Claude just made
“Where should I do this?”
easier.

Claude is bringing conversation, execution and finished work together

On September 16, Anthropic announced that chat and Cowork are merging into one Claude, alongside new Docs and Slides tools and Design inside conversations. A conversation can now grow from discussing a problem into carrying out the work and creating something you can edit, share or present. Anthropic's announcement

What changes in the way you work

The merge brings Cowork's existing capabilities into the conversation, with the context, skills and connectors you already use. You can start by asking a question, then ask Claude to find supporting information in an app you've connected, work through it and produce the result. You don't have to restart the task in Cowork when it becomes more involved.

Longer cloud tasks can continue after you leave the page, with progress available on your phone. Recurring work can run on a schedule. Work that needs local files or apps still requires Claude Desktop to stay open. These are capabilities brought together by the merge; they weren't all invented this week. How the merged experience works

The new creation tools add another part of the workflow. Docs supports direct editing and collaboration, Slides creates presentations, and Design works inside the conversation. You can keep developing the output where you discussed the task. In Docs, collaborators can edit together and ask Claude for changes through comments; you can export the result when it needs to live elsewhere. Claude Docs

Think about the whole job, then test one part

For a workshop, the full job might begin with your existing outline and participant questions, then move into a handout and teaching slides. You could review the explanation, correct it, and ask Claude to use the accepted version in the next output. The potential benefit is carrying the decisions through the work instead of repeatedly transferring them between tools.

That doesn't mean every workshop needs more materials. Choose what you're already responsible for delivering, and try one part first.

Take a task you already understand and follow it through.

A workshop handout is a useful example: bring the notes you already have, explain what participants should leave able to do, and work toward an editable document in the same conversation.

Try this when Claude Docs is available in your account:

Check the handout against the notes, make a correction, and open the exported file. The useful result is a handout you can actually give someone. This is a proposed exercise, not a test we have already run.

The merged experience is rolling out to Pro and Max first over the coming weeks. Docs, Slides and in-conversation Design are paid-plan betas; managed-workspace controls apply. Rollout details

Larger requests and multiple outputs still use your plan's allowance. Claude Docs also lacks version history for now, so keep an exported copy of an important accepted version. Docs usage and limitations

Notion can keep the instructions your team keeps rewriting

Notion 3.7, announced September 15, adds a shared library of skills: reusable instructions for work you do repeatedly. You can create one with Notion Agent, share it through your workspace, and choose it from the agent's slash menu.

You can also download a skill for tools such as Claude Code, Codex, Cursor and Gemini. Notion marks it when a newer version is available. The downloaded copy doesn't update itself. Notion's announcement

Start with the process you've already corrected

Suppose you repeatedly turn workshop notes into participant handouts. After you've reviewed a good result, save the instructions that made it useful: who the handout is for, what it must include, which facts it can use, and what you check before sharing it.

Run the saved instructions on a second set of notes. If they only work on the original example, they need more work before anyone else relies on them.

You don't need Notion to start preserving useful instructions. A maintained document works too. This update matters most if your team's work already lives in Notion and people need a common place to find the current version. Check the features available in your workspace before building around them.

ChatGPT now needs you to choose when to think longer

OpenAI's September 14 update removed automatic switching from Instant into Thinking for Plus and Pro users. Typing “think harder” doesn't move the control. Choose the Thinking level before starting the work. Managed workspaces can behave differently. September 14 release note

Give harder reasoning a job it can help with

Keep a short rewrite or straightforward question in Instant. Consider deeper Thinking when you need to compare conflicting sources, work through several constraints, or identify where a plan could fail.

For example, give it two approved versions of your cancellation policy and ask:

Then check the quoted sentences. More effort can't supply a missing policy decision.

What this uses

Plus includes Medium and High. Pro adds Extra High and Pro options. Manually chosen Thinking draws on the applicable Thinking allowance; ChatGPT may fall back to another model when you reach a limit. Check the model and reset information shown in your account. Chat and Work/Codex have separate usage rules, so an allowance in one isn't a promise of capacity in the other. Current controls and limits

AI can expand your job before the title changes

OpenAI published research on September 16 analyzing more than 1.5 million work-related ChatGPT messages. Among roughly 6,200 workers observed consistently from April through July, previously used tasks outside their usual occupation rose from 13.1% to 25.9% of occupation-specific AI activity.

That's a measure of activity in ChatGPT Business. It doesn't establish that people's total workloads doubled, that AI caused a role change, or that the resulting work was good. It does suggest that some experiments become recurring work. Read the research

Notice what became your responsibility

Maybe AI helped you write one promotional email, and now you're handling a weekly campaign. Maybe you made a useful comparison once, and now everyone expects you to produce it before each meeting.

Those can be worthwhile changes. They still need a decision about time, ownership and the expertise required to review the result.

Pick one task you've started doing because AI made it possible. Ask what it replaced. If the answer is “nothing,” decide what you'll stop, simplify, or hand back before making it a permanent workflow.

ChatGPT Ads are moving into the conversation after the click

On September 16, OpenAI announced new ways to create and manage ChatGPT Ads with natural-language instructions, AI help with creative, and HubSpot and U.S. Shopify integrations.

It also began testing Sponsored Agents with selected U.S. advertisers. After clicking an ad, someone can choose a clearly labeled conversation with a business-sponsored agent. That exchange is separate from ChatGPT's independent answers and the person's original conversation. This is a limited test, not a feature every business can turn on today. OpenAI's announcement

Your offer needs answers as well as ad copy

For a service business, the interesting question is what happens when a prospective buyer asks, “Is this right for someone at my stage?” or “What help do I actually get?”

Take one offer you already sell. Write down the five questions people ask before buying. For each, identify the current source for the answer and anything that needs a conversation with you.

For example, a workshop description may support an answer about session length and materials. It may not support a promise that someone will earn a certain amount afterward. That distinction matters whether a person or an agent answers.

If you're already advertising, investigate access and measurement before assigning budget. This announcement doesn't establish what a campaign would cost you or whether it would convert.

Give your useful custom GPTs a migration plan

This is a carryover from OpenAI's September 11 announcement. The current FAQ lists December 11, 2026 as the custom GPT retirement date, with any different account or workspace notice taking precedence. Existing GPTs remain usable until retirement, subject to their existing access.

Creators will be able to migrate GPTs to plugins as that option reaches their account. Migration uses the latest published version, not unpublished edits. Custom actions don't transfer automatically, and migration doesn't automatically install the plugin or give other people access. Migration FAQ

Start with the one you'd miss tomorrow

Keep its instructions, permitted source files and a few familiar prompts together. Record what a useful answer looks like so you can compare the replacement against real work.

If you use someone else's GPT, check their replacement guidance. Using it doesn't give you permission to migrate it yourself. If you've shared a GPT with customers or members, check their access separately when a replacement is ready.

Gemini's new Live models give voice a harder job

Google announced Gemini 3.8 Live and Live Extended Thinking on September 15 and updated the announcement September 17. Standard Live emphasizes responsive dialogue; Extended Thinking adds reasoning for more complex work while the conversation continues.

The rollout differs by product: standard Live is coming to Search Live; Extended Thinking to Gemini Live. Docs access is listed for Google AI Pro and Ultra, and Gmail and Keep for Google AI subscribers. Enterprise previews and future Workspace business availability are separate. Check what has reached your own account. Google's announcement

Rehearse something you already know

Use a short voice session to explain a workshop, prepare for common sales questions, or talk through a plan with several constraints. Extended Thinking is the relevant option for harder reasoning where it's available; a simple paragraph rewrite can stay in your existing text tool.

Here's a proposed rehearsal:

Correct misheard details as you go. Check the recap against what you meant, especially prices, dates and commitments. Smooth conversation doesn't establish accuracy, and a spoken progress update doesn't prove a connected action finished.

Limits and cost: check the product you're using

For the Gemini app, usage depends on your plan and the work you ask it to do. Google's general guidance describes five-hour and weekly limits. We haven't verified a universal daily Live-minute allowance. Start with access you already have; the exercise doesn't require a developer account. Gemini app limits

If you're building a voice tool through the API, Google's current paid rates for these models are $0.75 per million text-input tokens and $4.50 per million text-output tokens, including thinking. Audio is $3 input and $12 output per million tokens. Tokens are the units the model processes; these rates aren't your Gemini subscription allowance. Tools, search and other inputs can add cost. Developer pricing