What can you check?

1. Count the responses behind a survey finding

Google’s November 5 Google Forms announcement introduced a tool that writes and runs code to compare answers across survey questions, then suggests next steps. The rollout began November 2 for Rapid Release accounts. Scheduled Release is due to start November 19. Google lists Business Standard and Plus among the eligible plans, so Canadian teams on those plans should check their rollout settings before counting on access.

Illustrative scenario. An Ontario training provider asks attendees about session timing and course format. The tool could flag that people who wanted evening sessions also preferred online delivery. That gives the coordinator a place to start. If the pattern came from only four respondents, though, it may be a poor basis for moving an entire course online. The tool can summarize the replies it received. It cannot tell you what people who skipped the survey would have said.

First move this week. Pick one past survey cleared for use in the tool. Ask it for one comparison, then check the counts and the number of people who answered both questions. If your account is still waiting for access, record that manual comparison as your benchmark.

2. Ask what running AI in Canada would change

The federal government’s November 5 budget outlines a plan to give Canadian AI companies access to 500 megawatts of domestic computing capacity. The budget chapter on sovereign AI infrastructure names tools such as financial backstops and purchase commitments to help finance capacity. The figure describes power capacity. The announcement leaves the delivery timetable, individual customer access and pricing open.

Illustrative scenario. A Manitoba software company is choosing where to run a service that reads freight documents. More Canadian capacity could eventually give its vendors more hosting options. The tradeoff today is that narrowing the shortlist by location can shrink the range of models or suppliers the team can test. A Canadian processing location also leaves separate questions about where backups go and who can access support logs.

First move this week. Choose one workflow and write down where its inputs, outputs and backups may be kept. Ask your current supplier to answer for that workflow in writing. Keep the answer with your renewal documents so you can compare it with future options.

3. Measure the work after a model changes

On November 5, METR’s independent assessment of Anthropic’s research estimated that Anthropic’s models were improving on well-specified tasks at about twice their earlier pace. The researchers think AI assistance explains much of that acceleration, with significant uncertainty. Anthropic redacted much of the evidence the researchers found most informative. The estimate describes one lab’s model development and leaves open how much faster ordinary business work might get.

Illustrative scenario. A Canadian wholesale distributor uses an assistant to draft replies about returns. A newer model might handle exceptions better, giving staff fewer drafts to rewrite. It could still miss a credit limit or read a French request differently from an English one. The team has to account for the time it takes to catch those errors alongside the time it saves on drafting.

First move this week. Save ten anonymized requests with approved answers, including the kinds of exceptions that reach a supervisor. At the next model change, compare both versions against those answers and time the review. Keep the version that makes the full task easier to complete accurately.

The bigger pattern

Each decision gets easier when you keep something concrete to check. Keep the response count behind the summary, the supplier’s written answer on data location, or the approved replies you can use to test a model update.

What’s one AI-generated result your team had to redo, and what check would have caught the problem before handoff?