Your work became searchable.
Choose its home.

What actually changed

• A search box in the ChatGPT sidebar searches across chats, Projects, images, and documents at once, with filters to narrow by type. Click a result and it opens.
• You can pin projects, not just chats, and group recent chats by project.
• It is live on web, iOS, and Android, on all plans globally, including free.
• Claude had chat search and memory on paid plans, but it worked differently. You had to ask Claude to look, and it did not check past conversations unless you told it to.

Try this: Find one thing instead of rebuilding it

• Think of something you know you made in ChatGPT and would rather not make again. Choose something specific, not ‘my good prompts.’
• Search for it by what it was about, not by what you named the chat.
• If you find it, pin it or move it into a project so you do not lose it again.

What I found when I tried it

I ran this on my own account before writing it up. Searching across chats and projects at once is the part I love. One honest note so you set your expectations right: you still have to walk in with some idea of what you are looking for, and what comes back is the chat or project where the thing lives, not the exact answer pulled out for you. It gets you to the right conversation fast. You scroll the last stretch yourself.

My take

A search box over everything you have made is more useful than it sounds.

Use it for one specific thing first. Think of something you know you made and would rather not rebuild: a client summary, a launch outline, or a set of prompts you liked. If it comes back in ten seconds, you just learned your account has more finished work in it than you thought.

One caution. Being able to search something is not the same as having a copy of it. Search is a feature, and features change. If a document matters to your business, keep it somewhere you control and let the search box be for finding things, not storing them.

What actually changed

• beehiiv launched Community on July 16, plus an AI assistant called Copilot, a redesigned editor, and programmatic ads. Its framing was that creators should not have to juggle five or six platforms.
• Google renamed NotebookLM to Gemini Notebook on July 16 and pulled it closer to the Gemini brand. Notebooks moved into the Gemini app, with Search integration next. Nothing you built moved or broke. The name changed.
• Canva put website and app building inside the design tool.
• ChatGPT made your entire history searchable.
• Most solo businesses were running somewhere between five and seven tools, and the direction of travel was toward fewer.

Think about this

• Am I trying to consolidate the number of tools I pay for, or the number of places I have to look for things? Those have different solutions.
• If one of these platforms changed its pricing or layout next month, what of mine would be stuck inside it?

What actually changed

• After a week of complaints, OpenAI reversed most of the redesign on July 16 and 17. ChatGPT became the main screen again, chats and Projects returned to the sidebar, and Chat and Work history synced across web, mobile, and desktop.
• The fix was live on all plans, but the desktop app needed an update.
• My correction: Work starts at Plus, $20 a month. Free and Go users could not test it. If you went looking on a free plan, it was not there, and that was my error, not yours.

Try this only if you are on Plus or above

• Update the desktop app first, so you are not testing the broken version.
• Give it one small real deliverable, such as turning five bullet points into a one-page client summary.
• Decide whether this is your execution layer or Claude's. You still do not need both.

My take

Some people tested Work and then could not find what they made because the desktop app buried their history. Others could not test it because I got the plan wrong.

Neither of those was you. Enough people complained that OpenAI changed the app back, and the plan mistake was mine.

When a tool is confusing, most of the women I work with do not decide the tool is badly built. They decide they are behind, quietly stop using it, and do not mention it. If that was you, the tool was the problem. Try it again or do not, but decide on the merits.

What actually changed

• Canva rolled Code 2.0 out to every user on July 15 across free, pro, business, education, and enterprise.
• Describe what you want in plain language and it builds an interactive site, app, landing page, or form that you can edit like another Canva design.
• HTML import lets you bring in something built elsewhere, including output from Claude or ChatGPT, and edit the design without rebuilding it.
• Free means access, not unlimited use. AI credits still scale with the plan.

Try this: Rescue something you already built

• Find something you made with AI that works but looks generic: an artifact, a simple page, or a form.
• Import the HTML into Canva Code and apply your brand kit.
• Stop there. Do not redesign your whole site because you are already in the tool.

What I found when I tried it

I spent time in it before recommending it, and it is good. The thing that decides how useful it is turned out to be my setup, not the tool. Code designs from the context you hand it. If your Canva brand kit is thin, you get generic output and end up describing your brand from scratch. Mine could be built out better, and I could feel that in the results. Build the kit first, then let Code lean on it.

My take

I build tools in Claude Code, so I will be straight about this. Canva Code is a good on-ramp. It is not where you will end up if you keep going.

Most readers do not need a development environment. They need one working page for one offer, and they need it to look like their brand without a rebuild. Canva Code does that, including on the free plan.

The part I would use is HTML import. If you built something in Claude that works but looks generic, bring it into Canva and fix the design without starting over. Keep a copy of any business-critical file outside Canva.

What mattered

• The original deadline was June 22, then July 7, July 12, and July 19.
• After the included period, continued use moved to prepaid credits at $10 per million input tokens and $50 per million output tokens.
• Anthropic said it intended to bring the model back into subscriptions when it had the compute.

Think about this

• Is anything in my week actually running on Fable 5 specifically, or am I only aware of it?
• If a tool doubled in cost tomorrow, would I know which parts of my work depend on it?

My take

I gave you Anthropic's published deadline, and then Anthropic moved it again. I did not file that as my correction. I filed it as information about how the company was operating.

If you built a process around one model because of what it cost in June, you received several different answers in five weeks. Keep the work portable enough that a pricing change is annoying instead of expensive. For most of you, this was not your bill.

What actually changed

• Google added personal avatars to Google Vids on July 16. Record one selfie and a short voice sample, type a script, and the avatar delivers it.
• Gemini Omni also came to Vids, so you can generate clips from a written prompt and edit them by chatting.
• Generated clips carry an invisible SynthID watermark, and an avatar is locked to its account holder.
• It was available on Google AI Pro and Ultra and Workspace business tiers, in English, for people 18 and over, and not in the UK, Switzerland, or the EEA. Scheduled Release accounts were not due to see it until August 5.

Think about this

• What is actually stopping the video from getting made: the recording, or the being-seen part? An avatar helps with one of those.
• Which videos are genuinely repetitive, and which ones are people watching because it is you?

My take

For many people I work with, the reason there is no video is not the camera, lighting, or time. It is that being visible is uncomfortable. An avatar solves a scheduling problem. It does not solve that one, and if you use it to skip the discomfort, the videos may feel like nobody is in them.

Where it can help is repetitive work: course module intros, a welcome clip you would otherwise re-record with every offer update, or short explainers you already wrote out.

Every clip carries an invisible watermark, so platforms will know it is AI whether or not you say so. I would say so.