuffles it when a meeting moves. If your calendar runs you rather than the other way around, this is worth a look.
One automation tool
Zapier or Make connects your apps so information moves without you copying and pasting it. This is the tool most people skip, and then they wonder why nothing joins up. It is also where the biggest time savings usually hide.
If you want a fuller walkthrough of how these pieces fit together before you commit to any of them, I put the whole approach in one place: how to implement AI in your small business.
Your quick win this week
Do not try to roll AI across the whole business at once. That is how good tools get abandoned in month two. Prove it on the smallest job first.
Pick one task you do the same way every week. Quote follow ups. Booking reminders. Weekly reporting. Something with a clear start and a clear finish. Give it to a human owner, one person who watches the output and stays accountable for it. Set a 30 day window and measure the result against the real work, not a polished demo. If it saves you real hours, keep it and move to the next task. If it does not, drop it without guilt and try a different one.
One task. One owner. One 30 day test. That is the whole method, and it works precisely because it is small enough to actually finish.
This is the part I care about most, because it is where the systems thinking beats the shiny tool. A business owner in the Illawarra does not need forty AI subscriptions. They need one workflow that finally runs without them standing over it. Structure first. The tool second.
The scams got a voice
Here is the part I will not soften. The same tools saving you time are being pointed straight back at you, and the giveaways we all relied on are gone. Three to know this week.
Cloned voices on the phone
A few seconds of audio, pulled from a video, a voicemail, or a recorded talk, is now enough to fake someone's voice. In Australia, AI voice cloning scams have been linked to millions of dollars in reported losses, and the target is usually whoever can move money. Agree a code word with your family and your team. Verify any urgent money request by calling back on a number you already have, never the one in the message.
Australians reported around AUD 25.8 million in AI voice cloning losses in the first half of 2025. I would treat this figure as indicative and worth verifying, but the direction is not in doubt: this is rising fast.
Flawless phishing emails
AI now writes clean, personalised emails that copy the tone of your bank, your supplier, or your own staff. The clumsy typos that used to flag a scam have disappeared. Confirm any change to bank details or payment terms through a second channel before you pay anything. A thirty second phone call is cheaper than a wrong transfer.
Confident, wrong answers
AI can invent facts, figures, and sources that look completely right. If you publish them or act on them, the liability sits with your business, not the tool. Check anything an AI produces before it goes out the door. It is a first draft, never the final word.
The bigger picture
Strip away the product launches and one thing is clear. AI is becoming business infrastructure, not entertainment. The businesses pulling ahead are not the ones with the most tools. They are the ones who put AI inside a workflow that already made sense, kept a human in charge of the judgement, and left the rest alone.
That is the whole game for the back half of 2026. Not whether to use AI, but where you keep a person firmly in control. If you want a hand mapping which of your workflows are ready for it and which need fixing first, that is exactly the work I do with businesses across the Illawarra. You can see how that looks locally on the AI automation Wollongong page.
This is not a tool problem. It is a systems problem. Fix the system, then let the tool do its job.
See you next week. Keep the structure tight and the AI honest.
