Google turned Search into an agent

Google reframed Search at this week's I/O event as an agentic operating layer. By definition, that change demotes the destination page, and AI results become an end in themselves. It's happening everywhere. OpenAI's ChatGPT Personal Finance plugged 12,000-plus financial institutions through Plaid. Your institutional reader's daily workflow surface is moving inside conversational AI faster than your marketing plans might have expected.

The story that wins now is being agent-friendly. The format and structure of what you publish have to be readable by the systems that synthesize answers. A widely shared piece this week made the case that HTML beats Markdown for LLM context, because structured documents give agents richer scaffolding. AEO Monitor put the median time from publication to AI-answer citation at 6.81 days. Your publishing rhythm has to match that cadence.

Think about how that could affect your thought leadership timelines. When an institutional buyer asks an AI in your category about the problem you solve, does your firm show up in the answer? If not, you should assess what's missing in your corpus, your structure, your citations. Be sure to keep those fresh.

Your AI stack is now part of your story

Anthropic surpassed OpenAI on valuation this week, closing a $30B-plus round at over $900B. The Information reported that budget owners now care about Claude's reliability and tool use over IQ benchmark scores. KPMG plugged Claude into 276,000 workers. Microsoft cancelled its internal Claude Code licenses on cost grounds while continuing to sell Maia chips to Anthropic, a vendor-stack contradiction that's becoming the norm.

For marketing leaders, the AI vendor decisions inside your firm are no longer invisible. Buyers read them as procurement signals about your judgment. It's part of the broader conversation about how AI market share is shifting and the providers want to talk about it. If your team uses one provider for everything, the audience sees that. The choice used to be the IT director's problem. Now it's part of your category narrative whether you wanted it there or not.

What that means is your narrative needs to include and defend against questions about your AI stack. It's being read as a signal about concentration, switching, and supply chain.

The tsunami of slop has crested (for now)

Stop assuming AI volume will keep flooding the conversation. Graphite's analysis this week showed AI-generated articles have plateaued below 50% of the web, instead of climbing toward dominance. The reason might be what we see in new consumption data. TikTok's Scalable Summit numbers put narrative quality at 23x more views and 70x faster follower growth than production frequency. CreatorIQ found that 40% of YouTube views happen after the first 30 days, which means short-window attribution underprices durable material.

The publish-more, publish-faster, and flood the zone with AI content story might already be shifting. But this is what you should be thinking anyway. Publish what only your operators can write. Things like your editorial stance, the SME interview discipline, and the workflow details that make a piece useful to someone running a desk are the differentiators against AI-paced commodity. That's what the AI companies are doing. Anthropic published a Claude Code patterns guide for large codebases this week that reads like a blueprint for the kind of operator-grade reference material that earns citations and trust.

Treat your SME calendar like the most important meeting you run. The pieces that resonate are those with proprietary judgment and voice.