Practical advice and real-world insights on implementing AI workflows that deliver measurable business value.
On 30 June Anthropic released Claude Sonnet 5. Not the flagship. The mid-tier workhorse that most businesses actually run their day-to-day automations on. It's better at the agentic work that matters, and it's still cheap: $2 per million input tokens through 31 August. Most owners have no idea which model their automations are running on. Here's the ten-minute check, and why the mid-tier is the tier that matters to a small business.
On 1 July Google put Gemini Spark on the Mac, and for the first time it can act on the files sitting on your own computer, not only the ones in the cloud. For an accountancy practice, a law firm, an agency, this is the moment the AI moves from the chat window onto your desktop. Useful, and risky in a specific way. Here's the governance conversation to have before you switch it on.
On 22 June Google released Gemini 2.5 Pro with Deep Think — by several accounts the most capable public model going, with a two-million-token context window. The benchmark isn't the story for a small online retailer. The story is that buyers are starting to hand the shopping to an AI agent, and most catalogues weren't written for a machine to read. Here's the gap, and the unglamorous fix that gets you back on the shortlist.
For a year, the advice to anyone touching the EU market has been "get ready for 2 August 2026." That date is now two different things. The enforcement of general-purpose AI rules arrives as planned. The high-risk obligations most small businesses feared have been pushed to December 2027 under the Digital Omnibus. Here's the ten-minute self-check to find out which side of that line you're actually on.
On 13 June Chinese lab Zhipu released GLM-5.2 — an open-weight model that scores just behind the best closed model on serious coding benchmarks and ships with a licence to run it on your own hardware. For a business in finance, law, accountancy or healthcare, the headline isn't the benchmark. It's that near-frontier AI no longer has to live in someone else's cloud.
This month Anthropic put its managed agents on a timer. The question for a small business is which jobs want doing on their own and which absolutely should not. Here's the line I draw, and the three jobs I'd put on the night shift first.
On 12 June, OpenAI retired GPT-5.2 mid-conversation and moved everyone onto GPT-5.5. For casual users it was invisible. For businesses with workflows tuned to a specific model, it was a reminder: the model is rented, not owned, and the landlord can change the locks. The one rule I now design every client workflow around, and the ten-minute check to find out whether the next retirement would cost you a morning or a month.
Anthropic released Claude Fable 5 — the most expensive mainstream model on the market, double the price of Opus 4.8. Mythos-class capability is included free on paid Claude plans until 22 June. Here's the test I'm running to find out whether the dearest model in the world is worth it on your work.
OpenAI is redesigning ChatGPT into a "super app" with third-party apps and agents built in — but the new integrations aren't live in the UK, EEA or Switzerland yet. That's a few months of warning, which is rare. Here's the destination-or-task audit I'd use that head start to run.
On 4 June, OpenAI started rolling out "dreaming" — a memory upgrade that revises what ChatGPT knows about you in the background. Factual recall has climbed from 41.5% to 82.8%. It's a better product, and it changes the data-hygiene maths for any business whose team pastes customer details into the chat. The one rule I gave every client this week.
Realtor.com just put an AI concierge inside the portal. Rightmove and Zoopla read the same press releases. For a small UK agency that has lived on portal leads for two decades, the question is what's left to own when the portal becomes the adviser — more than you'd fear, but not the bits you might assume.
At Google I/O on 19 May, Google shipped Gemini Spark (a 24/7 cloud agent that will start making purchases on your behalf) and Daily Brief (a quieter overnight digest of your inbox and calendar). One I'd switch on for a client tomorrow. The other I'd keep at arm's length until I've watched it behave for a quarter.
OutSystems surveyed 1,886 IT leaders and found 94% worry about AI sprawl. The SME version is quieter but worse in proportion. I audited my own stack, cut £125 a month, and here are the three rules I'm now using with clients.
Anthropic launched Claude Design on Friday. It won't replace a senior designer thinking about your whole brand. It will flatten the fee you pay for the small, fast, "we need something by Thursday" work — the tier most SMEs spend five figures a year on and barely notice.
OpenAI beat humans at office work, Anthropic leaked an always-on agent, Apple confirmed Siri is getting Gemini, and $122 billion got raised. Most of it doesn't matter to your business yet. Three things do.
Apple is replacing Siri with Google's Gemini AI. 2.2 billion devices are about to get an assistant that can book, buy, and recommend — without ever opening a browser. If your business isn't structured for AI discovery, you're about to become invisible.
GPT-5.4 just scored higher than humans at real office work. Anthropic leaked an always-on AI agent. For SMEs, an agent costing a few hundred quid a month can now handle work that used to require hiring someone.
Agentic AI doesn't just answer questions — it takes action. I built an agent that monitors my clients' live deployments, investigates failures, and messages me when something breaks. That's not a chatbot.
The automations that actually move the needle aren't the impressive-sounding ones. They're the ones that remove human steps entirely. Here are the five that keep delivering.
Most small businesses should buy off-the-shelf where they can, build custom where they must, and use a hybrid approach for everything in between. Here's the honest version of the build-vs-buy decision.