AI DUNN Right Weekly — Issue #25

Practical AI insights for business growth

Hey AI Innovators! 👋

This week the AI world got political, practical, and a little bit personal. All at once. An OpenAI exec walked out over the Pentagon deal. Google quietly upgraded every app you already use. And Grammarly is in court for turning real writers into AI training data without asking first.

Here's what you need to know:

• An OpenAI hardware lead resigned over the company's Pentagon partnership

• Google rolled out Gemini directly inside Docs, Sheets, Slides, and Drive

• Microsoft launched Copilot Cowork, powered by Anthropic's Claude

• Claude can now respond with charts and diagrams built right into the chat

• Grammarly is being sued for using writers as "AI editors" without their consent

• ChatGPT launched interactive visual explanations for 70+ math and science topics

  
  

Read time: 6 minutes

---

Big story

An OpenAI exec just walked out. And the reason matters.

Most people in tech quietly disagree with decisions they don't love.

Caitlin Kalinowski, who led OpenAI's robotics and hardware division, did not do that. She resigned this week specifically because of OpenAI's deal with the Pentagon, and she said so publicly.

The Pentagon deal has been controversial from the start. Questions have swirled about how AI tools from major labs are being used in military contexts, and what guardrails, if any, are in place. Kalinowski made clear she could not stay at a company that signed that contract.

Here's why this matters to anyone watching the AI industry:

• Internal dissent at the top level of a major AI lab is rare

• It signals real tension between moving fast on government contracts and staying true to safety values

• It adds more pressure on OpenAI to explain exactly what they agreed to with the Pentagon

  
  

My take? When someone with Kalinowski's seniority leaves a job over a values disagreement and talks about it openly, you pay attention. This is a person saying "I cannot put my name on this." That means something. Whatever your politics, that kind of resignation tells you the ethical stakes around AI and military use are very, very real.

---

What's new this week

Google brings Gemini directly into your Workspace apps

Google rolled out new Gemini capabilities inside Docs, Sheets, Slides, and Drive this week. No separate chat window needed.

Here's what changed:

• In Docs: Describe what you want to create. Gemini pulls relevant info from your Gmail, Drive, and Chat and drafts it for you. There's also a "Match writing style" feature that unifies tone across a document with multiple contributors.

• In Sheets: A single prompt can pull data from across your workspace and build a fully formatted spreadsheet. The new "Fill with Gemini" feature populates tables and can pull live info from Google Search.

• In Slides: Gemini generates fully editable slides that match your existing deck theme. A future update will let you build a full presentation from one prompt.

• In Drive: Search using plain language and get an AI summary at the top of your results, without opening the file.

  
  

Currently in beta for Google AI Ultra and Pro subscribers. English worldwide for Docs, Sheets, and Slides. Drive features are U.S. only for now.

This is the practical shift a lot of us have been waiting for. AI built into the tools you already open every morning.

Microsoft adds Claude to Copilot

Microsoft announced Copilot Cowork this week, a new offering built on Anthropic's Claude technology.

What it does:

• Handles complex, multi-step tasks across your Microsoft 365 apps

• Creates apps, builds spreadsheets, and organizes large volumes of data

• Works with minimal back-and-forth from you

• Runs entirely in the cloud, so companies know exactly what data it can access

  
  

The cloud piece is deliberate. Most enterprises are nervous about AI agents running locally on devices, where it's harder to monitor what's happening. Microsoft's version is cloud-based with controlled access, which is what large organizations need to feel comfortable deploying it.

Copilot Cowork is currently in early access, with broader availability coming later this month. Microsoft is also now making Claude Sonnet models available to M365 Copilot users, meaning Microsoft is no longer running exclusively on OpenAI's GPT models.

If you're in a corporate environment, this one is worth watching closely.

Claude now responds with charts and diagrams

Anthropic rolled out visual output capabilities for Claude this week. Charts, diagrams, and other visuals now appear directly in the conversation, not as a separate file or external link.

For anyone using Claude for analysis, reporting, or explaining complex ideas to clients or colleagues, this is a genuine time saver. Claude can now show you something instead of only describing it.

---

Tool of the week

AWS launches OpenClaw on Amazon Lightsail

If you've heard of OpenClaw but found the setup too complicated, AWS just made it a lot easier.

OpenClaw is an open-source autonomous AI agent that runs privately on your own infrastructure. Think of it as a personal digital assistant that manages emails, browses the web, organizes files, and connects to messaging apps like WhatsApp, Discord, and Telegram. Your data stays on your own infrastructure.

AWS now offers pre-configured OpenClaw instances on Amazon Lightsail. Here's how the setup works:

• Launch an instance in the Lightsail console

• Choose OpenClaw from the blueprint options

• Pair your browser using the built-in SSH terminal

• Amazon Bedrock is integrated by default to power the AI capabilities

  
  

The 4GB memory plan is recommended for best performance. Pay-as-you-go pricing means you only pay for what you use.

If you've been curious about running a private AI agent but didn't want the technical headaches, this is the most accessible version of that option yet.

---

Quick hits worth your time

→ Grammarly is being sued by a writer who says the company used her work, and the work of other authors, to train its AI "expert review" feature without their consent. If you use any writing tool, this one is worth following. The question of who owns your content when you put it through an AI tool is far from settled.

→ DOGE used ChatGPT to help identify cuts at the National Endowment for the Humanities. AI is now being used at the government policy level to decide which programs survive and which don't. Worth being aware of, whatever your view on the politics.

→ Claude can now respond with charts and diagrams built directly into the conversation. If you use Claude for client-facing work or internal reporting, go test it this week.

→ ChatGPT launched interactive visual explanations for over 70 math and science topics. Adjust variables in real time and watch the formula update instantly. Topics include

compound interest, exponential decay, linear equations, and more. Available to all logged-in ChatGPT users now.

→ OpenAI's hardware lead Caitlin Kalinowski resigned this week over the Pentagon deal. She led the robotics division and made her reasons public. Worth watching for more fallout from this.

---

Prompt of the week

This week's prompt is for anyone who wants to understand where AI agents are actually headed, without needing a tech background to follow along:

Act as an AI education expert for entrepreneurs.

I have been using AI tools like [LIST THE TOOLS YOU USE]
for [LENGTH OF TIME] and I feel comfortable with the basics.

I want to understand what agentic AI is,
how it is different from tools I already use,
and what it can actually do for a [YOUR BUSINESS TYPE].

Explain it clearly without jargon.

Ask me any questions you have before you start.

Variables to fill in:

• LIST THE TOOLS YOU USE: ChatGPT, Claude, Gemini, etc.

• LENGTH OF TIME: How long you have been using them

• YOUR BUSINESS TYPE: Your industry or what you do

• 
  

Why this works: Everyone is talking about AI agents right now. Almost no one explains what they actually are in plain language. This prompt gets Claude or ChatGPT to explain agentic AI in the context of your specific business. The "ask me questions first" instruction is key. It gives the AI enough context before it starts, so the explanation actually fits your situation.

---

My take

Two things stood out to me this week.

Google making Gemini useful inside Docs, Sheets, and Drive is the kind of quiet upgrade that changes how you work every single day. No new app to learn. No tab switching. Just describe what you need and get a draft. That is where this technology is heading.

And then Caitlin Kalinowski walked out of OpenAI over a government contract. That is not a small thing. Decisions being made at the top of these companies affect more than just product features. They affect how AI gets used, and by whom. The people paying attention to both sides of this, the capabilities and the ethics, will be better positioned than those who only track the tools.

Keep experimenting. Keep learning. And ask the hard questions too.

See you next Monday.

Jackie