For two weeks now we have been getting the same questions from customers, usually in the same order. Can this write our quotes. Will it replace our support desk. Do we have to do something about it right now.
The short answers: partly, no, and yes but carefully. The longer version follows, with the caveat that nobody can currently say where this goes.
What actually happened
The technology is not new. Language models of this kind have been around for a few years, and anyone with a developer account could already experiment with them. What changed is access. You type a question into a plain text box and get an answer in fluent sentences. That was apparently enough to move the subject out of research talks and onto the kitchen table.
What these models can do today
They are good at language. That sounds obvious, and it is exactly the point. Hand one a rambling meeting transcript and you get a usable summary. Give it bullet points and it turns them into a paragraph. Ask it to rewrite a customer email in a friendlier tone and it will. Ask for ten headlines and you get ten, of which perhaps three are worth keeping.
Notice what all of those have in common. The material comes from you. The model rearranges it. That is where it is genuinely strong, and strong enough to save you real time.
What they cannot do
They do not know facts in any dependable sense. They produce text that sounds like a fact. Sometimes it is one.
They cannot do arithmetic you would put on an invoice. They have no access to your systems, your price list, your stock, your contracts or your project history. Ask about your own company and the model will answer, with great confidence, about a company that does not exist. There is no live connection to anything, so anything recent sits outside its world entirely.
Why the wrong answers sound so convincing
The model predicts, word by word, what plausibly comes next. It holds no internal notion of true or false. So a wrong answer arrives in the same fluent, assured prose as a right one. With a person you can usually hear the hesitation before a guess. Here you cannot.
That single property decides where the tool belongs in a business. It belongs where a human checks the output anyway, because checking is their job. It does not belong anywhere the output goes straight to a customer, to an authority or into a costing.
Where we would start
Drafts, not results. A first version of a product description that somebody then rewrites. A summary of a long document, read by the person who still opens the document afterwards. Reworking an awkward paragraph. Test data. A rough sketch of code that a developer reviews line by line. Small, low stakes, always with a second pair of eyes on it.
What we would not do is paste customer data, contracts, costings or source code from client projects into a public text box. You are handing material to a service whose terms you never negotiated, and you do not know what happens to your input. If that sounds overcautious, ask your legal counsel and watch their face. Whether any of this is worth your attention at all is exactly what our consulting is for.
Our reading of it
Interesting enough to watch closely and to test on something harmless. Too unfinished to build a process on. We are trying it in our own lab on our own texts before we recommend it to anyone for anything that matters. Ask us again in six months and the answer may well be different. That is not a dodge. It is the honest state of things in December.
Frequently asked questions
Can we feed it our own company data?
Not usefully through the public interface. You can paste text into the question and the model will work with it, but only for that one conversation. Afterwards it is gone, and the data protection question stays open regardless.
Can we publish what it writes?
Unclear, and we would not treat it as settled. Nobody has answered the copyright side properly yet. Treat the output as a draft you rewrite, not as text you ship as it is.
Should we cut staff because of this?
No. Anyone selling you that today is guessing.
If you want a sober assessment of whether this deserves any of your attention this year, we are happy to talk it through with you, no obligation and no sales pressure.
This article is part of our knowledge hub AI and digital transformation for SMEs.