Suprvisr AI Insights - November 17 2025

Hi from the land of double-doubles and double-checking AI outputs. This week almost every company uses AI, almost nobody is happy with the results, Canada quietly tightens the rules, and your web dev tools started asking if they can just build the whole app for you instead of watching you struggle.

The Week in One Number: 97%

That is the share of enterprises now using AI, according to a fresh Zapier survey of large U.S. companies. Nearly everyone has an AI strategy slide in the deck; only about half say the benefits show up across the whole org. Canada's adoption is much more hesitant, as we noted a couple weeks ago.

McKinsey backs this up with its 2025 State of AI report: adoption is basically universal, but most firms are still stuck in autopilot mode, with only a small elite seeing real bottom-line impact. Think of it as buying a Peloton but using it as a coat rack.

Why it matters: If your company "uses AI" but cannot point to fewer hours, fewer errors, or more revenue, you are not behind on models -- you are behind on execution, governance, and measurement.

See the survey ->

Pilot Purgatory & the Agentic Upgrade

While the average enterprise is running endless AI experiments, a smaller group is quietly leveling up into the agentic AI era -- where AI does not just autocomplete tasks, it plans and executes multi-step workflows end-to-end.

Zapier's survey says nearly all enterprises use AI yet struggle to integrate tools, measure ROI, and scale wins beyond a few teams. McKinsey finds only a small high-performer club converting AI into 5%+ EBIT bumps by treating it as business transformation, not a toy project. ISG's latest research shows agentic AI projects rising fast -- but only where companies invest in data architecture, governance frameworks, and talent, not just licenses.

Leader move: Stop asking "Which model is best?" and start asking "Which outcomes do we want to own, and what data + guardrails + workflows do we need so an agent can run them safely?"

Nerd out on the data ->

Canada Starts Reading the Fine Print on AI

Ontario's Working for Workers Four Act means that, starting January 1, 2026, many employers using AI to screen or assess candidates will have to say so in the job posting. No more mystery bots scanning resumes in the dark.

Zooming out, Wolseley Law points out that Canada's AI rules are still a patchwork of sector laws, privacy rules, and employment standards -- but that patchwork is getting tighter. Ottawa just launched an AI Strategy Task Force plus a national consultation sprint to shape the next federal AI strategy.

What this means for you: If you are using AI for hiring, performance, or employee monitoring, you are no longer in move-fast-and-break-HR territory. You are in document, disclose, and defend territory.

See the Canadian context ->

When AI Goes Off the Rails (a.k.a. Why Guardrails Exist)

If you have ever felt guilty about fact-checking an AI summary, do not. A new BBC/European Broadcasting Union study found that 45% of AI-generated news answers contained significant errors, with Gemini struggling the most and racking up sourcing problems. Great if you like chaos, less great if you are generating press releases or investor updates.

On the human risk side, OpenAI now faces lawsuits, including one from Ontario recruiter Allan Brooks, who alleges that design changes turned ChatGPT into a manipulative product that helped push him into a mental health crisis. Another high-profile case, Raine v. OpenAI, claims the chatbot played a role in a California teen's suicide.

Leader move: Treat safety, escalation, and red lines as product requirements, not nice-to-have checkboxes. Halfway guardrails are no guardrails.

See the error stats ->

Is SaaS Dead, or Just Being Overly Dramatic?

Agentic AI in IDEs and platforms can now take a plain-English request like "build us an internal lead dashboard that talks to Salesforce and deploy it" -- and then actually do that. No wonder investors (and Microsoft's CEO) are wondering whether classic pay-per-seat SaaS is about to get squeezed.

If your in-house dev agent can ship a custom tool in an afternoon, the old "it is faster to buy software" argument starts wobbling. For businesses, this is huge: it has never been easier to build lightweight, internal, exactly-what-we-need tools instead of stitching together subscriptions like a Franken-stack.

Leader lens: Treat agentic dev tooling as a governance challenge, not just a productivity boost. You still need data policies, audit logs, and clear owners when AI ships code.

Full breakdown ->

Leadership Takeaway

AI is no longer the experiment -- it is the infrastructure. The companies that win will not just have the flashiest model; they will have the boring-but-critical stuff nailed: clean data, clear guardrails, and teams that know when to trust the agent and when to say, "Yeah, we are going to double-check that."

If you want help getting from "we use AI somewhere" to "we can prove how AI is helping our people and our P&L," that is literally the job. Hit reply or book a chat, and we will sketch your governance-first roadmap over something caffeinated.

Talk to Suprvisr ->

Governance-first AI for Canadian leaders

Produced by Suprvisr AI -- Workplace AI for humans, with guardrails. You are receiving this because you subscribed to Suprvisr AI Insights. Data stays in Canada -- just how we like our AI and our maple syrup.

Book a 20-min intro