AI Insights & Resources

Microsoft retires the COPILOT() worksheet function on September 14. If you built it into live spreadsheets, the lesson is bigger than one formula.

New research finds AI is driving technical hiring up while early-career roles get squeezed. AI absorbed the work juniors used to learn on. That has a cost.

Most agent security advice tells you the risks. This is the containment answer: limit what an agent can run, what it can open, and where it can connect.

A 42-minute AI outage is a small event with a useful question attached. Most businesses cannot say which of their work halts when a provider goes offline.

OpenAI reportedly folded its preparedness team into other groups while METR raised $71M to test AI independently. Self-policing is thinning. Buyers should notice.

Stripe is reportedly paying over $7B for an AI gateway. That layer sits between you and the models, and it is where cost control and switching leverage live.

AI is automating AI research before it automates most jobs, and that is not a coincidence. Four traits predict which work in your business gets automated first.

New AI video tools accept dozens of reference images, so your product stays consistent across every shot. Consistency, not capability, was always the real blocker.

You cannot outbid the labs for AI researchers, and you do not have to. Co-op terms, capstone projects, and Mitacs put university AI talent inside small businesses.

ChatGPT Plus coûte environ 28 $ CAD/mois. Tous les forfaits ChatGPT en dollars canadiens — Gratuit, Go, Plus, Pro, Business et Enterprise — plus les frais de change que les guides oublient.

Combien coûte un consultant en IA au Québec (3 500 $ à 25 000 $ pour un mandat PME), comment le choisir, et pourquoi la conformité à la Loi 25 doit être intégrée dès la conception.

A new tool is replacing version control for code AI writes. The deeper problem applies everywhere: your review and audit habits assume a person made each change.

Pika launched an audio product priced at a fraction of the market leader. What a small business can newly afford to narrate, and where a real voice still wins.

AI assistants are arriving pre-installed on new hardware. Your AI policy probably covers tools people choose, not the ones that come in the box.

A frontier model now runs at 750 tokens per second on specialist hardware. Speed is not a vanity metric, it decides which AI products are possible at all.

Perplexity beat OpenAI and Anthropic on four of five search benchmarks. When AI looks things up for you, retrieval quality matters more than conversational polish.

One agent vendor now ships a tool measuring whether its agents actually make teams faster. The burden of proof is shifting, and buyers should push it further.

OpenAI shipped a feature that observes your computer to build context automatically. Genuinely useful, and a privacy decision your business should make deliberately.

New AI tools turn hand-drawn sketches into finished product campaigns. What it means for small retailers when the gap between idea and asset nearly disappears.

Gemini 3.7 Flash shipped three weeks after 3.6, twice as good at half the price. How to set an upgrade policy so you are not chasing releases or stuck on stale ones.

A DeepMind system doubled sign language translation accuracy. Accessible AI is quietly becoming practical, and for Canadian businesses it is also a legal obligation.

Head-to-head testing found a top model producing three times the words at lower complexity. Why AI output bloat costs you money and attention, and how to stop it.

New tools let you package and share an entire AI setup. That turns the prompts and context your team builds into a real asset, and one you are probably losing.

Researchers found exposed AI reasoning traces containing 62 API keys and 33 passwords. The hidden thinking your AI does is data, and it can leak like any other.