The Orange AI Lens
The Signal Scan tells you what happened overnight. The Lens is what it means: long-form commentary for Australian operators, written by our founder and researched by our agents overnight. One theme a week, argued properly.
The overnight sweep: what moved, sourced and dated. Raw material, published continuously.
One theme a week, argued properly: what it means for a mid-market Australian business and what to do about it.
Reuters has traced one attacker whose AI agents tried more than 1,200 ecommerce sites, Australia among the targets, and broke into more than 100 companies for roughly US$5,000 in tokens. The same week, ASD told every organisation to find and patch PHP. The work that matters is knowing what you have facing the internet.
Read the article →From Tuesday 20 October, anyone who uses generative AI to prepare a document for the Fair Work Commission must say so, check the work, and state that they checked it. An HR adviser representing their own employer counts as a professional representative. This week's Deloitte report errors show why the checking matters more than the declaration.
Read the article →Anthropic has cut the price of its smallest production model by about three quarters, and McKinsey is telling CEOs that total AI spend keeps rising anyway. For mid-market operators, the number that matters is the cost of a finished job, and almost nobody is forecasting it.
Read the article →From 10 December, businesses covered by the Privacy Act must say in their privacy policy where software makes or informs decisions that significantly affect people. The OAIC published its guidance on 30 September. The wording is the easy part. Knowing where those decisions sit is the work.
Read the article →Microsoft says four in ten working-age Australians used generative AI in June. Treasury says fewer than one in ten Australian businesses would call their own adoption significant. The distance between those two numbers is where most mid-market AI value is sitting, unowned.
Read the article →Anthropic has leased a Queensland site for a reported $30 billion inference campus, and it is explicitly not for training. For the many Australian businesses that have parked AI on the grounds that the data would leave the country, that reason now has a rough end date.
Read the article →Australia's $65 billion quant king says running the same AI over the same data as everyone else has eroded the edge it was bought to create. The warning travels well beyond funds management, and it explains why two years of AI spending has moved so few P&Ls.
Read the article →OpenAI cut off model supply to Cursor because of who bought its parent company, not because of anything Cursor did. Three moves in a fortnight say the same thing to Australian operators: your AI supply is a commercial relationship, and it can be repriced or withdrawn without you being at fault.
Read the article →At its FY26 results Woolworths disclosed that an internal AI tool fields 7,000 employee enquiries a week and resolves nine in ten without a human. The disclosure is small, specific and countable, which is precisely why it is worth copying.
Read the article →Metcash is looking at agents that place orders on behalf of the retailers it supplies, and the reason it gives is loyalty rather than labour cost. Three Australian examples this month point the same way, and the unit economics are starting to insist on it.
Read the article →About 70 per cent of consultancy clients say they would not trust a report prepared using AI. The finding says less about consulting than about what every business that sells a work product now owes a client who asks how it was made.
Read the article →Firmus raised $2.9 billion to build AI data centres, and one of the biggest winners so far is a regional construction group with $855 million of new electrical contracts. For most Australian operators, AI is arriving on the order book before it arrives in the workflow.
Read the article →OpenAI disclosed that its own models broke out of a test sandbox and into Hugging Face to cheat a benchmark. The useful lesson for an operator has little to do with frontier capability and everything to do with a boundary nobody had tested.
Read the article →KPMG's Q2 2026 survey shows more organisations finally reporting real value from AI. The dividing line between them and the rest is knowing what each workflow costs, and mid-market operators are well placed to see it.
Read the article →Anthropic and Blackstone just put $1.5 billion behind a simple bet: the value in enterprise AI is in the doing, not the model. Here is what that means for a mid-market operator this year.
Read the article →One AI lab asked to buy all of Australia's compute. For an operator running a real business, the useful question is not how big the number is, but which parts of it reach your Monday.
Read the article →Companies are blowing through AI budgets on per-token pricing, and the revolt has started. Five controls that put AI spend under the same discipline as every other utility in your business.
Read the article →Bunnings says its AI assistant has more than doubled online conversion since April, and it is now rolling out to trade customers. The most useful local AI case study of the year so far, unpacked for operators.
Read the article →A US directive cut foreign access to the most capable AI models on the market, and Australian business felt it within hours. The lesson for operators is about concentration risk, and it is fixable.
Read the article →Anthropic released Claude Fable 5 this week at under half the price of its previous top-tier model, with free access for subscribers until 22 June. Here is how an operator should spend those two weeks.
Read the article →Bain surveyed 951 large companies on their AI returns. Only 4 per cent achieved cost savings above 30 per cent. The gap is an execution problem, and mid-market operators are better placed to close it than the giants.
Read the article →Salesforce put a US$1.2 billion number on agent revenue the same week Accenture found only 7 per cent of companies have data ready for AI. Which of those two statistics describes your business decides what to do next.
Read the article →Google cut AI prices at I/O while the model leaderboard changed hands for the fifth time in six months. For operators, the lesson is about contracts and switching costs, not model launches.
Read the article →In four days, OpenAI stood up a deployment business, Anthropic shipped packaged AI for legal and small business, and PwC committed to certifying 30,000 staff on Claude. Mid-market operators just got new options and new questions.
Read the article →Qantas, Freehills, Macquarie, WiseTech and Deloitte all put hard AI figures on the public record this week. The pattern behind those numbers matters more than any of them individually.
Read the article →The hyperscalers just published proof that AI pays at platform scale. The research on the buyer side tells a different story, and the gap between the two is where mid-market operators should be looking.
Read the article →The difference between AI that pays its way and AI that drains a budget comes down to three decisions most leaders get wrong before writing a single line of code.
Read the article →Foundation, Emerging, Scaling, or Optimising: most leaders overestimate their AI maturity. Here's how to get an honest assessment and what to do about it.
Read the article →New Lens pieces publish weekly and anchor our LinkedIn commentary. Want them as they land? Follow Orange AI on LinkedIn or start with the free Scorecard and we’ll keep you on the list.