The 10-second version

  • Anthropic moved Claude Cowork to web and mobile, which means agents are starting to feel less like chatbots and more like background workers.

  • OpenAI’s latest Codex research says agent use grew more than fivefold in the first half of 2026, including outside software teams.

  • The smart move this week: stop asking AI for answers. Start giving it bounded work with a review step.

The Big Thing

Agents just got less theatrical, which is probably good.

Anthropic says Claude Cowork now works on web and mobile, with remote sessions that keep running when your laptop is closed and scheduled tasks that can run with no device online. That is not a “wow, it wrote a poem” feature. It is a “the spreadsheet got cleaned up while I was at lunch” feature.

OpenAI’s recent Codex research points in the same direction: agentic AI is moving from short chats to longer delegated tasks, with usage growing quickly and spreading beyond developers.

Why you care: the useful unit of AI work is shifting from “prompt” to “assignment.”

What to say in your next meeting: “Let’s stop measuring AI by who tried the newest model. Let’s pick three recurring tasks, define the handoff, define the review, and see what actually disappears from the calendar.”

Speed Round

  • Claude Cowork is now rolling out on web and mobile. Max users get it first, with more plans following. Translation: agents are becoming portable, not just powerful.

  • ECB told eurozone banks to submit AI cyber-defense plans by October 31. Regulators are treating advanced AI as operational risk infrastructure, not a thought leadership topic.

  • Anaplan introduced “Agentic Enterprise.” The pitch: agents grounded in shared enterprise data, deterministic calculations, and auditable business logic. Boring words, but powerful for the future of AI agents.

  • Bespoke Labs raised $40M for reliable-agent training environments. The market is moving toward simulation, evaluation, and “prove it before production.”

  • Social read: Hacker News is arguing about whether the agent harness matters more than the raw model. The practical answer is yes enough to matter. Your instructions, tools, review loop, and memory file are part of the product now.

The Two-Minute Win

Prompt of the Day: turn a 30-minute status update into a 2-minute review

Use this when you have meeting notes, Slack threads, email replies, or project updates scattered everywhere.

Paste this:

“Turn the notes below into a status update for my manager. Use this structure:

  1. What changed since the last update

  2. Decisions needed

  3. Blockers

  4. Risks that may become blockers

  5. Next 3 actions, each with owner and due date

  6. One sentence I can say live in the meeting

Rules:

  • Do not invent owners or dates. Use ‘needs owner’ or ‘needs date’ if missing.

  • Separate facts from assumptions.

  • Keep it skimmable.

  • End with a 5-bullet executive version.”

Then paste your raw notes.

This turns the “let me organize my thoughts first” tax into a review pass. Estimated time reclaimed: 20 to 30 minutes per weekly update.

Want more?

Two Minutes AI helps turn deep work into a two-minute task.

The good stuff is not “AI replaced your job.”
The good stuff is “AI removed the 17-minute task you hated, and nobody noticed except your blood pressure.”

Paste one messy task into Two Minutes AI and get the clean version back before your coffee starts lying about productivity.

Be an early adopter: https://twominutesai.app/

See you later, unless an agent schedules me into a meeting about agent strategy, in that case, the calendar has defeated me.

Toodles,
Don

Keep reading