OFO Collective

OFO Collective Supporting humanity’s transition into the AI era in a human-centred, ethical, and empowering way.

12/08/2026

Ask your accountant this: are you using AI on my work?

It is a fair question, and very few clients are asking it yet. Almost every professional service is quietly running some of your work through an AI tool now. That is not a scandal. Used well it is the reason a small firm can give you the turnaround a big one used to charge for. But you are the client, and you are allowed to know.

Three things worth asking, none of them confrontational.

Where in the process does it sit? There is a large difference between "we use it to draft, then a person reviews every line" and "we use it to produce the final document". Both can be fine. Only one of them needs you to trust the review.

Who carries it if it is wrong? A firm's professional obligations do not change because a machine wrote the first version. The answer you want is some version of "we do, same as always". If the answer starts drifting towards the software vendor, that is worth noticing.

Has the price changed? This is the awkward one, and it is the honest one. If a task that used to take a day now takes an hour, what it is worth to you has not dropped, but what it costs to produce has. Reasonable people land in different places on that. It is still a fair thing to raise.

What you are listening for is not a particular answer. It is whether the person can answer at all. Someone who has thought carefully about where AI sits in their work will tell you in a sentence. Someone who gets defensive has told you something too.

And it cuts both ways. If you use AI in your own business, it is worth having your own version of that sentence ready, because your customers will start asking.

Have you asked any of your suppliers yet, or would it feel rude?

11/08/2026

Three habits make your business hard to impersonate, even with AI.

The oldest scam in business is someone pretending to be a supplier, a client or the boss, and asking for a payment to go somewhere new. It has never needed clever technology. What has changed is the cost of doing it well. Writing a convincing email in fluent English, in the right tone, referencing the right project, used to take effort and some luck. It now takes very little of either.

Business email compromise has been a standing category in the Australian Signals Directorate's annual cyber threat reporting for years, and the useful thing to understand is that the target is almost never your software. It is your process.

So the defence is a process too, and it is duller than people expect.

Verify any change to bank details by phone, on the number you already had. Not the number in the email. Not a number in the signature block. The one in your own records, from before the request arrived. This single habit closes most of the risk.

Decide who is allowed to approve a payment that was not already scheduled, and make it two people above a threshold you choose. The threshold matters less than the fact that one person acting alone cannot move money out.

Agree that urgency is a reason to slow down, not to speed up. Every version of this scam depends on the recipient feeling that checking would be embarrassing or would cause a delay. Say out loud, once, as a team, that nobody will ever be criticised for making the call.

Notice what is not on that list. No new software, no subscription, no security product. The reason these three work is that they do not depend on spotting a fake. A well written impersonation is not meant to be spotted, and getting better at reading emails is not a strategy that scales.

What would happen in your business if a supplier emailed new bank details this afternoon? Who would check, and how?

There is one question that settles most worries about what you can safely type into an AI tool.Would you be comfortable ...
10/08/2026

There is one question that settles most worries about what you can safely type into an AI tool.

Would you be comfortable if this appeared in an email that got forwarded outside your business?

That is it. Not a policy document, not a risk matrix. One question, asked before you paste.

It works because it maps closely onto what actually happens to the text. Anything you put into a consumer AI tool may be stored, may be reviewed by a person for quality, and depending on the product and your settings may be used to improve the model. OpenAI's published policy is a fair guide to the pattern across the industry: business and enterprise plans are not trained on by default, while personal accounts are, unless you turn it off in settings. Worth reading yours once. The habit is worth more.

Run the test on a few things people paste every day.

A draft reply to an annoyed customer, with their name in it. The name is the problem, not the draft. Take it out and the question answers itself.

A supplier contract you want summarised. Usually fine on a business plan, rarely a good idea on a free one.

Your staff list with pay rates against names. No. That one is not a grey area.

A messy paragraph you want tidied up. Fine, and honestly that is most of what these tools get used for anyway.

The pattern underneath is simple. The risk is almost never the task, it is the identifying detail carried along with it. Strip the name, the account number, the address, the medical or financial specifics, and most work becomes safe to hand over without losing anything useful. The tool does not need to know who the customer is to write a better reply to them.

That is the whole habit. One question, then delete the names.



Photo: Crew

Most AI announcements are not products yet. They are plans with a date attached.That is the quiet frustration of running...
09/08/2026

Most AI announcements are not products yet. They are plans with a date attached.

That is the quiet frustration of running a business here and reading tech news written for California. Something remarkable gets announced, you go looking for it, and it is not there.

Three questions sort it out, and they take about a minute.

Is it shipped, or announced? Read the wording. "Rolling out", "coming weeks", "early access" and "preview" all mean not yet. Announcements get made when a company wants attention, which is often well before the thing exists for ordinary customers.

Is it available here? Plenty of new AI features launch in the United States first, and the gap before they reach Australia runs from weeks to never. Anything touching payments, bookings, maps or local data tends to be slowest, because each country needs its own partners.

Is it on your plan? The demo is almost always the top tier. The feature that impressed you may sit two subscription levels above the one your team is actually on.

None of this is a reason to ignore AI news. It is a reason to stop treating an announcement as something you have missed. The honest position most weeks is that nothing has changed for your business yet, and that is a perfectly good answer.

What deserves your attention is narrower and duller. It is the feature that quietly appeared inside software you already pay for. Nobody writes a headline about that one, and it is the one you can actually use.

Which AI feature did you go looking for and never find?



Photo: Kristian Buus / 1010 Climate Action

08/08/2026

Most AI tools will agree with you. That is worth knowing before you trust one with a decision.

Ask a chatbot whether your plan is any good and you will usually get a version of yes, with three supportive reasons attached. Not because it is lying to you. Because you handed it your framing, and it answered inside it.

The fix is a habit rather than a better tool. Ask for the other side, explicitly.

Four phrasings that do the work:

"Argue the case against this plan as strongly as you can."

"What would have to be true for this to be a bad idea?"

"If this fails in six months, what is the most likely reason?"

"What am I assuming here that I have not said out loud?"

The answers are usually more useful than the first response, and occasionally uncomfortable, which is rather the point. A tool that only ever confirms your thinking is entertainment. One that can brief you against yourself is closer to an adviser.

Worth doing before the decision rather than after it. Once you have committed, the case against reads like criticism. Beforehand it reads like a checklist.

Save these four questions somewhere you will find them before the next big call.

Most people use AI to write. The bigger win is reading.Writing is the obvious use, so it is the one everybody tries firs...
07/08/2026

Most people use AI to write. The bigger win is reading.

Writing is the obvious use, so it is the one everybody tries first. Draft an email, tidy a proposal, write a post. Useful, and also the use where the risk of sounding like everyone else is highest.

The quieter use is reading, and in a small business it is usually worth more.

A few examples of what that looks like in practice.

A supplier contract you have been meaning to read properly. Ask what the notice period is, what happens at renewal, and which obligations sit on you rather than on them. Then read those clauses yourself. You have gone from not reading it at all to reading the three parts that matter.

Two hundred customer reviews or survey answers. Ask for the recurring complaints and how often each one comes up. The answer is usually something you half suspected and had never counted.

A year of invoices or expenses. Ask what changed against the year before and where the biggest increases sit. It will not replace your bookkeeper. It will give you better questions to take to them.

A long tender or grant document. Ask for every requirement written out as a checklist, then work the checklist.

The reason reading beats writing here is the failure mode. If the tool writes something poor, you publish something poor. If it reads something poorly, you go back to the source, because you were always going to look at the important part yourself. Lower stakes, more time saved.

One rule that keeps it honest. Ask it to point at where in the document the answer came from. If it cannot, treat the answer as a lead rather than a fact.

What is the longest document sitting on your desk that you have been avoiding?



Photo: Cameron Porter, U.S. Army 405th Army Field Support Brigade

Hire a person or build a system? Most owners decide this on gut feel.There is a cleaner way to think about it, and it is...
06/08/2026

Hire a person or build a system? Most owners decide this on gut feel.

There is a cleaner way to think about it, and it is not really about cost.

Ask what the job actually consists of. If it is mostly judgement, relationships, or handling things that have never happened before, hire. That is what people are for and no system covers it.

If it is mostly moving information from one place to another, checking it against a rule, and doing that many times a week, that is a system. You can hire for it and it will work. It also means you have bought a person's attention and spent it on the dullest thing in the business.

Most roles are a mix, which is where this gets useful. Write the job out as a list of tasks and mark each one. Usually about a third of a role turns out to be genuinely repetitive. That third is what goes to software, and the result is not fewer people. It is the same person doing the part you actually hired them for.

The mistake in both directions looks the same from the outside. Automate judgement and you get confident nonsense. Hire for repetition and you get an expensive, bored person who eventually leaves.

One caution worth pricing in. A system needs someone to own it, check it and fix it when it breaks. That time is real, and it is the line most people forget to put in the sum.

If you wrote your own week out as a list of tasks, how much of it would be the repetitive third?



Photo: Tim Wright

One person understands your AI setup. That is a business risk.You know the pattern. Someone in the business gets good at...
05/08/2026

One person understands your AI setup. That is a business risk.

You know the pattern. Someone in the business gets good at this stuff. They write the prompts, wire up the automation, work out which tool does what and where the settings live. It works, so nobody looks closely at it again.

Then they take four weeks off. Or they leave.

What breaks is rarely the tool. It is the knowledge sitting around the tool. Why it was set up that way. What the fallback is when it fails. Which client the odd exception was built for, and why.

The fix is boring and takes an afternoon. Three documents, not thirty.

1. What it does and why. One paragraph per automation, plain English, written for someone who was not in the room when it was built.

2. Where it lives. The account it runs under, who pays for it, where the login sits, and what happens when that card expires. Business accounts, never a personal one.

3. What to do when it stops. Who to contact, how to switch it off, and how the job gets done by hand while it is off. If that last answer does not exist, you do not have an automation. You have a dependency.

Then have someone else follow the documents once, on their own, with no help. That is the only test that means anything.

The reason this gets skipped is that it feels like admin rather than technology. It is the difference between a system the business owns and a system one person is quietly renting to it.

Save this if one person in your business knows how everything is wired together.



Photo: Sebastiaan ter Burg

Siri finally works properly. The interesting bit is that nobody seems excited about it.Apple rebuilt it and shipped it i...
04/08/2026

Siri finally works properly. The interesting bit is that nobody seems excited about it.

Apple rebuilt it and shipped it in the iOS 27 public beta, with the full release due in September after a long run of delays. TechCrunch covered it on 3 August. It handles a real back and forth conversation, digs things out of your photos, email and calendar, and can identify what your camera is pointed at.

And the response has been a shrug.

Not because it is bad. Because a machine holding a decent conversation is not surprising anymore.

That is quietly useful if you buy software for a business. "It has AI" has stopped being a reason to pay more for anything, because everything has it now. Conversation is the floor.

Which changes the question worth asking a salesperson. Not "can it hold a conversation". Instead: "what does it finish on its own". Answering questions leaves the work with you. Completing a step, writing the quote, updating the record, sending the follow up, is what actually takes work off your desk. Similar price, completely different product.

Also worth saying plainly: the assistant on your phone is not going to run your business. It is good at finding your own information, not at operating your systems.



Photo: NordWood Themes

The same question does not always get the same answer. AI tools change underneath you.Worth knowing if you have built an...
02/08/2026

The same question does not always get the same answer. AI tools change underneath you.

Worth knowing if you have built any part of your week around one. The model behind the product gets updated by the vendor, and wording that worked beautifully six months ago can come back flatter, longer, or in a different format, with no announcement and nothing changed at your end. Nothing broke. The thing underneath moved.

Two habits turn that from a mystery into a non-event.

Keep your instructions somewhere you own. A document, not only pasted inside one product. If you ever switch tools, or the tool switches on you, you still have the thing that took you weeks to work out.

Re-read the output of the one step that matters most, every so often. Not everything. Only the place where a quiet change in tone or format would embarrass you in front of a customer.

That is the whole precaution. Ten minutes, and a baffling week of "the AI has got worse" becomes a small adjustment instead.

Has an AI tool you use ever quietly changed on you?



Photo: Glenn Carstens-Peters

Address

Suite 405H, 521 Toorak Road
Toorak, VIC
3142

Telephone

+61417487460

Website

Alerts

Be the first to know and let us send you an email when OFO Collective posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Contact The Business

Send a message to OFO Collective:

Shortcuts

Share

Category