You keep up with AI by checking two official pages once a week, testing one new thing against one real task, and ignoring everything else. You are not supposed to know about every launch. You're supposed to know about the ones that change your Tuesday.
Your feed will tell you otherwise. Every post says this tool changes everything, and that's the bit to ignore, because none of them know what your Tuesday looks like.
What should you ignore?
Ignore any launch that doesn't touch a task you actually do this month. That is most of them.
Here's what that looks like with real examples. On 20/10/2026, ChatGPT's release notes announced locked chats. On 21/10/2026, Claude's release notes announced a research preview of Claude Duet for Pro and Max plans. Both are real updates. If you're on a free plan, or you don't need a locked chat, neither one is your problem this week.
Ignoring something isn't falling behind. You've decided it doesn't fit yet. I covered that exact stance in AI strategy vs using AI: when a new tool drops, ask if it fits what you're building, and if not, leave it.
Three things I ignore on purpose:
- Rankings of "the 25 AI tools you need". Nobody needs 25.
- Anything that needs a new subscription before I've seen it work.
- Hot takes about which model is best, posted the day a model launches.
What's the 20-minute weekly check?
Pick one slot a week, the same one, and read the top of two pages. That's the whole habit.
The two pages are the release notes from the companies that make your tools:
- Claude release notes on Anthropic's help site, a dated log of product updates.
- ChatGPT release notes on OpenAI's help site, also dated.
Read them top down until you hit something you've already seen. Ask one question of each entry: does this touch something I do for clients or for my own business? If yes, write the name of it on a list. If no, move on.
You only need the page for the tool you use. If you live in Claude, skip the other one.
One more place worth your time is my YouTube channel. I test the new things on real business tasks, so you don't need to test all of them yourself. Watch the ones that match your list.
How do you test a new tool without wasting a day?
Test one new thing against one real task, once. Never test against a made-up task, because made-up tasks always look good.
Take something you actually have to do this week. A client email, a content plan, a summary of a call. Do it the way you normally do it. Then do it with the new feature. Compare the two.
Three questions, written down:
- Was the output better, or just different?
- How much did I have to fix?
- Would I use this again next week without anyone reminding me?
If the answer to the third question is no, you're finished. You tested it, it didn't earn a place, and you don't have to think about it again.
When should you actually switch tools?
Rarely. Switch when your current tool fails the same real task more than once and the new one handles it, not when a launch looks exciting.
My real example is Make and n8n. I used Make first and moved to n8n for one reason: when something broke, I found it much harder to troubleshoot in Make. That's a repeated failure on a real job. It wasn't a launch, and it wasn't a thumbnail. I wrote about it in the stack I use every day, including the tools I tried and dropped.
Before you switch, check you can take your setup with you. Save your instructions and example outputs as plain text first. If you can't move your work across in an afternoon, the switch costs more than it looks.
And one confession: I will happily sign up for AI tools on free trials, lose track, and get charged. I've said it publicly before. So my honest tip is to put a cancel reminder in your calendar the minute you start a trial. I don't always follow it. You can.
Quick answers
How often should I check for AI updates?
Once a week is plenty for most small businesses. Both release-notes pages above are dated logs, so you can see at a glance what has changed since your last look.
Do I need to learn every new AI feature?
No. You need to know the features that touch a task you do now. Everything else can sit on a "maybe later" list that you read once a month and mostly delete.
Is it a problem if I'm still on a free plan?
No. Plenty of updates are labelled for paid plans only, such as the Claude Duet preview for Pro and Max. Those are safe to skip until you have a reason to upgrade. If you want somewhere to start, can you learn AI for free covers the free path.
When keeping up is easier with company
The hard part of this system isn't the reading. It's testing something on a real task and having someone to ask when the result looks odd. That's what Wright Mode is built for: a monthly live build where we test something properly, fortnightly co-working and Q&A for your own tasks, and a private community that answers in writing.
