#14 😀Test test test test

AI News

Top posts from this week

🛠️ Claude now works inside Google Docs, Sheets, and Slides, and one setting decides whether it edits without asking.

Anthropic's Claude, the AI assistant, is in public beta on all paid plans as a sidebar add-on that can read the file you have open and edit it directly. Our walkthrough by Francois Buys explains that the default "Ask before edits" mode shows an approval card for each change, while "Accept all edits" applies them without stopping. Before anyone turns it on, decide which documents keep approval on (he suggests contracts and client-facing reports) and pilot with a small group, which the post says you can set up from the Google Admin console.

🛠️ Ask your AI team this at the next check-in: "What's the error rate right now, and how has it changed?"

Sara wrote a buyer's guide to AI projects that explains why an update saying the system got something wrong can be a healthy sign, as long as it comes with a number, an explanation, and a next step. Vague reassurance like "it's working pretty well" is one of the warning signs. Read it for the three things to ask for at a milestone: the current error rate on a representative sample, one or two recent examples of what the system got wrong, and what changed as a result.

🛠️ Three questions to ask before handing a repetitive task to AI

Francois explains that a lightweight AI setup suits a repeating task, meaning written instructions for Claude (the AI assistant) plus a read-only connection to the systems that hold your data, when you can describe every step, name every data source upfront, and are happy to have a person involved along the way. His team used it to cut a monthly client report to less than half its previous time. Read it for the three questions to ask about a task in your own business and the signs that you need something heavier.

🛠️ Free accessibility scanners miss most website problems, so who checks the rest?

Sara's guide to WCAG 2.2, the Web Content Accessibility Guidelines that many accessibility laws point to, says automated checkers like WAVE and Lighthouse (built into Chrome) typically catch only 30 to 40% of issues, so testing by hand with a screen reader and keyboard is still needed. It names Level AA as the usual target for commercial websites. Ask whoever runs your site which level they test against and whether a person checks by hand, and use the guide's list of free tools and project checklist to follow up.

📊 Students, not schools, seem to be driving AI adoption in education.

Amanda came back from South by Southwest EDU, an education conference in Austin, Texas, with an observation from talking to companies there: students find AI tools, use them, and then bring them to their teachers and institutions. Her recap covers the concerns raised in her conversations, including transparency, data privacy and security, and a disconnect between institutions and students. If you run a school or training program, ask your staff which AI tools students already use, then read it for those three concerns and the expo gap she saw in tools for university administrative tasks.

Do you know anyone who would love to receive this newsletter? Tell them to subscribe here 👇 and catch the latest issue right in their inbox.