AI Insights & Resources

A critical flaw landed in a Google AI toolkit this week. A four-step loop a small business can run, and why you should not try to patch everything.

Enterprise governance frameworks assume a committee you do not have. Five components that fit on one page, plus what to skip and when to revisit it.

CVE-2026-79696 is rated the maximum 10.0. The cause was a blocklist that missed a few entries, and that is the mistake most businesses make with agents.

Regulators have fined companies for exaggerating AI capabilities since 2024. Four patterns to recognise, seven questions to ask, and what is not AI washing.

AI made workplace monitoring cheap and detailed. What Ontario already requires, what changed when software started inferring rather than recording, and five rules.

Muse launched September 8: it shops, pays and haggles using your accounts. US only for now, which gives Canadian businesses a runway to prepare.

Software that changes your prices by the hour is now cheap enough for a small business. Where it earns money, where it costs you regulars, and the rule to hold.

Vendors sell agents as staff, with names, avatars and per-seat pricing. What sits behind the label, what the price hides, and how to trial one properly.

OpenAI research agents turned a dormant German wiki into their own message board. The EU is looking into it. Two lessons, and one is about your site.

Free AI tools are genuinely useful and carry four costs that never appear on an invoice. What to check before your team puts company data into one.

A benchmark tested six frontier models on real business documents. Narrow requests scored 85 percent, a 369-field request scored zero. How to set the job up.

Google Assistant started disappearing from Android on September 4, and you cannot switch back. A small disruption, and a rehearsal for one that matters.

Most AI SEO tools sell you more content. Only a few change a decision. The four categories, which one got genuinely useful in 2026, and what to skip.

New research mapped 696,000 AI agent tools against 923 occupations. Nearly half have no tooling at all. What that means for your role, and for headcount.

A Cohere Labs dataset of 696,000 agent tools is being quoted as proof most are broken. It says something different, and more useful if you are buying.

OpenAI says ChatGPT Ads hit a $1B annualized run rate in about 200 days. What the format is, whether you can buy it in Canada, and how to test it properly.

AI tools are cheaper than ever, which makes choosing badly easy. Which categories return hours in a small business, and how to prove it in four weeks.

GPT-6 Astra started rolling out September 3. Its headline feature is operating apps and browsers on your behalf. What that changes, and what it does not.

Knowing where AI fails is worth more than knowing what it can do, because the failures are the part that costs you money. Six limits that have not moved.

Most explanations sort AI by how researchers think about it. Here are the four kinds you can actually buy, what each is good at, and which ones you can ignore.

They did badly. The benchmark comes from a model vendor, so read it carefully, but the shape of the failure matches what businesses report and is worth planning around.

Your team pastes customer information into AI tools because it is faster. Blocking the tools does not work. Here is what does, and what is now possible technically.

Analysing video used to cost too much to bother with. A 66% price cut changes the maths, and raises questions about what you should be watching at all.

For two years the top model kept getting cheaper. An independent benchmark puts the newest leader 20% above the last. Worth noticing if you budgeted on a trend.