The Week AI Moved Into the Work Itself
ChatGPT got better at decisions, Grok moved into Word, Google shipped role-based Gemini training, and Perplexity made research easier to branch.
What matters today
ChatGPT got better at decisions, Grok moved into Word, Google shipped role-based Gemini training, and Perplexity made research easier to branch.
Key points
- Pick one decision you owe someone this week, not a toy example.
- Install the add-in in Word and open one real draft that is painful to clean up.
- Assign one marketer to complete the course and bring back one workflow, not a summary.
- Use Instant for routine drafting and High for decisions with cost, timing, or reputational risk.
- Start with one base question and tell Perplexity what decision the research should support.
Welcome byte
I kept coming back to the same pattern in this issue: the useful AI updates showed up inside the places teams already work. ChatGPT got better at messy decisions. Grok moved into Word, which is still where a lot of proposals and board memos get finished. Google turned Gemini adoption into a short role-based marketing course instead of another vague "learn AI" page.
The practical move is to pick one surface and run a live test. Use ChatGPT's model picker before a hard prompt. Try Grok inside one messy Word draft. Use Perplexity Computer to branch one research question into separate deliverables. Then do the privacy check before placing health data near any AI chat, and give the kids a model they can train with their own examples.
Quick Hits
Long ChatGPT pastes now become attachments.
OpenAI says pastes over 10,000 characters now convert into attachments for Free and Go users, with Plus, Pro, and Business already covered, so you can drop a long report or transcript into ChatGPT without wrecking the composer. Read more
ChatGPT adds better app controls and project organization.
The June 18 update added connected-app permission controls, easier project pinning, faster sharing, notes from responses, improved iOS photo upload, and one-off Android model selection for paid users. Read more
Codex Record & Replay turns a demonstrated workflow into a skill.
Eligible macOS users can record a repeatable workflow once and reuse it through Codex, Computer Use, browser actions, plugins, or a mix of tools. Read more
Grok Build gets a long-running /goal mode.
xAI says /goal lets Grok Build plan the work, keep a progress checklist, continue execution, and expose status, pause, resume, and clear commands for larger implementation tasks. Read more
Perplexity Computer moves into Microsoft 365.
Perplexity says Computer now works directly inside Word, Excel, PowerPoint, Outlook, and Teams, with a context panel and usage analytics so teams can see what the agent is using and producing. Read more
Top AI Updates
1. GPT-5.5 Instant gets better at decision work
OpenAI updated GPT-5.5 Instant on June 24 to improve answers when people are making decisions, asking for advice, planning, researching options, or shopping. The useful part is the default ChatGPT surface getting better at the work Executives already bring to chat: vendor comparisons, budget tradeoffs, hiring choices, and "what should I do next?" questions.
Why it matters: I would test this on one real decision before trusting it on ten. Give it options, constraints, risks, and what winning means, then force one recommendation. Because it is the default flow, casual ChatGPT users will touch it before they touch any advanced setting.
Action items:
- Pick one decision you owe someone this week, not a toy example.
- Paste the options, constraints, risks, and success criteria in one message.
- Ask for the one fact that would change the recommendation before you act.
2. Grok moves into Microsoft Word
xAI launched Grok for Word on June 18 as a free Microsoft 365 add-in. It can turn notes into structured text, rewrite for clarity, format a document, search the web or X, generate diagrams, and use connectors for recent emails or files in SharePoint and Google Drive.
Why it matters: The copy-paste loop between an AI chat and a document wastes attention. If your team lives in Word for proposals, board updates, policies, or client memos, Grok for Word is worth a 20-minute test on one ugly draft.
Action items:
- Install the add-in in Word and open one real draft that is painful to clean up.
- Ask Grok to produce a one-page executive version with open questions called out.
- Require it to label which claims came from the draft, the web, X, or connected files.
3. Google gives marketers a 45-minute Gemini workflow path
Google published AI Boost Bootcamp 201 for marketing teams on June 22. The course is built for role-specific use, not general AI literacy, and covers NotebookLM for feedback synthesis, Gemini Deep Research for competitive analysis, Canvas and Google Vids for campaign assets, =AI() in Sheets for sentiment analysis, Workspace Studio for campaign workflows, and Gemini Enterprise agents for marketing tasks.
Why it matters: Training only matters when it changes the work on Monday. This is a useful model for any team trying to move from "we should use AI" to one repeatable workflow per function.
Action items:
- Assign one marketer to complete the course and bring back one workflow, not a summary.
- Pick one course example to test: customer feedback synthesis, competitive research, sentiment analysis, or campaign handoff.
- Turn the result into a one-page internal recipe with owner, inputs, review step, and success metric.
Pro Tip
Use the model picker before the prompt
OpenAI simplified the ChatGPT model picker on June 10 so Plus and Pro users can choose Instant, Medium, High, and, on Pro, Extra High or Pro modes from the composer. This is the better habit: choose the reasoning level first, then write the prompt.
Prompt to unlock it: "Act as a decision editor for an Executive. Decision: [DECISION]. Context: [FACTS]. Constraints: [BUDGET, TIME, PEOPLE, RISK, CUSTOMER IMPACT]. Options: [A/B/C]. Success looks like: [SUCCESS CRITERIA]. If one missing fact would materially change your answer, ask for it first. Otherwise, state your assumptions and proceed. Compare the options in a table with upside, downside, hidden assumption, failure mode, and best-case use. Then give: 1. recommended option, 2. strongest argument against it, 3. what would make you reverse the call, 4. first action in the next 24 hours, and 5. owner plus check-in metric."
Action items:
- Use Instant for routine drafting and High for decisions with cost, timing, or reputational risk.
- Run the same decision once on Instant and once on High so you can feel the difference.
- Save the winning version as your reusable decision memo template.
Copy-ready guide Pro Tip: Use ChatGPT's Model Picker Before the Prompt
Productivity Gem
Turn Perplexity Computer into branching research
Perplexity's June 19 changelog added Deep Research inside Computer, plus faster command access, forking, inline actions, analytics APIs, and custom credit limits. My instruction: "Research [MARKET, ACCOUNT, OR COMPETITOR]. Create a base brief, then fork three branches: risks, opportunities, and next actions. Turn the strongest branch into a one-page memo with sources and a next-step checklist."
Action items:
- Start with one base question and tell Perplexity what decision the research should support.
- Fork the thread into three views: risk, opportunity, and next action.
- Export only the strongest branch, with citations and a short checklist for the person who owns the next move.
Put it to work Perplexity Computer Branching Research Workflow
Health Tip
Audit health-data sharing before AI
Keep this simple: open the Health app on your iPhone, tap Sharing, review every person, provider, app, and device with access, and remove anything stale. Then open ChatGPT settings, review Data Controls, use Temporary Chat for sensitive wellness summaries, and strip names, exact dates, addresses, and unnecessary identifiers before pasting anything. Use this prompt: "I am pasting a privacy-safe wellness summary. Do not diagnose. Do not suggest treatment. Help me organize the data into patterns, questions, and one low-risk habit to discuss with a qualified professional. If the information is not enough, ask clarifying questions instead of guessing."
Step-by-step guide The Health Data Sharing Audit Before AI
Kids Tip
Build a family image classifier in Google Teachable Machine
The child is building a simple AI model that learns to sort webcam photos into categories, such as mugs, books, and toy cars. Start at Google Teachable Machine, choose an image project, then create three safe household categories. Collect 20 to 30 examples for each class, train the model, and test it with new objects, new angles, and different lighting. The payoff is immediate: kids see that examples, labels, backgrounds, and test data shape what AI learns.
Try the activity Build a Family Image Classifier With Google Teachable Machine
Wrap Up
Pick one place you already work and give AI one real job there. If it saves time, keep it. If it creates cleanup, you learned that before rolling it out to the team.
About the author
Pierre has spent 25+ years building growth systems across fintech, real estate, lending, campaigns, and AI workflows, with $1.5B+ in client value delivered.
Email usThree deep dives. Four useful moves. One email worth opening.
PromptHacker turns the AI firehose into practical next steps for work, health, family, and everything time keeps trying to steal.