OpenAI has updated Scheduled Tasks in ChatGPT with a dedicated Scheduled page, improved task creation and editing, more flexible timing options, and better notifications.

That means ChatGPT is becoming a little less reactive and a little more proactive.

Instead of opening a chat every time you need something, you can schedule one-off or recurring prompts. You can also ask ChatGPT to check for changes and notify you when something meaningful happens.

This is not the flashiest AI update of the week, but it might be one of the most useful for everyday work.

Try setting up one recurring task this week.

A few practical examples:

Start simple. The goal is not to automate your whole business in one go. It’s to take one recurring task out of your head.

Perplexity has introduced Brain, a new memory system for its Computer agent.

Most AI memory today is about remembering the person: your preferences, your writing style, your role, maybe your favourite format. Brain is different because it focuses on remembering the work itself.

It can keep track of previous tasks, files, decisions, sources, corrections, failed attempts, and successful workflows. Perplexity says this gets organized into a context graph so the agent can start future tasks with better context instead of beginning from zero every time.

The company says early testing showed improvements in answer accuracy, recall, and cost on tasks that depend on historical context.

This is one of the more important agent stories because it addresses a very real problem.

A lot of AI tools feel impressive once, then frustrating the second or third time because they forget what already happened. You end up re-explaining the project, uploading the same files, correcting the same assumptions, and rebuilding context over and over again.

For AI agents to become genuinely useful in business, they need more than a good model. They need memory, permissions, source tracking, and a clear understanding of the work history.

That is what makes this interesting. The next step in AI agents may not be a bigger chatbot. It may be a system that knows what has already been tried, what worked, what failed, and where the source material came from.

Brain is rolling out in research preview for Perplexity Max and Enterprise Max users.

If you test it, do not just ask it random questions. Give it recurring work.

Try things like:

The question to ask is: does it actually get better the second, third, and fourth time?

Adobe is rolling out AI assistants across Photoshop, Premiere, Illustrator, InDesign, and Frame.io.

The important part is that these are not just generic chatbots sitting beside the app. Each assistant is designed around the specific workflow of the tool it lives in.

In Photoshop, that could mean organizing layers, swapping backgrounds, resizing assets, or making repeated edits across a design. In Premiere, it could mean organizing a timeline, renaming clips, marking moments in footage, or helping with repetitive editing tasks.

The idea is simple: instead of digging through menus or manually repeating steps, you describe the outcome you want and the assistant helps execute the workflow.

If you are already using Adobe Creative Cloud, watch for the beta features inside Photoshop, Premiere, Illustrator, InDesign, and Frame.io.

Good first tests:

The practical test is simple: does it save production time without taking creative control away from you?

Japanese AI startup Sakana AI released Fugu, a system designed to behave like one model while coordinating multiple models behind the scenes.

Instead of forcing users to choose one AI model for every task, Fugu can decide whether to answer directly or bring in a group of specialist models. It handles model selection, delegation, verification, and synthesis through a single API.

Sakana is positioning this as a way to get strong performance without relying on a single frontier AI provider.

That timing is interesting. After the Fable and Mythos access issue, the idea of having a swappable model pool suddenly feels more practical, not just technical.

This is more relevant for builders and teams developing AI products than everyday users.

Watch for:

For businesses, the takeaway is to start asking vendors: what happens if your primary model becomes unavailable?

Atlassian has launched Claude Agent for Jira, letting teams assign Jira tickets directly to Claude.

The idea is that the ticket becomes the starting point for the work. Claude can read the requirements, acceptance criteria, designs, and related context, then work in a secure environment and return a draft pull request for the team to review.

That is a much more practical version of AI coding than copying a task into chat, pasting in context, asking for code, then trying to fit the output back into the team’s existing workflow.

Here, the AI starts where the team already works.

The bottleneck with AI coding is not always the model.

A lot of the friction is everything around the model: getting the right context, understanding the task, following the team’s process, creating the branch, making the PR, and keeping humans in control.

This is why Jira integration matters. It brings AI into the existing software delivery workflow instead of asking teams to create a separate AI process.

The more useful pattern here is not “AI replaces developers.” It is “AI handles the first pass of well-scoped work, and humans review, guide, and merge.”

For operators, this is also a good lesson outside software: AI gets more useful when the work item, context, approval process, and output all live in the same system.

The key question is how well this works on real tickets.

Watch for:

Pinterest is testing Ask Pinterest, a standalone AI shopping and discovery app that lets users search through conversation instead of keywords.

Instead of typing a simple search like “living room decor,” a user could ask for help planning a room, choosing a style, or finding product ideas that match a specific taste.

The app can also use saved Pins and Boards to personalise suggestions when a user signs in.

Meta is moving in a similar direction with Facebook AI Mode, which turns the search bar into a conversational tool that can generate answers from public posts, Groups, Reels, and Marketplace content.

So across both platforms, search is starting to look less like a list of links and more like a guided conversation.

If you sell products or create content for a brand, start reviewing the information AI systems may use to understand you.

Look at:

Ask yourself: if an AI assistant was trying to recommend my business, would it have enough useful information to work with?