Opus 4.8 Works Longer On Its Own, and Copilot Got a Faster Redesign
Two flagship launches on the same day, plus the desktop agent that operates your Windows apps. Here is what is worth your Tuesday.
What matters today
Two flagship launches on the same day, plus the desktop agent that operates your Windows apps. Here is what is worth your Tuesday.
Welcome byte
Two things shipped on the same Thursday and both are worth two minutes. Claude Opus 4.8 can now run a multi-step job from start to finish without drifting, which means a half-day competitive teardown becomes a 20-minute supervised session, and it is honest when it is stuck. Microsoft rebuilt Copilot into a task-aware workspace that loads more than twice as fast, and its standout trick is turning a 4,000-row export into a board one-pager with no pivot tables. The Pro Tip hands you the exact Opus 4.8 teardown prompt. The Productivity Gem is the Copilot one-pager instruction you reuse every reporting cycle. Below those: ChatGPT's Codex can now drive your Windows apps while you supervise from your phone, a mood-pattern method using your Apple Watch, and a weekend comic project for the kids.
Quick Hits
Copilot pulls live data from your other tools. Microsoft's new federated connectors let Copilot query HubSpot, Notion, Canva, and Google Calendar in real time via MCP, so answers reflect today's data instead of a stale indexed copy. Ask your admin to enable the connectors your team actually uses.
Copilot answers your missed Teams calls. A new call delegation feature picks up Teams Phone calls you cannot take, captures the caller's intent, summarizes it, and can even book a follow-up through Microsoft Bookings. Useful for anyone drowning in interruptions.
GPT-5.5 Instant got clearer, and Canvas moved. OpenAI tuned GPT-5.5 Instant for more natural, better-paced replies with fewer bloated bullet lists. Canvas is gone from the 5.5 models; writing and coding now happen inline in the chat. Nothing to do, just expect cleaner answers.
xAI shipped its fastest coding model to everyone. Grok Build 0.1 hit public beta on May 29 via the xAI API. It is purpose-built for agentic coding with a 256,000-token context window and text plus image input. One for the builders on your team, not the board deck.
Gemini changed how it counts your usage. Google's Gemini app moved from daily prompt caps to a compute-used model that factors prompt complexity, the features you use, and chat length, with limits refreshing every five hours up to a weekly cap. Heavy users should plan around the new rhythm.
Top Updates
Anthropic released Opus 4.8 on May 28 with a new dynamic workflow tool, sharper judgment, and the ability to run multi-step tasks end to end without drifting. It was the only model to complete every case on the Super-Agent benchmark, at the same price as the last Opus and at cost parity with GPT-5.5. It is live in the Claude apps, the API, Bedrock, Vertex AI, Microsoft 365 Copilot, and GitHub Copilot.
Action Steps
- Open Opus 4.8 in Claude, or select it inside Microsoft 365 Copilot for a hard task.
- Hand it one genuinely multi-step job, such as a contract read or your Monday inbox pass.
- Instruct it to pause and summarize its plan before any send or overwrite.
- Ask for a confidence note per section so your review goes where it is needed.
- Save the prompt and rerun it weekly. The payoff is the repeat.
Read the premium deep dive [Premium]
On May 28 Microsoft replaced Copilot's static prompt box with a workspace that surfaces the right controls for the task in front of you, powered by an intelligence layer called Work IQ. It loads more than twice as fast, and in a controlled 12-enterprise study tasks ran 32 percent faster, with Excel analysis up 47 percent. You can now pick Opus 4.8 or GPT-5.5 per task. Rollout to Current Channel completes by mid-June.
Action Steps
- Confirm with IT whether your tenant is on Current Channel, which gets it first.
- Brief your team to use GPT-5.5 Instant for speed and Opus 4.8 for high-stakes work.
- Re-time one recurring Excel report or PowerPoint deck before and after it lands.
- Open Copilot in Word, Excel, PowerPoint, and Outlook to learn the new placement.
- Bank the speed and clutter wins; treat the 32 percent figure as a best case.
Read the premium deep dive [Premium]
OpenAI extended Codex with Computer Use on Windows, so it can see, click, and type inside native Windows applications, not just the browser. The real unlock is remote continuation: start a job at your desk and supervise it from ChatGPT on your phone or Codex on a Mac. Note the limit, it is not available in the EEA, UK, or Switzerland at launch, and it is a prosumer and developer feature first.
Action Steps
- Confirm the feature is available for your account and outside the restricted regions.
- Pick one repetitive desktop chore with no destructive step for the first run.
- Keep ChatGPT open on your phone to approve or redirect at each decision point.
- Check Codex Profiles so a long run does not surprise you on usage.
- Only let it run longer once a task has completed cleanly twice.
Read the premium deep dive [Premium]
Productivity Gem
The redesigned Copilot claims 47 percent faster Excel analysis. Skip the pivot tables entirely: open your raw export, paste one reusable instruction, and Copilot returns the five numbers that matter with period-over-period change, one clean chart, a three-sentence plain-language summary, and a data-quality flag list, with the math shown so you can verify it. Save the instruction once and rerun it every reporting cycle in under five minutes.
Read the premium deep dive [Premium]
Put it to work Strategic Brainstorming: Use AI to Generate and Refine Ideas
Pro Tip
Opus 4.8's longer autonomous runs turn a half-day competitor teardown into one supervised session. Feed it the competitor URL, their pricing, and two or three recent announcements, then run a single prompt that plans its approach, pauses for your approval, and returns a one-page teardown with confidence labels on every section. You correct any wrong assumption once, at the checkpoint, instead of re-reading ten separate answers. Most teardowns that took four to five hours land in about 20 minutes.
Read the premium deep dive [Premium]
Copy-ready guide AI for Executive Summaries: Distill Complex Information Fast
Health Tip
Your iPhone and Apple Watch already log State of Mind in the Mindfulness app in about ten seconds. Log it twice a day for two weeks, then hand the entries to ChatGPT and ask one question: what time of day and what tagged factors line up with your lowest and highest ratings, and the single change most likely to raise your average. Then change one thing and re-measure. You are using devices and a model you already pay for, no new app required.
Read the premium deep dive [Premium]
Kids Tip
A weekend project for ages 8 to 16: plan a three-panel story on paper, then direct the Gemini app's free image generator to illustrate each panel with specific prompts, keeping the same character description and changing only the action across panels. Assemble the images in Google Slides, add speech bubbles, and print it. The child is the author and art director, and they learn the single most valuable AI skill there is: describing exactly what they want.
Read the premium deep dive [Premium]
That is the week. Two flagship models, a faster Copilot, and a desktop agent you can run from your phone. Pick one thing above and actually try it before next Wednesday. That is the whole point of this newsletter.
See you next week, Pierre Bradshaw PromptHacker
Before next Wednesday
Three 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.
No comments yet