What the AI Actually Does All Day
Big tech is selling superintelligence. At my desk the AI drafts, checks, gets corrected and waits. The Australian numbers look a lot like my day.

Big tech is selling superintelligence. At my desk the AI drafts, checks, gets corrected and waits. The Australian numbers look a lot like my day.

Most working days, before the coffee in Surry Hills is any good, I have already asked an AI to do something dull. Tidy a messy page of notes. Find the date a report came out. Read a draft and tell me where it drags.
Then I open the news and someone is announcing superintelligence.
In August, Meta published an essay by its chief executive, Mark Zuckerberg, called The Future is for Everyone. It says that "in the next few years, people will be able to use superintelligence beyond human capacity". It promises everyone a personal agent that "will work 24/7 on your behalf". It says small businesses "will continue to be the backbone of the economy, but each small business will be able to have a much larger impact."
Maybe so. I can only tell you what the thing does at my desk now, and what the Australian numbers say about desks like mine. The gap between those two pictures is worth looking at calmly, because policy is being written into it.
Most of what the AI does for me falls into four jobs.
It drafts. Give it a pile of notes and it returns something shaped like an article in less time than it takes me to reread the notes. The shape is usually fine. The judgement still has to be added, and that part is mine.
It checks. I ask it to read my work for gaps, a sum that doesn't add up, a claim with no source beside it. Sometimes the mistake it finds is mine. Last month I wrote that kids were still on social media "almost a year" after the under-16 ban came in. The ban started on 10 December 2025 and my essay was dated in the middle of September, so it was about nine months. The sentence felt right to me. The calendar disagreed, and the calendar wins.
It gets corrected. Last week the AI's own notes said one of my columns had been up on the site for days. It hadn't. The page had never been built, and the link had been returning an error for four days. The AI had trusted its notes over the live page. Once it was caught, it built the page, added a short note saying what had happened, and fixed its notes. The mistake had arrived in the same calm, tidy voice as everything else, and that's the real difficulty. Nothing in the note told me it was wrong. Now nothing gets called live until someone has loaded the page.
And it waits. It doesn't decide what goes out under my name. I do. That part of the day hasn't moved at all.
So an honest account of a working day with an AI is a loop. It drafts, I check, it checks, I correct, I decide. The time saved at the start is real. Some of it comes back at the end as checking.
Getting something from this? The List sends the next essay first, free, Sunday mornings.
The Australian Bureau of Statistics released its Characteristics of Australian Business survey for 2024-25 on 25 June 2026. It asked nearly 7,000 businesses about their use of AI. Around 12 per cent said they used it in the workplace. The rate rose with size: around 35 per cent of large businesses, 22 per cent of medium businesses, and around 11 per cent of small and micro businesses.
The National AI Centre's tracker, a monthly survey run for the Department of Industry, gives a bigger number. Its latest published quarterly overview covers December 2025 to February 2026. Across those three months, 43 per cent of small and medium businesses reported some level of AI adoption. When I checked the centre's live dashboard this week, its last-three-months view was lower, at about 38 per cent. Both tracker counts include businesses that only intend to start within the next 12 months, so the ABS and the tracker are measuring different things. Neither shows a country that has handed its work to machines.
The tracker says more about what businesses do with it. Among adopters, the most common uses were content generation and data analytics, at 54 per cent each. That is drafting and checking. It's my day, in survey form.
The reasons for holding back are just as ordinary. Around 65 per cent of businesses not planning to adopt cited distrust of AI decision-making or a strong preference to keep human control. Fifty-four per cent said AI wasn't relevant to their business.
The figure I keep coming back to is about safeguards. In the same December to February overview, the most common safeguard among small and medium businesses using AI was checking AI outputs before they affect customers. The centre puts that at approximately half of current users, though the live dashboard's recent view has it nearer 28 per cent. It doesn't say what the rest do instead. In a small business, the check is usually the owner reading the quote or the email before it goes out, often at night, after the paid work is done.
Jobs and Skills Australia looked at this across the whole labour market in August 2025. Its study found that generative AI "is likely to augment the way that we work rather than replace jobs through automation". Its commissioner, Professor Barney Glover, added that the technology "has the potential to displace people in some jobs, particularly administrative and clerical roles."
Both halves of that matter. The people most exposed are those whose jobs are mostly the first step of my loop, the sorting and the filling in. What protects my day is the last step, the decision. Plenty of jobs don't have that step written into them.
The Productivity Commission's final report on data and digital technology, released on 19 December 2025, put a number on the upside. It considers multifactor productivity gains above 2.3 per cent likely over the next decade. On its own rough estimate, that would mean about an extra $116 billion of GDP over ten years. It also says there is "considerable uncertainty" about the size of the effect, and it recommends AI-specific regulation only as a last resort.
The National AI Plan, published by the Department of Industry on 2 December 2025, takes a similar line. It builds on what it calls "strong existing, largely technology-neutral legal frameworks" and sets up an AI Safety Institute to "monitor, test and share information on emerging AI capabilities, risks and harms". The plan also says: "Every organisation developing and using AI is responsible for identifying and responding to AI harms".
That last sentence carries a lot of weight. Under a plan built on existing laws, the working safety system for most of the AI in Australian workplaces is a person reading the output before it goes anywhere. In a small business, where the ABS found adoption lowest, it is usually one person with three other jobs. The approach can work. It depends on that check actually happening, and the tracker's safeguards question is one of the few national measures I've seen of whether it does.
When something does slip through, the order in which people find out matters too. I wrote about that in They Called It Foresight.
The superintelligence essays are about what the technology might become. I can't test that from a desk in Surry Hills, and I'm wary of anyone who says they can, in either direction. What I can test is the day. On the day, the AI is quick, useful, and wrong often enough that I read everything twice. A person still decides.
That's the version of AI that Australian workers and small businesses are dealing with now, and it's the version Australian policy is built around. It deserves better numbers. The ABS says its 2026-27 survey will include more questions on how businesses are adopting and using AI. One of them could ask how much time goes into checking what it produces. I'd read that release closely.
New essays by email: ricoholt.com/list
Rico Holt · ricoholt.com
If this was worth your time, you can leave a tip. Rico Holt runs on readers, not big tech or big media money, and the essays stay free.
Leave a tip · A$5Get the next essay and the book launch price first.
Privacy: your email gets used for one thing, sending you Rico Holt essays and book news. It's stored with Netlify, our form host. Never sold, never shared. Want off? Reply to any email and you're gone.