🚨Apple poaches Microsoft exec Amar Subramanya—who spent 16 years at Google leading Gemini engineering—as new AI chief replacing John Giannandrea. The hire signals Apple's recognition that it's fallen behind in the AI race and needs someone who knows the competition's playbook inside and out.

Anthropic just shared their playbook for how they use AI to automate their own marketing for Claude.

The crazy part? It's done by one non-technical employee. Not a team. One person.

Here's the workflow they shared:

Automated ad generation and optimization

Claude scans through CSV files of their ads, identifies underperforming ads, and uses two specialized sub-agents — one for headlines, one for descriptions — to write hundreds of new ad variations in minutes.

The human reviews their favorite options. Claude creates the new ads and replaces the underperforming ones. Fully automated.

The result: Anthropic is running hundreds of live Facebook ads right now, constantly iterating and finding winners through AI-generated variations.

Why this is different from traditional ad automation

Traditional ad automation optimizes targeting and bidding. The creative is still manual.

Anthropic's workflow automates the creative itself. The AI writes the headlines, the descriptions, and the variations. The human just approves or rejects.

This is the shift from automation to autonomy. You're not setting rules for the AI to follow. You're delegating the entire task and reviewing the output.

The broader pattern

Anthropic isn't just building AI tools. They're using AI tools to run their own business. And they're sharing the playbook publicly.

This is the "eat your own dog food" strategy. If Claude can't automate Anthropic's marketing, why would anyone else trust it to automate theirs?

The fact that one non-technical employee can run their entire ad operation is the proof point. It's not a demo. It's production.

What this unlocks

Marketing teams used to need copywriters, designers, media buyers, and analysts. Now you need one person who can manage AI agents.

The bottleneck isn't creativity anymore. It's judgment. The AI generates the ideas. The human decides which ones are worth running.

This is the same shift happening in coding, design, and content creation. The skill isn't execution. It's direction.

How to set this up yourself

If you're new to building AI workflows, here's a simplified version of Anthropic's ad automation setup using Claude Code:

1. Set up your data source: Export your current ad performance data to a CSV (campaign name, headline, description, CTR, conversions, cost per result).

2. Create a Claude Code project: Install Claude Code and initialize a new project in your terminal.

3. Write the analysis prompt: 

```

"Analyze this ad performance data. Identify the bottom 20% of ads by CTR and conversion rate. For each underperforming ad, generate 5 new headline variations and 5 new description variations that maintain brand voice but test different angles: pain point, benefit, social proof, urgency, and curiosity."

```

4. Set up sub-agents: Create two specialized prompts — one focused on headlines (30 character limit, punchy, benefit-driven) and one focused on descriptions (90 character limit, clear CTA, specific value prop).

5. Automate the review process: Have Claude output the variations to a new CSV with columns for original ad, new headline options, new description options, and a "recommended" flag for the AI's top pick.

6. Human review and approval: Review the recommended variations, approve the ones that fit your brand, and flag any that need revision.

7. Upload to ad platform: Use Claude Code to format the approved variations for your ad platform's bulk upload format (Facebook CSV, Google Ads Editor, etc.).

8. Schedule recurring runs: Set up a cron job or scheduled task to run this workflow every 48-72 hours, continuously optimizing your ad creative based on performance data.

Key considerations:

- Start with a small test batch (10-20 ads) before scaling to your full account

- Maintain a brand voice document that Claude can reference for consistency

- Set performance thresholds (e.g., only replace ads below 1% CTR or above $50 CPA)

- Keep a human in the loop for final approval — AI can generate, but you own the brand

If you're running ads, you're competing against teams that are using AI to generate and test hundreds of variations. Manual creative is too slow.

If you're building marketing tools, the moat isn't better targeting or better analytics. It's better creative automation.

If you're a marketer, your job is shifting from making ads to managing AI agents that make ads. The skill is knowing what to test, not how to write the copy.

What's next

Expect more companies to share their internal AI workflows publicly. The competitive advantage isn't the workflow itself. It's the execution speed and the judgment layer.

The best marketing teams won't be the ones with the most people. They'll be the ones with the best AI workflows and the clearest judgment on what works.

BTW: Anthropic's ad operation is proof that AI isn't just a tool for startups. It's a tool for AI companies to run themselves. If the people building AI are using AI to replace their own jobs, that's the signal.

The Recap: OpenAI has declared 'code red' internally as Sam Altman pushes staff to improve ChatGPT amid mounting competitive pressure from Google's Gemini and Anthropic's Claude. The once-unassailable leader in conversational AI now finds itself defending market position as rivals close the gap with superior benchmarks and user experience.

Unpacked:

Bottom line: The 'code red' declaration reveals how quickly AI leadership can erode when competitors have comparable resources and talent. OpenAI's challenge isn't just technical—it's maintaining developer and enterprise loyalty while Google and Anthropic offer increasingly compelling alternatives backed by massive infrastructure and data advantages.

Cloudflare acquires Replicate in major signal that cloud players are buying up the AI inference layer to own the deployment stack.

Data center energy demand projected to surge 300% through 2035 as the AI boom's infrastructure reality hits the power grid hard.

US government becomes major shareholder in chip startup xLight—Silicon Valley's libertarian ethos meets industrial policy reality in semiconductor race.

India mandates state cyber-safety app preloaded on every smartphone—a billion-user surveillance infrastructure wrapped in security theater.

Google's AI advantage isn't the models—it's decades of personal data creating assistants that know you better than you know yourself.

OpenAI's circular deal with Thrive Holdings raises questions whether portfolio firms build real businesses or just pump valuations with OpenAI partnerships.

Zillow kills climate risk scores after real estate agents complained it hurt sales—information asymmetry wins over homebuyer protection.

Construction workers cashing in on AI boom with data center jobs—infrastructure buildout creating blue-collar windfall as supervisors oversee 200-person crews.

Best of AI™ Team

Was this email forwarded to you? Sign up here.