The Week AI Stopped Being Just Tech

Five days. That's all it took.

OpenAI signed a Pentagon deal to deploy models on classified networks. CNBC
Trump ordered every federal agency to immediately stop using Anthropic. CNBC
Perplexity launched an agent that coordinates 19 AI models.
Google released a free image generator that rivals paid tools. Google Claude Cowork quietly became the best second-brain setup nobody's covering.

OpenAI Takes the Pentagon. Anthropic Gets Blacklisted.

Anthropic and the Pentagon spent months negotiating how the military could use Claude in classified environments. NPR Anthropic held two red lines: no mass domestic surveillance, no fully autonomous weapons. Al Jazeera

The Pentagon wanted unrestricted "any lawful use." Anthropic wouldn't budge.

On Friday, it escalated fast. Trump directed all federal agencies to immediately cease using Anthropic The Hill, calling them "left-wing nut jobs." Defense Secretary Hegseth designated Anthropic a supply-chain risk The Washington Post — a label normally reserved for foreign adversaries. That designation bans military contractors from any commercial activity with the company. CNBC

Hours later, OpenAI CEO Sam Altman announced his own Pentagon deal. CNBC

The twist? Altman's statement suggests the Pentagon agreed to similar restrictions with OpenAI. CNN Cloud-only deployment, cleared OpenAI personnel in the loop, and prohibitions on surveillance and autonomous weapons. OpenAI

Altman admitted on X it was "definitely rushed" and the optics weren't great. TechCrunch He also asked the Pentagon to offer the same terms to all AI companies — including Anthropic.

The question every builder should sit with: What are your red lines? And would you hold them when the government is on the other side?

Perplexity Computer: One Prompt, 19 Models, Zero Switching


While everyone watched the Pentagon drama, Perplexity quietly launched the most ambitious AI product of the year. Perplexity

Computer doesn't replace your AI tools. It orchestrates all of them.

The system uses Claude Opus 4.6 for reasoning, Gemini for deep research, Nano Banana for images, Veo 3.1 for video, Grok for speed, and GPT-5.2 for long-context recall. Perplexity 19 models total.

You describe an outcome. Computer breaks it into subtasks, assigns each to the best model, and runs in the background VentureBeat — checking in only when it genuinely needs you.

It can run for hours or even months Perplexity on complex projects. Think: full competitive analyses, investment briefs, multi-step research with citations verified by a secondary agent.

The catch: $200/month on Perplexity Max only. TechCrunch

The signal: OpenAI already hired OpenClaw's developer Technology Org, betting big on multi-agent orchestration. This approach is coming to every major platform. Computer is just first to ship it polished.

Claude Cowork + Obsidian: Your Notes, But Smarter


If you use Obsidian, this is the highest-ROI setup of the week.

Claude Cowork can read every note in your vault once you grant folder access. No plugins, no MCP servers, no terminal. Coworkhow

What this unlocks:

Ask Cowork to find themes across your notes. Surface connections you've missed. Generate a monthly review from everything you wrote. It doesn't just read your notes — it thinks with them, finding forgotten threads and creating new connections. Towards AI

Then level up: let Cowork create and modify files — auto-generating weekly reviews, linking related notes, building project summaries Coworkhow from your existing vault.

Setup takes 5 minutes: Open Claude Desktop → Cowork tab → Grant vault folder access → Start asking questions about your notes.

Pro tip: Set up git version control on your vault before allowing write access. Coworkhow Safety net you'll be glad you have.

This is the closest thing to a second brain that actually thinks — not just stores.

Nano Banana 2: Free 4K Image Gen That Rivals Paid Tools


Google launched Nano Banana 2 — officially Gemini 3.1 Flash Image — combining Nano Banana Pro's quality with Flash's speed. Google

The highlights:

Resolutions from 512px to 4K. TechCrunch Real-time web search integration for accurately rendering real locations, products, and current subjects. Google Character consistency across up to five characters and fidelity for up to 14 objects in one workflow. TechCrunch Dramatically improved text rendering Google — clean, legible type directly in generated images.

The kicker: It's now the default image model across Gemini's free and paid plans Android Central, plus Search AI Mode, Google Lens, and developer APIs.

Equivalent quality from competitors costs $20–30/month. Google just made it free for everyone.

For anyone creating marketing visuals, pitch decks, blog headers, or product mockups — this changes the math overnight.

5 More Tools on Our Radar This Week

The big stories got the headlines. These shipped quietly and deserve your attention:

→ Claudian — Obsidian plugin that embeds Claude Code directly in your vault with full agentic capabilities GitHub — read, write, search, and multi-step workflows without leaving Obsidian.

→ OpenClaw — The viral open-source autonomous agent Technology Org that runs locally with deep system access. Powerful but configure carefully. OpenAI hired its creator Technology Org — expect this to merge into ChatGPT.

→ Cursor + Supabase — Still the fastest path from "I have an idea" to "it's live." Both expanded free tiers this month. If you haven't tried vibe-coding a working app in a weekend, this is your stack.

→ Gamma — AI presentations that don't look like AI presentations. Paste a doc or outline, get a polished deck. Useful for pitch prep and internal decks where speed matters more than custom design.

→ n8n — Open-source workflow automation that's becoming the default for AI-powered task chains. If Zapier feels limiting, this is the next step. Self-hostable, free for internal use.

3 Patterns From This Week That Actually Matter

Zoom out. Here's what the noise adds up to:

1 → AI governance is now a power struggle, not a policy debate. The Anthropic ban is the first time safety principles triggered a federal blacklisting. Every AI company is recalculating their red lines. Every builder should be thinking about theirs.

2 → Multi-model orchestration is the next platform war. Perplexity Computer, OpenClaw, Claude Cowork — three different architectures for the same idea. The winners won't have the best single model. They'll have the best router.

3 → The free tier just got dangerously good. Nano Banana 2 for free is better than most paid image tools were six months ago. Cowork gives Claude Code power without a terminal. If your product competes on "basic AI features," brace yourself.

🐝 One Last Thing...

Thanks for reading! If you found this deep dive valuable, don’t keep the alpha to yourself. Knowledge is the only asset that grows when you share it.

Forward this to someone you actually like.

See you in the next one

Keep Reading